Interventions for improving medication‐taking ability and adherence in older adults prescribed multiple medications
Abstract
Background
Older people taking multiple medications represent a large and growing proportion of the population. Managing multiple medications can be challenging, and this is especially the case for older people, who have higher rates of comorbidity and physical and cognitive impairment than younger adults. Good medication‐taking ability and medication adherence are necessary to ensure safe and effective use of medications.
Objectives
To evaluate the effectiveness of interventions designed to improve medication‐taking ability and/or medication adherence in older community‐dwelling adults prescribed multiple long‐term medications.
Search methods
We searched MEDLINE, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), PsycINFO, CINAHL Plus, and International Pharmaceutical Abstracts from inception until June 2019. We also searched grey literature, online trial registries, and reference lists of included studies.
Selection criteria
We included randomised controlled trials (RCTs), quasi‐RCTs, and cluster‐RCTs. Eligible studies tested interventions aimed at improving medication‐taking ability and/or medication adherence among people aged ≥ 65 years (or of mean/median age > 65 years), living in the community or being discharged from hospital back into the community, and taking four or more regular prescription medications (or with group mean/median of more than four medications). Interventions targeting carers of older people who met these criteria were also included.
Data collection and analysis
Two review authors independently reviewed abstracts and full texts of eligible studies, extracted data, and assessed risk of bias of included studies. We conducted meta‐analyses when possible and used a random‐effects model to yield summary estimates of effect, risk ratios (RRs) for dichotomous outcomes, and mean differences (MDs) or standardised mean differences (SMDs) for continuous outcomes, along with 95% confidence intervals (CIs). Narrative synthesis was performed when meta‐analysis was not possible. We assessed overall certainty of evidence for each outcome using Grades of Recommendation, Assessment, Development and Evaluation (GRADE). Primary outcomes were medication‐taking ability and medication adherence. Secondary outcomes included health‐related quality of life (HRQoL), emergency department (ED)/hospital admissions, and mortality.
Main results
We identified 50 studies (14,269 participants) comprising 40 RCTs, six cluster‐RCTs, and four quasi‐RCTs. All included studies evaluated interventions versus usual care; six studies also reported a comparison between two interventions as part of a three‐arm RCT design.
Interventions were grouped on the basis of their educational and/or behavioural components: 14 involved educational components only, 7 used behavioural strategies only, and 29 provided mixed educational and behavioural interventions. Overall, our confidence in results regarding the effectiveness of interventions was low to very low due to a high degree of heterogeneity of included studies and high or unclear risk of bias across multiple domains in most studies.
Five studies evaluated interventions for improving medication‐taking ability, and 48 evaluated interventions for improving medication adherence (three studies evaluated both outcomes).
No studies involved educational or behavioural interventions alone for improving medication‐taking ability. Low‐quality evidence from five studies, each using a different measure of medication‐taking ability, meant that we were unable to determine the effects of mixed interventions on medication‐taking ability.
Low‐quality evidence suggests that behavioural only interventions (RR 1.22, 95% CI 1.07 to 1.38; 4 studies) and mixed interventions (RR 1.22, 95% CI 1.08 to 1.37; 12 studies) may increase the proportions of people who are adherent compared with usual care. We could not include in the meta‐analysis results from two studies involving mixed interventions: one had a positive effect on adherence, and the other had little or no effect. Very low‐quality evidence means that we are uncertain of the effects of educational only interventions (5 studies) on the proportions of people who are adherent.
Low‐quality evidence suggests that educational only interventions (SMD 0.16, 95% CI ‐0.12 to 0.43; 5 studies) and mixed interventions (SMD 0.47, 95% CI ‐0.08 to 1.02; 7 studies) may have little or no impact on medication adherence assessed through continuous measures of adherence. We excluded 10 studies (4 educational only and 6 mixed interventions) from the meta‐analysis including four studies with unclear or no available results. Very low‐quality evidence means that we are uncertain of the effects of behavioural only interventions (3 studies) on medication adherence when assessed through continuous outcomes.
Low‐quality evidence suggests that mixed interventions may reduce the number of ED/hospital admissions (RR 0.67, 95% CI 0.50 to 0.90; 11 studies) compared with usual care, although results from six further studies that we were unable to include in meta‐analyses indicate that the intervention may have a smaller, or even no, effect on these outcomes. Similarly, low‐quality evidence suggests that mixed interventions may lead to little or no change in HRQoL (7 studies), and very low‐quality evidence means that we are uncertain of the effects on mortality (RR 0.93, 95% CI 0.67 to 1.30; 7 studies).
Moderate‐quality evidence shows that educational interventions alone probably have little or no effect on HRQoL (6 studies) or on ED/hospital admissions (4 studies) when compared with usual care. Very low‐quality evidence means that we are uncertain of the effects of behavioural interventions on HRQoL (1 study) or on ED/hospital admissions (2 studies). We identified no studies evaluating effects of educational or behavioural interventions alone on mortality.
Six studies reported a comparison between two interventions; however due to the limited number of studies assessing the same types of interventions and comparisons, we are unable to draw firm conclusions for any outcomes.
Authors' conclusions
Behavioural only or mixed educational and behavioural interventions may improve the proportion of people who satisfactorily adhere to their prescribed medications, but we are uncertain of the effects of educational only interventions. No type of intervention was found to improve adherence when it was measured as a continuous variable, with educational only and mixed interventions having little or no impact and evidence of insufficient quality to determine the effects of behavioural only interventions. We were unable to determine the impact of interventions on medication‐taking ability. The quality of evidence for these findings is low due to heterogeneity and methodological limitations of studies included in the review. Further well‐designed RCTs are needed to investigate the effects of interventions for improving medication‐taking ability and medication adherence in older adults prescribed multiple medications.
PICO
Plain language summary
Interventions for helping older adults prescribed multiple medications to use and take their medications
Background: Older people are often prescribed multiple medications, which can be challenging to manage. Medication‐taking errors and non‐adherence (under‐use or over‐use of medication) can lead to negative health outcomes. Assisting older people to better use and adhere to their medications could reduce adverse medication events, such as medication‐related hospital admissions, and could improve health outcomes.
Question: What are the findings of studies testing ways to improve older people's ability to use and adhere to multiple medications?
Search strategy: To find relevant studies, we searched seven online databases, trial registries, and the reference lists of previous reviews, retrieving studies published until June 2019.
Selection criteria: We included randomised controlled trials (RCT) or studies of similar design comparing a group of people receiving an intervention to improve medication‐taking ability or medication adherence with a group receiving usual care (no intervention) or receiving a different intervention. We included trials that studied older adults (≥ 65 years) living at home (or being discharged from hospital back to home) who were using four or more regular prescription medications.
Main results: We identified 50 studies, involving 14,269 participants. All studies tested interventions versus usual care, with six studies also comparing two different types of interventions.
Fourteen studies tested educational interventions whereby people received education regarding their medications or a health professional reviewed their medications. Seven studies tested behavioural interventions such as changing dosing times, re‐packaging medications into multi‐compartment pill boxes to make medication regimens easier to take, or sending text message adherence reminders. Twenty‐nine studies tested mixed educational and behavioural interventions.
The studies identified were very different in terms of what interventions people received, where interventions were delivered, and how and when people's medication‐taking ability or adherence was measured. Due to these differences and problems with how the trials were conducted, the quality of the evidence was considered low or very low overall.
Low‐quality evidence means that the impact of mixed interventions on medication‐taking ability could not be determined, and no studies were identified that assessed educational only or behavioural only interventions for improving medication‐taking ability.
Low‐quality evidence suggests that compared with usual care, behavioural only and mixed interventions may improve the proportions of people who satisfactorily adhere to their prescribed medication, but very low‐quality evidence means that the effects of educational only interventions are uncertain. Low‐ and very low‐quality evidence means that no interventions were found to be effective in improving medication adherence when assessed by continuous measures such as percentage of medications consumed.
Low‐quality evidence also suggests that mixed interventions may reduce the number of emergency department visits or hospital admissions, and may lead to little or no change in health‐related quality of life (HRQoL). Moderate‐quality evidence shows that educational interventions alone probably have little or no effect on HRQoL or on emergency department or hospital admissions. The effects of behavioural interventions alone on HRQoL or emergency department or hospital admissions are uncertain because of very low‐quality evidence. We are uncertain of the effects of behavioural, educational, or mixed interventions on mortality.
Studies comparing one type of intervention with another were limited in number, and we are unable to draw firm conclusions for any key outcomes.
Authors' conclusions: Interventions varied greatly among studies, and there were problems regarding how the trials were conducted, which may have affected their results. We were unable to determine the impact of interventions on medication‐taking ability. Low‐quality evidence suggests that behavioural only and mixed educational and behavioural interventions may improve the proportions of people who adhere to their prescribed medication regimen. Low‐ and very low‐quality evidence found no type of intervention to be effective in improving medication adherence when this was assessed by a continuous measure. High‐quality studies are necessary to identify the most effective way to improve medication‐taking ability and medication adherence among older adults prescribed multiple medications.
Authors' conclusions
Summary of findings
| Mixed educational and behavioural interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving both educational and behavioural components Comparison: usual care | |||
| Outcomes | Impacts | No of Studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: 2 weeks to 12 months | The effects of mixed interventions on medication‐taking ability were unable to be determined. Meta‐analysis was not possible due to all 5 studies using different outcome measures. Of the 5 studies, 1 demonstrated significant improvement in medication‐taking ability, 2 showed no significant impact, 1 did not test for differences between groups, and 1 did not report results | 5 | Lowa,b |
| Medication adherence (dichotomous) Follow‐up: 1 to 18 months | Mixed interventions may improve the proportion of people who are adherent (dichotomous adherence outcome) Twelve studies (3147 participants) were included in a meta‐analysis. Risk ratio was 1.22 (95% CI 1.08 to 1.37), indicating interventions increased the absolute number of adherent participants by 12.8% (4.6% to 21.5%) Two studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study reported the intervention increased the number of medications taken correctly; 1 study showed no differences between groups | 14 | Lowa,b |
| Medication adherence (continuous) Follow‐up: 1 to 12 months | Mixed interventions may have little or no impact on medication adherence measured by continuous adherence outcomes (e.g. proportion of pills dispensed or taken) Seven studies (1825 participants) were included in a meta‐analysis. Standardised mean difference was 0.47 (95% CI ‐0.08 to 1.02), indicating that the mean adherence score in the intervention group was 0.47 standard deviations higher (0.08 lower to 1.02 higher) than in the usual care group Four studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study showed fewer medication errors as a proportion of total doses with the intervention; 3 studies showed no significant effect on adherence. Two additional studies were excluded due to unclear reporting of results | 13 | Lowb,c |
| Health‐related quality of life Follow‐up: 6 to 18 months | Mixed interventions may lead to little or no change in health‐related quality of life. Six of 7 studies showed no significant impact on this outcome. One study reported the intervention may improve both physical and mental summary scores on the SF‐36 at 12 months. Meta‐analysis was not possible due to differences in scales used and differences in reporting of results | 7 | Lowa,b |
| Emergency department (ED)/Hospital admissions Follow‐up: 1 to 24 months | Mixed interventions may reduce the number of emergency department (ED) and/or hospital admissions. Eleven studies (1827 participants) were included in meta‐analysis. Risk ratio was 0.67 (95% CI 0.50 to 0.90), indicating mixed interventions may reduce the absolute number of patients admitted to ED/hospital by 12.3% (18.7% to 3.7% fewer). Six studies were excluded from the meta‐analysis due to alternate reporting of outcome data; none of these studies reported differences between groups in ED/hospital admissions | 17 | Lowa,b |
| Mortality Follow‐up: 3 to 24 months | We are uncertain of the effects of mixed interventions on mortality. Seven studies (1776 participants) were included in a meta‐analysis. Risk ratio was 0.93 (95% CI 0.67 to 1.30), with an anticipated absolute effect of 0.9% fewer deaths (4.1% fewer to 3.8% more). One study was excluded from meta‐analysis due to incomplete information | 8 | Very lowa,b,d |
| CI: confidence interval; SF‐36: Short Form Health Survey‐36. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures, and because of high levels of heterogeneity in results. cOne mark deducted due to high or unclear risk of bias across multiple domains and inclusion of studies at risk of attrition bias in meta‐analysis. dOne mark deducted due to imprecision with limits of confidence intervals including both substantial potential benefit and harm. | |||
| Educational interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving educational components only Comparison: usual care | |||
| Outcomes | Impacts | No of studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: N/A | No studies that evaluated medication‐taking ability were found | ‐ | ‐ |
| Medication adherence (dichotomous) Follow‐up: 1 to 6 months | We are uncertain of the effects of educational interventions on the proportion of people who are adherent Two studies (182 participants) using dichotomous measures of adherence were included in a meta‐analysis. Risk ratio was 1.66 (95% CI 1.33 to 2.06), indicating that educational interventions increased the absolute number of adherent participants by 31.1% (15.6% to 50.1% more) Three studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study reported that the intervention increased the number of resolved medication issues (including non‐adherence); 2 studies reported no significant effect on adherence | 5 | Very lowa,b,c |
| Medication adherence (continuous) Follow‐up: 1 to 12 months | Educational interventions may have little or no impact on medication adherence measured by continuous adherence outcomes (e.g. proportion of pills dispensed or taken) Five studies (1165 participants) using continuous measures of adherence were included in a meta‐analysis. Standardised mean difference was 0.16 (95% CI ‐0.12 to 0.43), indicating that the mean adherence score in the intervention group was 0.16 standard deviations higher (0.12 lower to 0.43 higher) than in the usual care group Four studies were excluded from the meta‐analysis: 2 due to alternate reporting of outcome data (neither showed a difference between groups); 2 did not report results | 9 | Lowa,b |
| Health‐related quality of life Follow‐up: 3 to 12 months | Educational interventions probably have little or no effect on health‐related quality of life, with all 6 studies reporting no differences between groups. Meta‐analysis was not possible due to differences in scales used and differences in reporting of results | 6 | Moderatea |
| ED/Hospital admissions Follow‐up: 4 to 28 weeks | Educational interventions probably have little or no effect on ED/hospital admissions. Three studies (554 participants) were included in a meta‐analysis. Risk ratio was 1.02 (95% CI 0.71 to 1.48), indicating no change in the number of patients admitted to ED/hospital. One further study not included in meta‐analysis, reporting mean number of days in hospital, found no differences between groups | 4 | Moderatea |
| Mortality Follow‐up: N/A | No studies that evaluated the effects of educational interventions on mortality were found | ‐ | ‐ |
| CI: confidence interval; ED: emergency department. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures, and because of high levels of heterogeneity in results. cOne mark deducted due to imprecision ‐ small total number of participants and only two studies in meta‐analysis (one had very wide confidence interval and low events) plus the number of adherent patients (i.e. events) were not clearly reported in the two studies excluded from meta‐analysis. | |||
| Behavioural interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving behavioural components only Comparison: usual care | |||
| Outcomes | Impacts | No of studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: N/A | No studies that evaluated medication‐taking ability were found | ‐ | ‐ |
| Medication adherence (dichotomous) Follow‐up: 3 to 18 months | Behavioural interventions may improve the proportion of people who are adherent (dichotomous adherence outcome) Four studies (528 participants) were included in a meta‐analysis. Risk ratio was 1.22 (95% CI 1.07 to 1.38), indicating behavioural interventions increased the absolute number of adherent participants by 10.5% (3.3% to 18.1% more) | 4 | Lowa,b |
| Medication adherence (continuous) Follow‐up: 6 to 12 months | We are uncertain of the effects of behavioural interventions on medication adherence when continuous measures of adherence are used Three studies were identified, but results could not be pooled in a meta‐analysis due to differences in reporting. All 3 reported significant impact on medication adherence, 2 showed large effects on adherence based on pill count, and 1 showed moderate improvement in self‐reported adherence using daily logbooks to calculate percentage of days adherent | 3 | Very lowa,b,c |
| Health‐related quality of life Follow‐up: 3 months | We are uncertain of the effects of behavioural interventions on health‐related quality of life. Only 1 study was identified, which found that the intervention resulted in worsening quality of life using the Minnesota Living With Heart Failure Questionnaire | 1 | Very lowa,d,e |
| ED/Hospital admissions Follow‐up: 3 to 6 months | We are uncertain of the effects of behavioural interventions on ED/hospital admissions. Two studies (70 participants) were included in a meta‐analysis. Risk ratio was 0.21 (95% CI 0.08 to 0.55), indicating behavioural interventions may reduce the absolute number of patients admitted to ED/hospital by 42.9% (49.9% to 24.4% fewer) | 2 | Very lowa,f |
| Mortality Follow‐up: N/A | No studies that evaluated the effects of behavioural interventions on mortality were found | ‐ | ‐ |
| CI: confidence interval; ED: emergency department. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures. cOne mark deducted due to low participant numbers. dOne mark deducted due to indirectness of evidence as the Minnesota Living With Heart Failure Questionnaire is specific for heart failure populations and results may not be generalisable to general population of older people. eOne mark deducted due to low participant numbers from a single small study. fTwo marks deducted due to low participant numbers and low number of events. | |||
Background
Description of the condition
Older people, conventionally defined as those aged 65 years and older, often have multiple chronic health problems that require ongoing healthcare interventions (Hilmer 2007; WHO 2000). Increasing multi‐morbidity and an expanding evidence base supporting multi‐drug regimens in the management of many chronic diseases mean that polypharmacy (use of multiple medications) is often unavoidable in older people. Polypharmacy has a range of definitions but is commonly defined as the use of four or more medications (Department of Health (UK) 2001; Patterson 2014). The prevalence of polypharmacy is increasing. For example, in the United Kingdom, the number of adults aged over 65 years and taking five or more medications daily has quadrupled from 12% to 49% over the past two decades (Gao 2017). There is also a substantial subgroup of the older population who are prescribed an average of 10 or more different medications, which is sometimes referred to as hyperpolypharmacy (Elliott 2014).
Medication‐taking ability refers to a person’s ability to accurately follow a prescribed medication regimen. It includes knowing what medications to take and when to take them and being able to correctly administer the medication (Maddigan 2003). Managing multiple long‐term medications can be a complex and challenging task, especially for older people, who may experience a decline in the cognitive and physical abilities required for taking medication (Barbas 2001; Beckman 2005). More than a quarter of older people experience difficulties when opening medication packages, including opening bottles and removing medication from blister packs (Philbert 2014). Older people with visual impairment are more than twice as likely to require help in managing their medication as those without visual impairment (McCann 2012). Many older people receive assistance from informal or non‐professional carers when taking medication (ACSQHC 2012). Thus, interventions that aim to improve medication‐taking ability in older adults may need to target carers as well as consumers.
Medication adherence refers to the extent to which a person’s medication‐taking behaviour corresponds with agreed upon treatment recommendations from a healthcare provider (WHO 2003). Non‐adherence refers to deviations from that agreed upon treatment and includes under‐utilisation, over‐utilisation, and incorrect use of medication. There are two broad types of non‐adherence: unintentional non‐adherence – which may be due to factors such as forgetfulness, lack of understanding, physical problems, or the complexity of the regimen; and intentional non‐adherence – which occurs when a person decides not to take his or her treatment as instructed (Wroe 2002). A person is generally considered adherent if he or she takes between 80% and 120% of prescribed medication over a given time period (WHO 2003). Non‐adherence to medications has been reported in up to 50% of older people in different countries and settings (George 2006; Gilbert 1993;Gray 2001; Hemminki 1975;Lau 1996; Lee 2010; Mansur 2008; McElnay 1997;Okuno 1999;Sewitch 2008; Spagnoli 1989;Stoehr 2008; Thorpe 2009; Vik 2006). The World Health Organization (WHO) has recognised the importance of enhancing adherence as a strategy to tackle chronic health conditions effectively (WHO 2003).
Consequences of poor medication‐taking ability and of non‐adherence may include suboptimal response to treatment, recurrence of illness, adverse drug events (ADEs), increased healthcare service utilisation, unplanned hospitalisations, increased morbidity and mortality, and increased healthcare costs (Balkrishnan 2003; Col 1990;DiMatteo 2002;Howard 2003; Leendertse 2008; Tafreshi 1999). Among older adults, ADEs are a significant and increasing problem (Burgess 2005;Elliott 2014). Almost a quarter of preventable ADEs in older people are attributable to consumer error (Field 2007; Gurwitz 2003). Between US$100 and US$300 billion of avoidable healthcare costs has been attributed to non‐adherence in the United States annually (IMS 2013).
Medication‐taking ability and adherence are influenced by a range of factors related to healthcare consumers, their therapies, their medical conditions, social factors, and healthcare provider‐, and health system‐related factors (Balkrishnan 1998; Jin 2008; WHO 2003). Medication‐taking ability and adherence can be inter‐related. For example non‐adherence may result from a patient being unable to follow instructions or remove medications from packaging. Age itself is generally not an independent predictor of poor medication‐taking ability nor of non‐adherence (DiMatteo 2004; Vik 2004). Nevertheless, the prevalence of risk factors for medication use problems increases with age (Col 1990). These include polypharmacy (Gray 2001;Vik 2006), medication regimen complexity (Corsonello 2009; Jansa 2010; Vik 2006), cognitive and functional decline (Gray 2001; Hutchison 2006; Spiers 1995;Vik 2006), inadequate contact with health professionals (George 2006), depressive symptoms (Vik 2006), poor social support (DiMatteo 2000; Spiers 1995), and absence of assistance with administration of medications (Vik 2006). The risk factors for suboptimal use of medications by older people have been studied extensively in cross‐sectional studies (George 2006; Gilbert 1993; Gray 2001; Hemminki 1975; Jerant 2011; Lau 1996; McElnay 1997; Okuno 1999; Sears 2012; Spagnoli 1989; Tavares 2013; Vik 2006). Many adverse health outcomes may be preventable if appropriate measures are taken to address these risk factors and to optimise medication‐taking ability and adherence (George 2008; Jokanovic 2016; Sorensen 2004).
Description of the intervention
A range of simple to complex behavioural and educational interventions, given alone or in combination, have been tested for improving the medication‐taking ability and adherence of consumers (George 2008). Behavioural strategies include:
-
alarm/beeper;
-
calendar/diary;
-
reminder chart/medication list;
-
large print labels;
-
packaging change;
-
multi‐compartment pillbox/calendar pack/compliance aid (also known as dose administration aid (DAA));
-
contracting (verbal or written agreement);
-
adherence monitoring with or without feedback;
-
reminders (mail, telephone, email);
-
inpatient programs of self‐administration of medications;
-
simplification of medication regimens;
-
skill building (supervised, group);
-
tailoring (routinisation); and
-
follow‐up (home visit, scheduled clinic visit, video/teleconferencing).
Educational strategies comprise group (inpatient, family, and support group) and/or individual (verbal, audiovisual, visual, written, telephone, mail) education provided by physicians, pharmacists, nurses, allied health professionals, and others.
How the intervention might work
Behavioural and educational interventions, used alone or in combination, are intended to improve the ability of older people (and/or the ability of their carers) to manage medications and adhere to medication regimens. Interventions may target medication‐taking ability, medication adherence, or both.
These interventions may also lead to improvements in knowledge about medications and in confidence regarding medication management; greater satisfaction with treatment; better health‐related quality of life (HRQoL); reductions in the incidence of ADEs; and reductions in health service utilisation.
Why it is important to do this review
Older people taking multiple medications represent a large and growing proportion of consumers seen by health professionals in clinical practice. They are also the group most likely to experience ADEs. Evidence from well‐designed studies testing interventions to improve medication‐taking ability and adherence among older people prescribed multiple long‐term medications could provide valuable information for practitioners, researchers, and consumers to help optimise medication use among older people living in the community.
Interventions to improve medication adherence have been widely investigated (Nieuwlaat 2014; Ryan 2014). However previous reviews of interventions focusing on medication adherence in older people taking multiple medications are more than 10 years old (George 2008), or they have identified few studies for inclusion (Patton 2017; Zelko 2016). Older people form a heterogeneous population in terms of their medication consumption and disease patterns; therefore studies recruiting relatively homogenous samples of people experiencing one specific disease or consuming one type of medication have limited generalisability.
To date, no systematic review has included measures of medication‐taking other than adherence, such as ability to manage medications. Standardised methods for measuring the ability of people to manage medications have been developed (Elliott 2009; Elliott 2015), some of which have been used in studies of medication use in older people (Lam 2011). Two of the best‐studied assessment tools for evaluating medication‐taking are the Drug Regimen Unassisted Grading Scale (DRUGS) (Edelberg 1999), which utilises a person’s own medications, and the Medication Management Ability Assessment (MMAA) (Patterson 2002), which uses a simulated medication regimen.
Our review will focus on interventions to improve medication‐taking ability or adherence, or both, in older adults who are prescribed multiple medications, or their carers (who are not health professionals).
This review will complement previous Cochrane reviews looking at interventions for improving medication adherence in the general population (Nieuwlaat 2014), including the impact of dose reminder packaging (Mahtani 2011), and interventions for improving clinical outcomes in people with multi‐morbidities (Smith 2016). The appropriateness of people's medication regimens is dependent on a number of factors and will not be considered as part of this review, but only their ability to take (or use) the medications and their adherence to the agreed regimen. There has been a previous Cochrane review of interventions targeted at health professionals, designed to improve the appropriateness of prescribing and polypharmacy (Patterson 2014).
Objectives
To evaluate the effectiveness of interventions designed to improve medication‐taking ability and/or medication adherence in older community‐dwelling people (or their carers) whose treatment consists of multiple long‐term prescribed medications.
Methods
Criteria for considering studies for this review
Types of studies
We included randomised controlled trials (RCTs), cluster‐RCTs, and quasi‐RCTs, as specified by the Cochrane Consumers and Communication Group (CCCRG 2014).
Types of participants
We included studies in which:
-
most participants (≥ 80%) were aged 65 years and over, or the mean/median age was over 65 years. Studies were identified that did not meet these criteria but had relevant data regarding older people that could be extracted separately; these were also included;
-
participants were living in the community or were discharged from a hospital or other healthcare facility to the community (living in the community includes in a person’s own home or retirement village/independent living unit, with or without additional support; it does not include situations in which professional carers or nurses administer the person's medications, such as in nursing homes, residential care facilities or full nursing care in the home); and
-
participants used at least four long‐term regular prescription medications, or the group mean/median was greater than four (irrespective of participants’ number of medical conditions).
Studies that involved carers of consumers who met these criteria were also included. Carers were defined as “people who provide unpaid care and support to family members and friends who have disability, mental illness, a chronic condition, terminal illness, or general frailty” (ACSQHC 2012).
Types of interventions
We included studies that tested single interventions or combinations of interventions directed at the consumer or the carer that sought to improve medication‐taking ability and/or adherence by the consumer.
Examples included:
-
support for behaviour change;
-
provision of medication aids (e.g. DAAs, medication lists);
-
medication regimen simplification;
-
remote monitoring of medication use with or without feedback;
-
facilitation of communication and decision‐making about medications;
-
provision of information or education; and
-
acquisition of skills and competencies.
This list of interventions is not exhaustive. Therefore the search strategy (see Appendix 1) also focused on terms that described the outcomes of interest to avoid missing potentially relevant studies that tested novel interventions.
We included the following comparisons.
-
Interventions to improve medication‐taking ability and/or adherence versus standard or usual care.
-
One form of intervention to improve medication‐taking ability and/or adherence versus another − including simple versus complex interventions.
In future updates, we will consider including interventions to improve medication‐taking ability and/or adherence versus no intervention, but for this review, we identified no studies of this nature.
Types of outcome measures
Primary outcomes
This review focused on two outcomes directly related to medication‐taking behaviour of older adults (or their carers): ability to manage medications and adherence to medication regimens. To be eligible for inclusion, studies had to have assessed at least one of these outcome measures for at least four regular prescribed medications (which could be the person’s own medications or, for assessment of ability to manage medication, a validated, simulated medication regimen instrument) (Elliott 2009). These two outcomes were evaluated separately.
Ability to manage medications
This outcome assessed participants' (or carers') ability to manage medications using objective and/or subjective measures.
-
Objective measures: direct observation using standardised assessment instruments/methods (e.g. Drug Regimen Unassisted Grading Scale (DRUGS), Medication Management Ability Assessment (MMAA), device technique checklists (Elliott 2006; Elliott 2009; Patterson 2002)).
-
Subjective measures: self‐reported ability or self‐efficacy (e.g. Self‐Efficacy for Appropriate Medication Use Scale (SEAMS)) (Risser 2007).
Adherence to medication regimens
This outcome assessed consumer adherence to prescribed medication regimens, using objective and/or subjective measures.
-
Objective measures: refill data, pharmaceutical claims data, electronic monitoring, biological assay, measure of used/unused medications (e.g. pill count).
-
Subjective measures: self‐report of missed/used doses, validated questionnaires (e.g. Morisky scale (Morisky 1986)).
If an included study measured adherence and/or ability by using more than one type of outcome measure, review authors extracted the most reliable measure (i.e. objective measures were preferentially reported over subjective measures).
Secondary outcomes
We analysed the following secondary outcomes from studies that also measured at least one of the primary outcomes listed above.
-
Consumer (or carer) knowledge about their medications.
-
Consumer (or carer) satisfaction with the intervention.
-
Health‐related quality of life (HRQoL).
-
Adverse clinical health outcomes (e.g. unplanned hospital or emergency department presentations, general practitioner visits, ADEs).
-
Condition‐specific outcomes (e.g. cardiovascular events, blood pressure, blood glucose levels, lung function).
-
Cost‐effectiveness of the intervention.
Timing of outcome assessment
For adherence outcomes, the minimum duration of follow‐up was four weeks. For medication‐taking ability outcomes, a follow‐up period of at least 48 hours after the intervention was required.
If an included study measured adherence and/or ability more than once, we extracted the outcome measure with the longest follow‐up.
Search methods for identification of studies
Electronic searches
We searched the following electronic databases in June 2019, using search strategies tailored to each database.
-
Cochrane Central Register of Controlled Trials (CENTRAL), in the Cochrane Library (to 2019).
-
MEDLINE (OvidSP) (1966 to 2019).
-
Embase (OvidSP) (1973 to 2019).
-
PsycINFO (OvidSP) (1967 to 2019).
-
Cumulative Index to Nursing and Allied Health Literature (CINAHL) Plus (EBSCOhost) (1981 to 2019).
-
International Pharmaceutical Abstracts (IPA) (ProQuest) (1971 to 2019).
The search strategies are presented in Appendix 1.
We applied no language restrictions (provided title and abstract were in English).
Searching other resources
We searched grey literature and online trial registries in November 2017.
For grey literature, we searched:
-
Joanna Briggs Institute Evidence Based Practice Database; and
-
conference proceedings (Scopus).
We checked the reference lists of included studies and of previously published relevant systematic reviews to locate potentially eligible studies that were not identified via electronic searches.
We also searched the following online trial registries for ongoing and recently completed studies.
-
World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP).
-
ClinicalTrials.gov.
-
ClinicalTrials.com.
-
TrialsCentral.
-
Australian New Zealand Clinical Trials Registry (ANZCTR).
-
United Kingdom Clinical Research Network (UKCRN).
-
Networked Digital Library of Theses and Dissertations (NDLTD).
-
International Standard Randomised Controlled Trial Number (ISRCTN) registry.
Non‐English language studies were translated and included if they met the eligibility criteria. Studies that were translated are noted in the Characteristics of included studies tables.
Data collection and analysis
Selection of studies
Two review authors (AC and KP, LK, or JG) independently screened abstracts and retrieved the full text of any papers identified as potentially relevant by at least one review author. Two review authors independently screened full‐text articles for inclusion or exclusion (AC and RE, KP, LK, or JG), with discrepancies resolved by discussion and by consulting a third review author if necessary to reach consensus (RE or JG). Review authors were not responsible for screening studies in which they were involved or that they were associated with. We listed as excluded studies all potentially relevant papers excluded from the review and provided reasons in the Characteristics of excluded studies table. We also provided citation details and any available information about ongoing studies, and we collated and reported details of duplicate publications, so that each study (rather than each report or manuscript) was the unit of interest in the review. We reported the screening and selection process in an adapted PRISMA flow chart (Liberati 2009).
Data extraction and management
Two review authors extracted data independently from included studies (AC and, RE, KP, LK, or JG). We resolved any discrepancies by discussion until consensus was reached, or through consultation with a third review author when necessary (RE or JG). We developed and piloted a data extraction form using the Cochrane Consumers and Communication Review Group Data Extraction Template (cccrg.cochrane.org/author-resources). Data extracted included the following study details: aim of the intervention, study design, study population, intervention details, control/comparison group(s), outcome(s), and follow‐up period(s). One review author (AC) entered all extracted data into Review Manager version 5.3 (RevMan 2014), and an independent person checked these data for accuracy against the data extraction sheets.
Assessment of risk of bias in included studies
We assessed and reported on the methodological risks of bias of included studies in accordance with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), and according to Cochrane Consumers and Communication Group guidelines (CCCRG 2014), which recommend explicit reporting of the following individual elements for RCTs: random sequence generation; allocation sequence concealment; blinding (participants, personnel); blinding (outcome assessment); completeness of outcome data; selective outcome reporting; and other sources of bias. Other sources of bias included concerns related to sample size, fidelity, potential conflict of interest (e.g. influence of funding bodies), changes to methods (e.g. trial being ceased early), or trial results not published in a peer‐reviewed journal (e.g. thesis). We considered blinding separately for different outcomes when appropriate (e.g. blinding may have the potential to differently affect subjective versus objective outcome measures). We judged each item as being at high, low, or unclear risk of bias, as set out in the criteria provided by Higgins 2011, and we provided a quote or information from the study report and a justification for our judgement for each item in the "Risk of bias" table.
We deemed studies to be at highest risk of bias if they were scored at high or unclear risk of bias for the sequence generation or the allocation concealment domain, based on growing empirical evidence that these factors are particularly important potential sources of bias (Higgins 2011).
We assessed and reported quasi‐RCTs as being at high risk of bias on the random sequence generation item of the "Risk of bias" tool. For cluster‐RCTs, we also assessed and reported risk of bias associated with another domain: selective recruitment of cluster participants.
In all cases, two review authors independently assessed the risk of bias of included studies, with disagreements resolved by discussion to reach consensus. We contacted study authors for additional information about the included studies, or for clarification of study methods as required. We incorporated results of the "Risk of bias" assessment into the review through standard tables and systematic narrative description and commentary about each of the elements, leading to an overall assessment of the risk of bias of included studies and a judgement about the internal validity of review results.
Measures of treatment effect
We considered the primary outcomes as dichotomous variables when possible, that is, the person (or the carer) was assessed as able to manage medications or not, and similarly to have satisfactory adherence (80% to 120%) or not (< 80% or > 120%). If a study did not report its outcome as dichotomous, we extracted and analysed continuous outcomes.
For dichotomous outcomes, we analysed data based on the number of events and the number of people assessed in the intervention and comparison groups. We used these data to calculate the risk ratio (RR) and the 95% confidence interval (CI). Given heterogeneity in study measures, we analysed data for continuous measures using the standardised mean difference (SMD) and 95% CI approach via the inverse variance method in Review Manager 5.
Unit of analysis issues
For included cluster‐RCTs, we checked for unit of analysis errors. If errors were found, and if sufficient information was available, we re‐analysed data using the appropriate unit of analysis by taking into account the intracluster correlation coefficient (ICC). We obtained estimates of the ICC by contacting authors of included studies. When this was not possible, we reported effect estimates and annotated "unit of analysis errors." For future updates, we may impute missing ICCs using estimates from external sources, but this was not required for any of the trials included in this review.
Of the six cluster‐RCTs, three reported ICCs but did not report effective sample sizes. We recalculated effective sample sizes based on information reported in each study and divided the reported sample size by the design effect (Higgins 2011). We reported the adjusted sample sizes in meta‐analyses and reduced the weightings given to these studies.
Muth 2016 reported an ICC of 0.00; thus no adjustment was required.
Moral 2015 reported an ICC of 0.05 and had an average cluster size of 5.4 for intervention and 6.0 for control. There were 70 participants in the intervention group and 84 participants in the control group, which we adjusted to 57 for intervention and 67 for control for all outcomes.
Willeboordse 2017 reported an ICC of 0.08 but was not included in any meta‐analyses due to alternate reporting methods; thus effective sample sizes were not calculated.
For the three studies that did not report an ICC (Bernsten 2001;Volume 2001; Wood 1992), we contacted study authors for further information. As we received no response, we could make no adjustments. We conducted sensitivity analyses while excluding these studies to adjust for possible unit of analysis errors.
Dealing with missing data
We attempted to contact study authors to obtain missing data (participant, outcome, or summary data). When possible, we conducted analyses of participant data on an intention‐to‐treat (ITT) basis; otherwise we analysed data as reported. We reported on levels of loss to follow‐up and assessed this as a potential source of bias.
For missing outcome or summary data, we planned to impute missing data when possible and to report any assumptions, but this was not required or possible for any included studies.
We conducted sensitivity analyses that excluded studies presenting data with loss to follow‐up greater than 20% for the primary outcomes (medication‐taking ability and/or medication adherence) including total reported lost to follow‐up and differential loss to follow‐up between groups. This was due to potential serious threats to validity associated with high proportions of participants lost to follow‐up (Sackett 2000).
Assessment of heterogeneity
We identified substantial variations in types of interventions, populations studied, and study designs and settings. When studies were considered similar enough to enable data pooling via meta‐analysis, we assessed the degree of heterogeneity by visually inspecting forest plots and examining the Chi² test for heterogeneity. We quantified heterogeneity by using the I² statistic. We considered an I² value of 50% or more to represent substantial levels of heterogeneity, but we interpreted this value in light of the size and direction of effects and the strength of evidence for heterogeneity, based on the P value from the Chi² test (Higgins 2011). When heterogeneity was present in pooled effect estimates, we explored possible reasons for variability by conducting subgroup analyses.
When we detected substantial heterogeneity, particularly in relation to types of outcome measures or methods of reporting outcome measures, we did not report pooled results from meta‐analysis but instead used a narrative approach to data synthesis.
Assessment of reporting biases
We assessed reporting bias qualitatively based on the characteristics of included studies (e.g. if only small studies with positive findings were identified for inclusion), and based on information obtained by contacting authors of studies which suggested there might have been relevant unpublished data or studies.
We did not construct funnel plots to investigate publication bias because we found insufficient studies per outcome and intervention type and because multiple studies were not suitable for inclusion in meta‐analyses.
For future updates, if we should identify sufficient studies (at least 10) for inclusion in the review, we will construct a funnel plot to investigate small‐study effects and to formally test for funnel plot asymmetry, with test selection based on advice provided in Higgins 2011, while bearing in mind there may be several reasons for funnel plot asymmetry when results are interpreted.
Data synthesis
We conducted meta‐analyses on extracted data for some outcomes. Due to variability in the interventions of included studies, we used a random‐effects model for all meta‐analyses.
For studies not included in meta‐analyses, and for outcomes for which we were unable to pool data, we have presented data in tables and have narratively summarised the results for each outcome.
Subgroup analysis and investigation of heterogeneity
We conducted subgroup analyses to investigate heterogeneity of mixed interventions for medication adherence. We conducted three planned subgroup analyses.
-
Duration of intervention (short versus long).
-
Type of outcome measure (objective versus subjective).
-
Health professional group/system delivering the intervention (e.g. pharmacist versus nurse versus medical professional versus automated).
We were not able to conduct additional planned subgroup analyses to investigate heterogeneity because we found insufficient studies or poor reporting of participant characteristics. For future updates, should more studies be included (especially for other outcomes or other intervention types), we plan to also look at the following.
-
Duration of follow‐up (short, medium, and long term) as described under "Timing of outcome assessment."
-
Person managing the medication (consumer versus carer).
-
Number of medications (up to 10 versus 11 or more medications).
-
Frailty and/or functional ability (e.g. level of home assistance required) and/or cognitive function/ability.
Sensitivity analysis
We conducted sensitivity analyses for the primary outcomes that excluded studies assessed to have losses to follow‐up greater than 20%, and that excluded studies with unit of analysis errors. We had planned to also conduct sensitivity analyses that excluded studies with high risk of bias; however there were too few included studies assessed as having low risk of bias. Future updates of this review should conduct these sensitivity analyses if sufficient studies with low risk of bias are identified.
"Summary of findings" table
We prepared "Summary of findings" tables to present results of meta‐analyses based on methods described in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2011). We presented results of meta‐analyses for the major comparisons of the review and for each of the primary outcomes, as outlined under Types of outcome measures. We provided a source and a rationale for each assumed risk cited in the tables and used Grades of Recommendation, Assessment, Development and Evaluation (GRADE) criteria to rank the quality of evidence using GRADEprofile (GRADEpro) software (Schünemann 2011). When meta‐analysis was not possible, we present results in a narrative "Summary of findings" table format, such as that used by Chan 2011.
Ensuring relevance to decisions in health care
Authors of this review received feedback from a consumer referee and a health professional as part of the standard editorial process of the Cochrane Consumers and Communication Group.
Results
Description of studies
See Characteristics of included studies,Characteristics of excluded studies,Characteristics of ongoing studies, and Characteristics of studies awaiting classification.
Results of the search
The database search yielded 27,854 titles. We found 65 additional records through a search of grey literature. After removing duplicates, we screened 19,192 studies and reviewed 482 full‐text articles. We excluded 373 studies that did not meet the inclusion criteria and recorded our reasons for exclusion. We included 50 independent studies (from 68 citations); 40 were randomised controlled trials (RCTs), four were considered quasi‐RCTs (Begley 1997; Shimp 2012; Volume 2001; Winland‐Brown 2000), and six were cluster‐RCTs (Bernsten 2001; Moral 2015; Muth 2016; Volume 2001; Willeboordse 2017; Wood 1992). Eight studies (from 12 citations) are ongoing (see Characteristics of ongoing studies). Twenty‐six studies (from 29 citations) are awaiting classification; eight have no published results; 15 may be eligible for inclusion but provide insufficient information to allow determination of eligibility; and three will be included in the next update of this review (see Characteristics of studies awaiting classification) (Char 2017; Marusic 2018; Muth 2018). Refer to Figure 1 for a PRISMA diagram.
Study flow diagram.
Included studies
Participants
A total of 14,269 participants were included in the 50 studies. Fifteen studies involved fewer than 100 participants, and six studies involved more than 500 participants. In 38 studies, the intervention was directed at patients; in one study, the intervention was directed at family caregivers (George 2016); and in 11 studies, the intervention involved both patients and caregivers. The mean/median age of included patients ranged from 65.6 to 87.0 years, and 52.4% (6893/13,143) of patients were female (three studies did not provide clear details on gender). The mean/median number of medications ranged from 4.2 to 16.3, but the definition of 'medication' varied greatly between studies and was poorly described in 24 studies (48%). Eighteen studies clearly referred to prescribed medications, but many restricted the count to regular and/or oral medications only. Non‐prescription/over‐the‐counter (OTC) medications were included in the total count in five studies (Haag 2016; Khdour 2009; Krska 2001; Lingler 2016; Marek 2013); these were reported separately in three studies (Begley 1997; Chrischilles 2014; Volume 2001). Four studies did not provide mean/median values but were included based on the published range of the number of medications being taken (Winland‐Brown 2000), or on the fact that inclusion criteria ‐ Hale 2016 ‐ or additional information provided by study authors ‐ Blalock 2010, Shively 2013 ‐ indicated that the mean/median number of medications would be greater than four. One study was included because the subgroup of people taking more than eight medications met our inclusion criteria (Truelove 2015).
Sixteen studies reported some measure of frailty and/or functional ability for included participants, but variation in the scales used prevented comparison. Twenty studies included a measure of cognitive function of participants or listed the proportion of people with cognitive impairment, but the heterogeneous nature of reporting cognitive impairment prevented comparison. Seventeen studies excluded people with cognitive impairment, and 13 did not specify any details. The total mean/median number of chronic conditions (or co‐morbidities) was mentioned in only 10 studies and ranged from three to nine chronic conditions.
Setting
Included studies were carried out across four continents: North America, Europe, Asia, and Australia (see Table A). Most studies were conducted in the USA (21), the UK (8), Canada (5), and Australia (3). Ten studies were conducted in European countries: Spain (4), Croatia (1), Denmark (1), Germany (1), the Netherlands (1), Portugal (1), and Switzerland (1). Two studies were conducted in Asian countries: China (1) and Singapore (1). One study was conducted across seven countries (Bernsten 2001).
Study healthcare settings were categorised according to where the interventions were initiated during the patient's healthcare journey. Twenty‐six studies were initiated at the interface between hospital and community: in hospital (2), immediately before discharge (11), post discharge (6), or in hospital outpatient clinics (7). Twenty‐four studies were initiated in the community/primary care setting including general practice/medical clinics/centres (11), community pharmacies (5), home healthcare services (2), a university clinic (1), an independent living facility (1), and in the home (4, with 2 delivered online and 2 involving visiting health professionals).
Table A. Study design, setting, and participants
| Study ID | Study design | Target participants | Country | Stage of the patient's healthcare journey/healthcare setting where the intervention was initiated | |
| Hospital/Communityinterface | Community/Primary care | ||||
| Al‐Rashed 2002 | RCT | Patient | UK | Discharge | |
| Begley 1997 | RCT | Patient | UK | Post discharge | |
| Bernsten 2001 | Cluster‐RCT | Patient | 7 countries | Pharmacy | |
| Blalock 2010 | RCT | Patient | USA | Pharmacy | |
| Bond 2007 | RCT | Patient | UK | Pharmacy | |
| Cargill 1992 | RCT | Patient + Carer | USA | Outpatient clinic* | |
| Chrischilles 2014 | RCT | Patient | USA | Home (online) | |
| Cohen 2011 | RCT | Patient | USA | Medical centre | |
| Cossette 2015 | RCT | Patient | Canada | ED discharge | |
| George 2016 | RCT | Carer | USA | Online (home) | |
| Grymonpre 2001 | RCT | Patient | Canada | Health clinic | |
| Haag 2016 | RCT | Patient | USA | Post discharge | |
| Hale 2016 | RCT | Patient | USA | Post discharge | |
| Hanlon 1996 | RCT | Patient + Carer | USA | General medicine clinic* | |
| Holland 2007 | RCT | Patient + Carer | UK | Post discharge | |
| Khdour 2009 | RCT | Patient | UK | Outpatient clinic | |
| Krska 2001 | RCT | Patient | UK | General practice clinic | |
| Lee 2006 | RCT | Patient | USA | Outpatient pharmacy* | |
| Lim 2004 | RCT | Patient | Singapore | Outpatient clinic | |
| Lingler 2016 | RCT | Patient + Carer | USA | Community | |
| Lipton 1994 | RCT | Patient + Carer | USA | Discharge | |
| Lopez Cabezas 2006 | RCT | Patient | Spain | Discharge | |
| Manning 2007 | RCT | Patient | USA | Discharge | |
| Marek 2013 | RCT | Patient | USA | Home healthcare service | |
| Marusic 2013 | RCT | Patient | Croatia | Discharge | |
| Messerli 2016 | RCT | Patient | Switzerland | Pharmacy | |
| Moral 2015 | Cluster‐RCT | Patient | Spain | General practice | |
| Morales Suarez‐Vurela 2009 | RCT | Patient + Carer | Spain | Home healthcare service | |
| Murray 1993 | RCT | Patient | USA | Outreach centre | |
| Muth 2016 | Cluster‐RCT | Patient | Germany | General practice | |
| Nascimento 2016 | RCT | Patient + Carer | Portugal | Diabetes clinic | |
| Naunton 2003 | RCT | Patient + Carer | Australia | Post discharge | |
| Nazareth 2001 | RCT | Patient + Carer | UK | Discharge | |
| Olesen 2014 | RCT | Patient | Denmark | Community | |
| Pandey 2017 | RCT | Patient | Canada | Post discharge | |
| Pereles 1996 | RCT | Patient | Canada | Inpatient | |
| Rich 1996 | RCT | Patient | USA | Discharge | |
| Saez de la Fuente 2011 | RCT | Patient + Carer | Spain | Discharge | |
| Shimp 2012 | RCT | Patient | USA | University clinic | |
| Shively 2013 | RCT | Patient | USA | Primary care clinic | |
| Taylor 2003 | RCT | Patient | USA | General practice clinic | |
| Truelove 2015 | RCT | Patient | Australia | General practice clinic | |
| Vinluan 2015 | RCT | Patient | USA | Discharge | |
| Volume 2001 | Cluster‐RCT | Patient | Canada | Pharmacy | |
| Willeboordse 2017 | Cluster‐RCT | Patient | Netherlands | General practice clinic | |
| Williams 2012 | RCT | Patient | Australia | Outpatient clinic | |
| Winland‐Brown 2000 | RCT | Patient | USA | Independent living facility | |
| Wood 1992 | Cluster‐RCT | Patient | UK | Inpatient | |
| Wu 2006 | RCT | Patient + Carer | China | Outpatient clinic | |
| Young 2016 | RCT | Patient | USA | Discharge | |
*Unclear whether health service provided primary or secondary care or both.
ED: emergency department.
RCT: randomised controlled trial.
Interventions
A range of simple to complex interventions were used across the included studies. Due to the heterogeneous nature of the interventions, we categorised them into three broad groups: educational interventions, behavioural interventions, and mixed interventions (both educational and behavioural). These categories have been used in a previous systematic review of interventions to improve medication‐taking in elderly patients prescribed multiple medications (George 2008).
Fourteen studies involved educational interventions comprising medication/health education (provided in writing or verbally) and/or medication review only; seven studies involved behavioural interventions only; and 29 studies had both educational and behavioural elements.
Educational interventions were identified in 38 studies delivering patient or carer education regarding medications and/or health conditions, and in 26 studies involving a review of patient medications.
A range of behavioural interventions were used (either alone or in combination) in 24 studies utilising follow‐up or monitoring; in seven studies providing regimen simplification (Begley 1997;Bernsten 2001;Lim 2004;Lipton 1994;Murray 1993;Olesen 2014; Rich 1996); in five studies practising motivational interviewing (Khdour 2009;Moral 2015;Olesen 2014;Shively 2013;Williams 2012); and in two studies implementing three‐step self‐administration of medications (Pereles 1996; Wood 1992). All participants in six studies utilised DAAs including simple pill boxes (Lee 2006; Marek 2013; Morales Suarez‐Vurela 2009; Winland‐Brown 2000), unit of use packages (Murray 1993), automated dosing devices (Marek 2013; Winland‐Brown 2000), and remotely monitored electronic devices (Hale 2016). Two studies utilised DAAs for some participants as required (Cargill 1992; Naunton 2003), and two studies provided participants with electronic pill reminder devices (Olesen 2014; Young 2016). Other types of interventions included text message adherence reminders (Pandey 2017), a four‐ingredient poly‐pill (Truelove 2015), use of online personal health records (Chrischilles 2014), and a three‐dimensional (3D ‐ durable display at discharge) medication discharge tool that involved patients affixing a tablet/capsule of each medication onto the 3D tool (Manning 2007).
Most interventions were delivered by pharmacists (31 studies), nurses (17 studies), and physicians (15 studies), either alone (31 studies) or in multi‐disciplinary teams of two or more health professionals (15 studies). Two interventions were delivered online (Chrischilles 2014; George 2016); one study involved text message reminders (Pandey 2017); and one study involved a remotely monitored electronic device (Hale 2016). Interventions varied in duration, ranging from one‐off ‐ Al‐Rashed 2002, Blalock 2010, Haag 2016, Lim 2004, Manning 2007, Marusic 2013, Muth 2016, Nascimento 2016, Naunton 2003, Saez de la Fuente 2011, Willeboordse 2017 ‐ to two years ‐ Wu 2006, and were most commonly delivered in the home (face‐to‐face or by phone calls), hospital, medical centre, or community pharmacy. Four studies were delivered across two settings (Lipton 1994; Moral 2015; Nazareth 2001; Rich 1996), and 11 studies involved both face‐to‐face meetings and phone calls (Cargill 1992; Cossette 2015; Khdour 2009; Lingler 2016; Lipton 1994; Lopez Cabezas 2006; Olesen 2014; Shively 2013; Vinluan 2015; Williams 2012; Young 2016).
An additional table summarising the intervention features of all included studies is located at https://latrobe.figshare.com/articles/Additional_tables_Cross_et_al_2020_docx/12247385.
Primary outcomes
Medication‐taking ability was measured in five studies. Four studies used objective measures including a five‐item dexterity test that assessed skills such as ability to open child‐resistant closures on containers (Begley 1997), a medication‐taking behaviour score (Cargill 1992), the Medication Management Instrument for Deficiencies in the Elderly (MedMaIDE) (Lingler 2016), performance in an inpatient self‐administration of medications programme, and/or pharmacist assessment (with input from other team members) of ability to self‐administer medications (Pereles 1996). One study used a subjective measure ‐ a self‐reported assessment of safety in taking medication (Manning 2007). Medication‐taking ability was typically measured at short follow‐up points (e.g. 7 to 14 days; Manning 2007), except for one study, which had an extended measure at 12 months (Begley 1997).
Medication adherence was measured in 48 studies (Table C). Twenty studies used an objective measure of adherence such as pill count (Begley 1997;Cargill 1992;Cohen 2011;Lee 2006;Lopez Cabezas 2006;Marusic 2013;Moral 2015;Murray 1993;Olesen 2014;Pereles 1996;Rich 1996;Williams 2012;Winland‐Brown 2000;Wood 1992), prescription claims/refills (Al‐Rashed 2002;Grymonpre 2001;Messerli 2016;Shimp 2012;Vinluan 2015), or machine‐recorded correct doses (Marek 2013). Twenty‐eight studies used a subjective measure of adherence; 16 of these used an original or modified version of the Morisky Medication Adherence Scale ‐ a validated measure of adherence (Bernsten 2001;Chrischilles 2014;Cossette 2015;George 2016;Haag 2016;Hale 2016;Khdour 2009;Morales Suarez‐Vurela 2009;Muth 2016;Nascimento 2016;Saez de la Fuente 2011;Volume 2001), the Medication Adherence Rating Scale (Bond 2007;Holland 2007;Muth 2016), the Brief Medication Questionnaire (Blalock 2010), and the Medical Outcome Study‐Specific Adherence Scale (Shively 2013). Four studies used structured interviews to enquire about adherence (Hanlon 1996;Lipton 1994;Nazareth 2001;Willeboordse 2017), and six studies asked participants a single question about forgotten or missed doses (Lim 2004;Naunton 2003;Taylor 2003;Truelove 2015;Wu 2006;Young 2016). One study used a patient‐completed daily log‐book of medication consumption (Pandey 2017), and one study included pharmacist review for pharmaceutical care issues including potential or actual adherence issues (Krska 2001). The longest follow‐up time points of post‐intervention adherence outcomes ranged from 1 month to 18 months, with the median time point of 6 months.
Secondary outcomes
Knowledge about medications was measured in 13 studies. Knowledge was assessed most often by asking participants about one or more of the following: name of medication, appearance of medication, purpose of medication, dose, dose frequency/interval, side effects, drug interactions, and special comments or cautions (Al‐Rashed 2002; Begley 1997; Bernsten 2001; Grymonpre 2001; Hanlon 1996; Lim 2004; Manning 2007; Messerli 2016; Nazareth 2001; Pereles 1996; Taylor 2003). One study asked participants on a 5‐point Likert scale if they "knew more about their medicines compared to a year ago" (Bond 2007), and another study assessed medication knowledge as part of a larger chronic obstructive pulmonary disease (COPD) knowledge questionnaire (Khdour 2009).
Satisfaction with the intervention was measured in 13 studies. Six studies used a previously validated measure (George 2016;Hanlon 1996;Lopez Cabezas 2006;Nazareth 2001;Volume 2001;Willeboordse 2017), but no two studies used the same measure. Satisfaction was most commonly assessed on a 5‐point Likert‐type scale (Bernsten 2001;Bond 2007;Hanlon 1996;Manning 2007), or on a 7‐point Likert‐type scale (George 2016;Volume 2001;Willeboordse 2017), which included between one ‐ Manning 2007, Willeboordse 2017 ‐ and 15 items (Bond 2007). Lopez Cabezas 2006 used a 0 to 10 analogue scale, and four studies did not adequately describe the measure used (Holland 2007;Lingler 2016;Naunton 2003;Taylor 2003).
Health‐related quality of life (HRQoL) was measured in 14 studies. The two most common measures were the validated Short Form Health Survey involving 36 items (SF‐36) used in eight studies ‐ Bernsten 2001, Bond 2007, Cohen 2011, Hanlon 1996, Krska 2001, Marek 2013, Taylor 2003, Volume 2001 ‐ and the European Quality of Life 5‐Dimension Instrument (EQ‐5D) used in five studies ‐ Bond 2007, Holland 2007, Lopez Cabezas 2006, Muth 2016, Willeboordse 2017. Other measures used by individual studies included the 12‐item Short Form Health Survey (SF‐12; Willeboordse 2017), as well as disease‐specific quality of life measures including the Minnesota Living With Heart Failure Questionnaire (MLHFQ) ‐ Hale 2016, Holland 2007 ‐ and St George's Respiratory Questionnaire (SGRQ) ‐ Khdour 2009.
Adverse clinical health outcomes were measured in 28 studies and included measures such as emergency department (ED) and/or hospital admissions (Al‐Rashed 2002; Bernsten 2001; Cossette 2015; Haag 2016; Hale 2016; Holland 2007; Khdour 2009; Lipton 1994; Lopez Cabezas 2006; Marusic 2013; Messerli 2016; Muth 2016; Naunton 2003; Nazareth 2001; Olesen 2014; Rich 1996; Saez de la Fuente 2011; Shively 2013; Taylor 2003; Vinluan 2015; Winland‐Brown 2000; Wu 2006; Young 2016), mortality (Holland 2007; Lopez Cabezas 2006; Naunton 2003; Nazareth 2001; Olesen 2014; Saez de la Fuente 2011; Vinluan 2015; Wu 2006), adverse drug reactions (Chrischilles 2014; Hanlon 1996; Lim 2004; Marusic 2013; Murray 1993; Willeboordse 2017), and physician visits (Al‐Rashed 2002; Khdour 2009; Nazareth 2001; Winland‐Brown 2000).
Condition‐specific outcomes were measured in seven studies and included changes in blood pressure (Lee 2006;Taylor 2003;Williams 2012), diabetes control ‐ glycosylated haemoglobin (HbA1c)/blood glucose (Nascimento 2016;Taylor 2003;Williams 2012), low‐density lipoprotein (LDL) cholesterol (Lee 2006;Taylor 2003), falls (Blalock 2010), international normalised ratio (INR) of time taken for blood to clot (Taylor 2003), and renal function (Williams 2012). Two studies reported composite measures of reaching multiple 'health' targets (Bond 2007;Cohen 2011).
Cost‐effectiveness of the intervention was measured in four studies; three studies used costs of the intervention, medicines, hospitalisations, and/or health consultations (Bernsten 2001;Bond 2007;Lopez Cabezas 2006), and one study used US Medicare Part B costs and total hospital inpatient costs (Lipton 1994).
Other outcome measures extracted included medication management problems from a list of eight problems (Chrischilles 2014), a medication deficiency checklist (Lingler 2016), medication errors defined as both prescriber and patient errors (Moral 2015), and medication misadventures defined as one or more medication errors, adverse drug events, or adverse drug reactions (Taylor 2003).
An additional table summarising the type and timing of primary and secondary outcomes assessed by included studies is located at https://latrobe.figshare.com/articles/Additional_tables_Cross_et_al_2020_docx/12247385.
Excluded studies
We excluded 373 studies in total (see Characteristics of excluded studies). We excluded 24 studies because study design did not meet Cochrane criteria for an RCT, a cluster‐RCT, or a quasi‐RCT. We excluded 149 studies on the basis of the age of participants; 88 studies based on the number of regular prescription medications (including 13 studies for which the number of medications was unknown and attempts to contact study authors were unsuccessful); and 10 studies because study authors did not collect information on the number of medications. We excluded 94 studies because they did not include a measure of medication‐taking ability or medication adherence as an outcome, and 8 studies because the follow‐up period for outcome measures was too short (i.e. < 48 hours for medication‐taking ability, < 4 weeks for adherence). We excluded 6 studies because the intervention did not target consumers, along with 4 studies because participants were not community‐dwelling.
Risk of bias in included studies
See Characteristics of included studies table, Figure 2, and Figure 3 for a summary assessment of the risk of bias of included studies.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Allocation
Risk of bias for random sequence generation was low in 22 studies (44%), unclear in 24 studies (48%), and high in four studies (8%). For concealment of allocation, risk of bias was low in 17 studies (34%), unclear in 30 studies (60%), and high in three studies (6%). Selective recruitment of cluster participants was assessed for the six cluster‐RCTs ‐ three were considered at high risk (Moral 2015;Willeboordse 2017;Volume 2001), two were considered at low risk (Muth 2016;Wood 1992), and one was considered at unclear risk of recruitment bias (Bernsten 2001).
Blinding
Blinding of both participants and personnel could not be achieved through the study design in 49 of the 50 studies (98%), leading to high risk of performance bias. One study was considered to have low risk of performance bias, as the intervention was delivered online and both intervention and control participants viewed the same interface and thus were unaware of their allocation (George 2016).
Seventeen studies (34%) stated that there was no blinding of outcome assessment; we considered these studies to have high risk of detection bias. Twenty studies (40%) were assessed as having 'unclear' risk of detection bias due to insufficient details regarding the method of outcome assessment. Studies with unclear detection bias included one study for which data collection was performed "where possible" by a member of staff other than the intervention pharmacist (Bernsten 2001), one study involving caregiver‐reported patient adherence when caregivers were assumed to be unaware of allocation (George 2016), and two studies in which assessors were reported as blinded but contamination from unblinded participants was thought to be highly likely (Saez de la Fuente 2011; Young 2016). We assessed 13 studies (26%) as having low risk of detection bias; five involved an objective measure of the primary outcome (Grymonpre 2001;Marusic 2013;Messerli 2016;Rich 1996; Williams 2012), and eight involved subjective measures but data were collected/analysed by blinded investigators (Bond 2007;Chrischilles 2014;Cossette 2015;Haag 2016;Hanlon 1996;Lipton 1994;Manning 2007;Nazareth 2001).
Incomplete outcome data
Twenty‐two (44%) studies were considered to have incomplete outcome data and therefore high risk of attrition bias – 19 of these cases were due to high loss to follow‐up. A further three (6%) studies were assessed as having high risk of attrition bias due to inconsistency between the average number of medicines and the number assessed for adherence (Grymonpre 2001), inconsistency between the number of patients with adherence reported and the number who saved their medication boxes enabling accurate pill count (Williams 2012), and lack of details regarding attrition of the control group (Shimp 2012). Thirteen studies were assessed as having unclear risk of bias mainly due to low to moderate attrition, which may have had an impact on the results, or insufficient details on number of, or reasons for, attrition. Fifteen studies reported minimal incomplete outcome data and/or adequately addressed this (low risk of bias).
Selective reporting
Fourteen studies (28%) were considered to have high risk of reporting bias ‐ 12 due to missing outcome data (Begley 1997;Blalock 2010;Cohen 2011;Krska 2001;Lim 2004;Lopez Cabezas 2006;Messerli 2016;Morales Suarez‐Vurela 2009;Murray 1993;Olesen 2014;Shimp 2012; Winland‐Brown 2000), one because study authors did not clearly specify how data were obtained (Vinluan 2015), and one because researchers changed the inclusion criteria mid‐way through the study to increase recruitment (Williams 2012).
Twelve studies (24%) were considered to have unclear risk of reporting bias due to minor deviations from study methods (Bond 2007;Grymonpre 2001;Lipton 1994;Wood 1992), missing information in the methods section (Bernsten 2001;Willeboordse 2017), missing baseline data (Lee 2006), and unclear reporting of results (Cargill 1992;Lingler 2016;Marek 2013;Moral 2015;Willeboordse 2017).
Although 24 studies (48%) were assessed as having low risk of selective reporting, 15 of these did not have a published protocol nor trial registration; thus it was difficult to accurately assess reporting bias.
Other potential sources of bias
Four studies were identified as having high risk of other types of bias. One study was assessed as having high risk because it was a research thesis and had not been published in a peer‐reviewed journal (George 2016), two studies because of poor intervention fidelity (Chrischilles 2014;Nazareth 2001), and one study because researchers measured adherence only for the first three medications mentioned by the patient (Lipton 1994).
Twenty‐seven studies were considered to have unclear risk of other types of bias. Sixteen studies did not reach their specified target sample size (Bernsten 2001;Blalock 2010;Bond 2007;Cossette 2015;Hale 2016;Khdour 2009;Lopez Cabezas 2006;Marek 2013;Messerli 2016;Moral 2015;Morales Suarez‐Vurela 2009;Pereles 1996;Truelove 2015;Volume 2001;Willeboordse 2017; Wu 2006); three studies had unbalanced participant groups likely influencing outcomes (Haag 2016; Murray 1993; Winland‐Brown 2000); and four studies had potential conflicts due to funding arrangements (Holland 2007; Shimp 2012) or participant compensation (Messerli 2016; Shively 2013), which may have biased results of the study. Two studies provided limited information regarding intervention fidelity (Manning 2007; Nascimento 2016), and four studies expressed concerns regarding the appropriateness of the adherence assessment (Nascimento 2016; Rich 1996; Pandey 2017; Williams 2012).
Trial authors also noted that 10 studies did not declare a funding source. However, given the differences in journal requirements and the age of those studies, this was not considered to introduce risk of bias for this review.
Effects of interventions
See: Summary of findings 1 Summary of findings: mixed interventions; Summary of findings 2 Summary of findings: educational interventions alone; Summary of findings 3 Summary of findings: behavioural interventions alone
See "Summary of findings" tables for the main comparison.
COMPARISON 1. Intervention versus usual care
Primary outcome ‐ medication‐taking ability
Educational interventions
No studies were identified.
Behavioural interventions
No studies were identified.
Mixed educational and behavioural interventions
Mixed educational and behavioural interventions were identified in five studies, which showed mixed impact on medication‐taking ability (Table 1; low‐certainty evidence). One study involving an educational intervention combined with telephone follow‐up directed at both patients and caregivers (group 3) showed slightly greater improvement in medication‐taking ability compared to usual care (group 1), as measured by a behaviour score at four to six weeks (mean scores presented visually; mean 86/100 versus 74/100; P = 0.01) (Cargill 1992). Another study, also involving an educational intervention plus follow‐up, demonstrated significant decreases in medication management problems, as measured by the Management Instrument for Deficiencies in the Elderly (MedMaIDE), in both intervention and usual care groups but did not report between‐group comparisons (Lingler 2016). Two studies showed no significant difference in patients' medication‐taking ability (Manning 2007; Pereles 1996). In Manning 2007, a medication chart with tablets/capsules affixed and medication discharge education had no significant impact on the number of self‐reported mistakes in taking medication. In Pereles 1996, an inpatient self‐administration of medication programme was reported to have no impact on the number of participants able to self‐administer their medications; however, different methods of assessing the outcome were used for intervention and control groups. One study involving patient‐focused education and regimen simplification did not report results (Begley 1997).
| Study | Measure of medication‐taking ability | Outcome |
|---|---|---|
| Begley 1997 | Objective measure: 5‐task dexterity test (e.g. opening child‐resistant closure), 1 point awarded for each successfully completed activity. Note: no difference across groups at baseline ‐ mean (SD) group A: 7.8 (1.3), group B: 7.5 (1.5), group C: 8.0 (1.4) | Objective measure: follow‐up results not reported |
| Cargill 1992 | Objective measure: behaviour score/100 for congruency between supply of medications on hand and prescribed medications (/40), verbalising correct regimen (/30), maintaining each prescribed med (/20), appropriate use of OTC (/10). Points deducted for sequestering old scripts, inappropriate use of alternative medications, or mixing medications together | Mean read from graph: Control: 74 vs intervention (group 3); 86 vs intervention (group 2); 84 vs intervention (group 3): 86 |
| Lingler 2016 | Objective measure: Medication Management Instrument for Deficiencies in the Elderly (MedMaIDE). MedMaiDE uses interview and observation to assess ability to self‐administer medications in 3 areas: knowledge of medications, how to take medications, and how to procure medications. Each medication is reviewed during administration. Scores 0 to 13, max total deficiency score is 13 | Baseline: mean ± SD intervention 0.833 ± 0.745 vs control 0.692 ± 0.768 Unpublished follow‐up results: mean ± SD: intervention 0.595 ± 0.725 vs control 0.297 ± 0.777; both groups showed significant decreases in number of medication management problems at 2 months (P < 0.01) |
| Manning 2007 | Subjective measure: self‐reported safety. Since discharge, how many mistakes have you made taking your medications (score 0 to 4)? | Mean ± SD: intervention 0.78 ± 0.4187 (n = 72) vs control 0.79 ± 0.4113 (n = 57) |
| Pereles 1996 | Objective measure: assessed differently for each group: intervention = pharmacist assessment with input from other team members, primarily based on having made 2 or fewer errors at stage 2 of the inpatient self‐medication programme ‐ considered able to self‐medicate at discharge. Control = pharmacist assessment with input from other team members at time of discharge counselling. YES/NO ‐ self‐medicating at discharge (note: there could be reasons other than failing the SMP that might explain why they were not self‐medicating at discharge, such as patient preference) | n (%): intervention 39 (76.5%) vs control 39 (69.6%) |
OTC: over‐the‐counter.
SD: standard deviation.
Subgroup analysis was not possible due to the small number of eligible studies.
Overall,
-
no studies were identified that evaluated the impact of educational or behavioural interventions alone on medication‐taking ability; and
-
the effect of mixed interventions on medication‐taking ability was unable to be determined (low‐quality evidence). The evidence was downgraded due to high or unclear risk of bias across multiple domains (‐1) and for inconsistency (‐1) (variations in interventions, outcome measures, settings, duration, etc.).
Primary outcome ‐ adherence
Forty‐eight studies included a measure of adherence and were analysed based on the type of intervention provided: educational (n = 14 studies), behavioural (n = 7 studies), or mixed (n = 27 studies).
Meta‐analyses were possible for 31 studies: 18 involving dichotomous measures (Analysis 1.1), and 13 involving continuous measures (Analysis 1.2). The 17 studies not included in meta‐analyses are briefly summarised in Table 2 and in the following subsections.
| Study | Measure of adherence | Outcome |
|---|---|---|
| Al‐Rashed 2002 | Objective measure: percentage compliance using home medicines stocks and refill prescriptions between visits 1 and 2 | Intervention: 70% (n = 342 medications) |
| Blalock 2010 | Subjective measure: Brief Medication Questionnaire (5‐item regimen screen that assesses how medication is used) | Not reported |
| Bond 2007 | Subjective measure: Extended Medication Adherence Report Scale (MARS) questionnaire (12 statements about medicine‐taking; score range 12 to 60) | Med (IQR): intervention 59 (57 to 60) vs control 59 (57 to 60) |
| Cargill 1992 | Objective measure: pill count, percentage of pills taken compared to those prescribed | Mean scores: control: 74.5; intervention (group 3): 76.2; intervention (group 2): 74; intervention (group 3): 76.2 |
| Cohen 2011 | Objective measure: medication possession ratios | Not clearly reported |
| Hanlon 1996 | Subjective measure: self‐report proportion of medications for which the patient's response agreed with the directions for use on the action profile | Intervention 77.4% (n = 86 people) vs usual care 76.1% (n = 83 people) |
| Holland 2007 | Subjective measure: Medication Adherence Report Scale (MARS) scores from 5 (very poor adherence) to 25 (perfect adherence) | Mean (median): 23.74 (25), n = 101 vs 23.55 (25), n = 103 |
| Krska 2001 | Subjective measure: pharmaceutical care issues including potential or actual compliance issues, number of baseline issues resolved at 3 months | 51 of 74 issues resolved (n = 168) vs 21 of 69 issues resolved (n = 164) |
| Lim 2004 | Subjective measure: self‐reported; patients asked if they 'forgot to take medication as directed'. Then categorised as least compliant (compliant base, not at 2 months), not compliant (not compliant at base or at 2 months), compliant (compliant at 2 months) | Not clearly reported; unadjusted OR 1.50, 90% CI 0.73 to 3.08 |
| Marek 2013 | Objective measure: machine recorded or nurse pill count, average percentage of correct doses per month | Not reported for control group MD.2: 98.8% (SD 0.32), planner: 97.4% (SD 5.19) |
| Pandey 2017 | Subjective measure: participants used a logbook to record name and timing of medications taken on a daily basis. Absolute medication adherence calculated as percentage of total prescribed doses that were actually taken each month. 12 months adherence calculated as the mean of each of the 12 monthly measurements. Adherence outcome is % of days covered | Intervention: 91% (n = 9), control: 73% (n = 8) |
| Pereles 1996 | Objective measure: patients discharged with 40 days worth of medication, pill count conducted in home at 40 days. Number of medication errors as a proportion of the total doses administered | Not clearly reported After controlling for age and MMSE ‐ I: 0.045, C: 0.086; P < 0.001 |
| Shimp 2012 | Objective measure: medication possession ratios defined as sum of all days of medication supply received during year divided by total numbers of days supply needed ‐ calculated for top 8 drug classes for chronic conditions | Not reported ‐ MPRs were very high for both groups (range 0.84 to 0.96), and no clinically meaningful changes were observed over time for either group. Fewer patients reported missed doses after the intervention |
| Taylor 2003 | Subjective measure: self‐reported number of medication doses missed. Presented as % adherence | Intervention mean 100 vs control mean 88.9 (± SD 6.3) |
| Volume 2001 | Subjective measure: Morisky Adherence, scores 0 to 4; lower scores = better adherence | Mean SD: 0.56 ± 0.75 vs 0.47 ± 0.69; number of participants in each group unclear |
| Willeboordse 2017 | Subjective measure: self‐reported adherence problems | Persistence of adherence problems = OR 0.83 (0.54 to 1.27) (P = 0.38) (unpublished = adherence worsened or persisted: 65 vs 54; adherence improved or remained the same: 143 vs 144) |
| Winland‐Brown 2000 | Objective measure: pill count, average number of missed doses (unclear over what time period). Please note: group 2 vs group 3 was used for comparison of intervention vs usual care; group 1 vs group 2 was used for comparison of intervention vs intervention | Mean: group 1 = 15.1, group 2 = 1.7, control = 19.7 |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
C: control.
CI: confidence interval.
I: intervention.
IQR: interquartile ratio.
MMSE: Mini‐Mental State Examination.
MPR: medication possession ratio.
OR: odds ratio.
SD: standard deviation.
Educational interventions
Educational interventions were identified in 14 studies; seven studies were included in meta‐analyses. Two studies ‐ Haag 2016; Marusic 2013 ‐ involving dichotomous measures of adherence and five studies ‐ George 2016; Grymonpre 2001; Messerli 2016; Muth 2016; Nascimento 2016 ‐ involving continuous measures of adherence were included in the meta‐analyses. Analysis 1.1.1 indicated that educational interventions increase the proportion of patients who are adherent (risk ratio (RR) 1.66, 95% confidence interval (CI) 1.33 to 2.06); however results were strongly influenced by Marusic 2013. Three additional studies reported dichotomous data (Table 2), which could not be included in the meta‐analyses due to incomplete data or reporting of results in a format that could not be meta‐analysed. Krska 2001 increased the number of participants who had pharmaceutical care issues (e.g. non‐adherence) resolved (intervention: 68.9% versus usual care: 30.4% resolved), and two other studies reported no differences between groups (Hanlon 1996; Willeboordse 2017).
Overall, the quality of evidence was rated as very low, meaning we are uncertain of the effects of educational interventions on adherence measured dichotomously. We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment, (by ‐1) for inconsistency (high I² and variations in intervention, providers, settings, duration, and outcome measures), and (by ‐1) for imprecision (only two studies in the meta‐analysis; one with very wide confidence interval and low events).
Analysis 1.2.1 showed that educational interventions may have little or no impact on adherence when assessed via continuous measures (standardised mean difference (SMD) 0.16, 95% CI ‐0.12 to 0.43), but heterogeneity was substantial (I² = 74%). Sensitivity analysis performed after one study with high attrition ‐ George 2016 ‐ was removed did not substantially alter the result (SMD 0.16, 95% CI ‐0.14 to 0.47). Four further studies reported continuous measures of adherence and could not be included in the meta‐analysis (Table 2) due to incomplete data or reporting of results in a format that could not be meta‐analysed (e.g. median, interquartile ratio (IQR)). Two studies reported no differences between groups (Bond 2007; Volume 2001), and two studies had no clear results available (Blalock 2010; Shimp 2012), but one study did report that a high medication possession ratio across both intervention and usual care groups meant that no clinically meaningful differences were observable (Shimp 2012).
Overall, educational interventions may have little or no effect on medication adherence measured by continuous adherence outcomes, with quality of evidence assessed as low. We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains and (by ‐1) for inconsistency (high I² and variations in interventions, providers, settings, duration, and outcome measures).
In total, 3 of the 14 educational interventions had positive effects on adherence, and all three were delivered as one‐off interventions. In Krska 2001 and Nascimento 2016, pharmacists provided individualised medication management education and medication reviews at home; Nascimento 2016 also provided therapeutic education on diabetes care. In Marusic 2013, physicians who were specialists in clinical pharmacology provided pre‐discharge counselling (e.g. medication indications, dosages, administration, importance of compliance, possible adverse drug reactions (ADRs)) to participants 24 hours before discharge. Adherence was measured through different measures and at different time points in each study; pharmacist assessment of pharmaceutical care issues included actual or potential adherence issues at three months (Krska 2001), pill count at 30 days (Marusic 2013), and subjective use of a Portuguese/Spanish variation of the Morisky adherence measure at six months (Nascimento 2016).
Behavioural interventions
Behavioural interventions were identified in seven studies, four of which were suitable for inclusion in meta‐analyses. Four studies used a dichotomous measure of adherence, and the meta‐analysis showed that behavioural interventions increased the proportion of adherent patients (Analysis 1.1.2; RR 1.22, 95% CI 1.07 to 1.38) (Hale 2016; Moral 2015; Morales Suarez‐Vurela 2009; Truelove 2015). Sensitivity analysis after removal of one study with high attrition ‐ Truelove 2015 ‐ did not alter the result substantially (RR 1.22, 95% CI 1.02 to 1.45).
Three studies were unable to be included in the meta‐analyses, one due to non‐reporting of standard deviations ‐ Winland‐Brown 2000 ‐ and one due to reporting of the percentage of days covered rather than the percentage of participants adherent (Pandey 2017). Both reported positive effects on adherence. Winland‐Brown 2000 reported fewer missed doses assessed via pill count among participants in the intervention group (1.7 intervention group versus 19.7 usual care), and Pandey 2017 reported a higher percentage of days adherent (91% absolute adherence intervention versus 73% usual care). The remaining single study that used a continuous measure reported a large effect on adherence (mean difference (MD) 13.60, 95% CI 7.78 to 19.42) (Murray 1993).
Overall, behavioural interventions may increase the proportion of people who are adherent to medications, but we are uncertain of the effects of behavioural interventions on medication adherence measured via continuous adherence outcomes. Quality of evidence was assessed as low for dichotomous outcomes and very low for continuous outcomes. We downgraded the evidence for both outcomes (by ‐1) due to high or unclear risk of bias across multiple domains and (by ‐1) for inconsistency (high I² and variations in interventions, providers, settings, duration, and outcome measures). We also downgraded the evidence for continuous outcomes (by ‐1) for imprecision (low participant numbers).
In total, five of the seven behavioural interventions had an individual positive impact on adherence. Two studies involved DAAs: Murray 1993 involved pharmacist‐led regimen simplification to twice‐daily dosing intervals and provided medications in unit of use packaging (translucent plastic cups with lids containing all medications for that dosing time), and Winland‐Brown 2000 used an automated dispenser with audible reminders. The remaining three studies all used different interventions: Truelove 2015 involved a cardiovascular four‐ingredient poly‐pill, Pandey 2017 sent once‐daily text message adherence reminders to participants, and Moral 2015 involved physician/nurse‐led motivational interviewing and follow‐up. Adherence was measured objectively at six months via pill count in three studies (Moral 2015; Murray 1993; Winland‐Brown 2000), and it was measured subjectively via patient log‐book records at 12 months (Pandey 2017), or by self‐reported use of medication at 18 months (Truelove 2015).
Mixed educational and behavioural interventions
Mixed educational and behavioural interventions were identified in 27 studies; 19 studies were included in meta‐analyses. Twelve studies used a dichotomous measure of adherence, and meta‐analysis of data from these studies shows that mixed interventions may increase the proportion of adherent patients (Analysis 1.1.3; RR 1.22, 95% CI 1.08 to 1.37; I² = 77%) (Bernsten 2001; Cossette 2015; Khdour 2009; Lopez Cabezas 2006; Naunton 2003; Olesen 2014; Rich 1996; Saez de la Fuente 2011; Vinluan 2015; Wood 1992; Wu 2006; Young 2016). A sensitivity analysis was conducted after removal of two cluster‐RCTs that had potential unit of analysis errors (Bernsten 2001; Wood 1992), but this had little impact on the risk ratio (RR 1.25, 95% CI 1.09 to 1.44). A second sensitivity analysis performed after removal of six studies with high attrition further strengthened the above finding further (RR 1.33, 95% CI 1.20 to 1.49) (Bernsten 2001; Cossette 2015; Lopez Cabezas 2006; Olesen 2014; Vinluan 2015; Wood 1992).
Seven studies used a continuous measure of adherence, and meta‐analysis of data from these studies shows no significant impact on adherence (Analysis 1.2.3; SMD 0.47, 95% CI ‐0.08 to 1.02; I² = 95%) (Begley 1997; Chrischilles 2014; Lee 2006; Lipton 1994; Nazareth 2001; Shively 2013; Williams 2012). A sensitivity analysis after removal of three studies with high attrition did not substantially alter the findings (SMD 0.70, 95% CI ‐0.25 to 1.65) (Lipton 1994; Nazareth 2001; Williams 2012).
Of the eight studies not included in meta‐analyses, two reported a positive impact on adherence with the intervention (Al‐Rashed 2002; Pereles 1996), four reported no differences between intervention and usual care groups (Cargill 1992; Holland 2007; Lim 2004; Taylor 2003), and two did not have clearly reported results (Cohen 2011; Marek 2013). Of the two studies showing positive impact, Al‐Rashed 2002 reported a significantly higher number of medications taken correctly in the intervention group (70% versus 15.8%) and Pereles 1996 reported a significantly lower number of medication errors in the intervention group, as a proportion of total doses administered (P < 0.001).
Overall, mixed interventions may increase the proportion of people who are adherent to medications but may have little or no impact on medication adherence measured via continuous adherence outcomes. Quality of evidence was assessed as low for both outcomes ‐ downgraded (by ‐1) due to high or unclear risk of bias across multiple domains, and (by ‐1) for inconsistency (high I² and variations in interventions, outcome measures, settings, duration, etc.).
In total, 11 of the 27 mixed interventions had a positive impact on adherence. All 11 studies were conducted at the hospital‐community interface (e.g. in hospital, at discharge, post discharge, at outpatient clinics) and involved elements of education/counselling. Five studies involved pharmacist medication review (Khdour 2009; Lee 2006; Lipton 1994; Naunton 2003; Rich 1996), and three involved regimen simplification (Begley 1997; Lipton 1994; Rich 1996). Interventions varied in duration from one‐off (Al‐Rashed 2002; Naunton 2003; Saez de la Fuente 2011), to within three months (Lipton 1994; Pereles 1996; Young 2016), to 6 to 12 months (Begley 1997; Khdour 2009; Lee 2006), to two years (Wu 2006). Duration of the intervention in one study was unclear (Rich 1996). Other behavioural interventions in these studies included blister‐packed DAAs (Lee 2006, and as needed in Naunton 2003), motivational interviewing (Khdour 2009), and use of a medication reminder card (Al‐Rashed 2002).
We identified sufficient mixed intervention studies to conduct three of the planned subgroup analyses.
-
Duration of intervention: short (≤ 3 months) versus long (> 3 months). When considered separately by subgroups based on intervention duration, there was no difference in adherence between those receiving interventions of a short duration and those receiving a long duration intervention when measured either as a dichotomous outcome (Analysis 1.3; RR 1.4 versus 1.11; test for subgroup differences P = 0.06; I² = 70.7%) or as a continuous outcome (Analysis 1.4; SMD 0.18 versus 0.70; test for subgroup differences P = 0.39; I² = 0%). Heterogeneity remained high, and it is not possible to state whether intervention duration is a major contributing factor to heterogeneity in effects on adherence.
-
Type of outcome measure: objective versus subjective. When considered separately by subgroups based on type of outcome measure, there was no difference in adherence between those studies using an objective measure of adherence and those using a subjective measure of adherence when measured either as a dichotomous outcome (Analysis 1.5; RR 1.13 versus 1.26; test for subgroup differences P = 0.38; I² = 0%) or as a continuous outcome (Analysis 1.6; SMD 0.82 versus 0.21; test for subgroup differences P = 0.39; I² = 0%). Heterogeneity remained high, and it is not possible to state whether type of outcome measure is a major contributing factor to heterogeneity in effects on adherence.
-
Health professional delivering the intervention: when considered separately by subgroups based on health professional delivering the intervention, there was no difference in adherence between those interventions delivered by pharmacists, nurses, or two or more health professionals when measured either as a dichotomous outcome (Analysis 1.7; RR 1.21 versus 1.19 versus 1.38; test for subgroup differences P = 0.83; I² = 0%) or as a continuous outcome (Analysis 1.8; SMD 1.38 versus ‐0.13 versus 0.42; test for subgroup differences P = 0.08; I² = 61.4%). Heterogeneity remained high, and it is not possible to state whether the type of health professional delivering the intervention is a major contributing factor to heterogeneity in effects on adherence.
Secondary outcome ‐ medication knowledge
Thirteen studies included a measure of medication knowledge (Table 3). Meta‐analysis was not possible due to large variations in outcome measures and reporting.
| Study | Measure of medication knowledge | Outcome |
|---|---|---|
| Al‐Rashed 2002 | Pharmacist‐delivered questionnaire; percentage scores for correct answers (drug use, dose, dosage interval) | Drug use: 97.4% vs 69.5%; dosage interval: 97.4% vs 86.0%; dose: 98.5% vs 91.5% |
| Begley 1997 | Patients asked about name, purpose, dose, dosage frequency, and side effects. Reported as percentage of correct answers. Accuracy compared to hospital discharge or GP instructions | Group A 70%, Group B 68%, Group C 66% (usual care) |
| Bernsten 2001 | Interview‐based questionnaire calculating percentage correct (looking at 4 areas: indication, number of dosage units taken per dose, number of doses per day, and awareness of potential adverse effects). Higher scores = better knowledge | Mean ± SD change at 18 months: +3.19 ± 15.18 (n = 704) vs +3.16 ± 16.19 (n = 636) |
| Bond 2007 | Patients were asked whether they "knew more about their medicines compared with a year ago" on 5‐point Likert scale. Those who said agree/strongly agree | Trial report: 73% vs 65% |
| Grymonpre 2001 | Knows purpose of prescribed drugs (yes/no), expressed as number and percentage of drugs correct | 304/327 (93%) vs 335/373 (90%) |
| Hanlon 1996 | Self‐report knowledge of 'how they took each analysed medication and what the medication was for'; percentage of correct responses | 89.4% (n = 86) vs 90.6% (n = 83) |
| Khdour 2009 | COPD knowledge questionnaire (validated) ‐ effectiveness of education in helping persons with COPD. 16 T/F questions, correct response = 1, range 0 to 16, higher score = better knowledge | Median (IQR): 75.0 (32.0) vs 59.3 (33.0) |
| Lim 2004 | Composite knowledge of dose (D), frequency (F), and indication (I), percentage correct | Not reported |
| Manning 2007 | Assessment of knowledge of indication, dosage frequency, and special comments or cautions. 0 (for no correct responses) to 3 (all correct responses) | Mean ± SD: 1.96 ± 0.7561 vs 1.66 ± 0.6851 |
| Messerli 2016 | Knowledge of medicines and daily use ‐ phone questionnaire. 58 questions ‐ included assessing knowledge | Not reported |
| Nazareth 2001 | Prescription medicine interview ‐ patient's knowledge of prescribed drugs. Validated self‐report semi‐structured interview (knowledge score is out of 1, with 1 being 'total/highest' knowledge). Mean (SD) out of 1 | Mean ± SD: 0.69 ± 0.35 (n = 65) vs 0.68 ± 0.32 (n = 68) |
| Pereles 1996 | "Short medication knowledge questionnaire" = Patients asked to name and describe appearance and purpose of their medication, to describe their regimen and any potential side effects or drug interactions. Percentage of correct responses in each knowledge category | Discharge: name: 69% vs 55%; appearance: 77% vs 66%; times: 80% vs 69%; purpose: 77% vs 72%; side effects: 6% vs 4% |
| Taylor 2003 | Self‐reports used to assess medication knowledge. Score determined by dividing the number of medications for which a patient reported the correct name, purpose, dose, and frequency by the total number of medications and multiplying by 100 | Mean ± SD: 92.6 ± 3.4 vs 42.9 ± 12.8 |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
COPD: chronic obstructive pulmonary disease.
GP: general practitioner.
IQR: interquartile ratio.
SD: standard deviation.
T/F: true/false.
Educational interventions
Educational interventions were identified in four studies. In Bond 2007, pharmacist‐led medication management review (including assessment of medication appropriateness, adherence, lifestyle, and social support) conducted in the community pharmacy resulted in more patients agreeing that they knew more about their medications compared to what they knew one year ago (intervention 73% versus usual care 65%). Two studies showed no significant impact on medication knowledge, potentially due to the high levels of knowledge reported in both intervention and usual care groups (knowledge of medications was around 90% in all groups in both studies) (Grymonpre 2001; Hanlon 1996). In Grymonpre 2001, home medication histories were obtained by trained staff and were reviewed by pharmacists; pharmacists then sent a letter summarising the information and providing recommendations to the patient's general practitioner. In Hanlon 1996, the intervention involved pharmacist education provided to the participant and medication review at a general medicine clinic before/after physician appointments. One study involving pharmacist medication review and counselling in the pharmacy did not report results (Messerli 2016).
Overall, educational interventions may have little or no impact on medication knowledge (4 studies; low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains including sequence generation, allocation, attrition, and outcome reporting, and (by ‐1) for inconsistency (variations in settings, providers, duration, and outcome measures).
Behavioural interventions
We identified no behavioural interventions that included a measure of medication knowledge.
Mixed educational and behavioural interventions
Mixed educational and behavioural interventions were identified in nine studies. Five studies reported a positive impact on medication knowledge; four reported small to moderate effects (Al‐Rashed 2002; Khdour 2009; Manning 2007; Pereles 1996), and one reported a large effect (mean ± standard deviation (SD) knowledge score/100: intervention 92.6 ± 3.4 versus usual care 42.9 ± 12.8) (Taylor 2003). Two studies used one‐off pre‐discharge education: Al‐Rashed 2002 involved a 30‐minute pharmacist counselling session (focusing on indications, side effects, dose, dosage times, importance of adherence, provision of a medication card etc.), and Manning 2007 involved nurse education via a three‐dimensional (3D) medication discharge education tool, whereby participants could affix a tablet or capsule of each medication onto the tool to assist with tablet identification. Three studies involved follow‐up/monitoring: Pereles 1996 involved a three‐stage self‐administration of medications programme in hospital, Khdour 2009 involved medication review and motivational interviewing conducted four times over nine months in an outpatient clinic or via phone, and Taylor 2003 involved pharmacist education and medication review at scheduled medical clinic visits over 12 months. Three studies reported no impact on medication knowledge ‐ one involving medication review and regimen simplification conducted at home post discharge for 12 months (Begley 1997), one involving medication review and regimen simplification conducted continuously for 18 months in the community pharmacy (Bernsten 2001), and one involving hospital pharmacist medication review with community pharmacist home visit follow‐up 7 to 14 days post discharge (Nazareth 2001). One study did not report any follow‐up results on medication knowledge (Lim 2004).
Overall, mixed interventions may improve medication knowledge but to a variable degree (9 studies; low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains including sequence generation, allocation, and outcome reporting and (by ‐1) for inconsistency (variations in settings, providers, duration, and outcome measures).
Secondary outcome ‐ consumer satisfaction
Thirteen studies included a measure of consumer satisfaction, but only 10 studies measured the outcome in both intervention and control groups (Bernsten 2001;Bond 2007;George 2016;Hanlon 1996;Lopez Cabezas 2006;Manning 2007;Nazareth 2001;Taylor 2003;Volume 2001;Willeboordse 2017). Three studies measured satisfaction in the intervention group only (Holland 2007;Lingler 2016;Naunton 2003). Meta‐analysis was not possible due to large variation in outcome measures (Table 4).
| Study | Measure of satisfaction | Outcome (intervention vs usual care) |
|---|---|---|
| Bernsten 2001 | Self‐reported rating of services provided, satisfaction with services, and general opinion of pharmaceutical care. Questionnaire administered by pharmacist. Results presented as percentage who agree/mainly agree | Rating of services as excellent: 73.8% vs 64.6%; satisfaction with services: 93.9% vs > 90%; general opinion: 77% (intervention group only) |
| Bond 2007 | Overall score on 15 positive and negative statements of most recent pharmacy visit (total score 15 to 75, higher scores better) | Median (IQR): 46 (40 to 55) vs 43 (38 to 49) |
| George 2016 | User satisfaction regarding use of the computer programme questionnaire (USUCPQ): an 8‐item measure based on 7‐point Likert score (max score 56, higher scores better) | Mean ± SD total satisfaction: 45.33 ± 7.81 vs 44.68 ± 6.75 |
| Hanlon 1996 | Health Care Attitude Questionnaire: 3 questions on pharmacy‐related healthcare satisfaction (directions received, explanation of SES, numbers/types of drugs) based on 5‐point Likert scale (lower scores better) | Mean ± SD total score: 5.2 ± 1.5 vs 5.4 ± 1.7 |
| Holland 2007 | Satisfaction questionnaire; usefulness of community pharmacist visits | 75 (64%) considered the visits to have been extremely or very useful |
| Lingler 2016 | Acceptability of the intervention using a set of Likert scale questions and eliciting open‐ended comments | 88% of caregivers reported intervention topics useful and relevant; 92% reported that the intervention was helpful for managing the patient's treatment plan |
| Lopez Cabezas 2006 | Catalan Health Department satisfaction survey, asking participants about the care and information received and asking them to provide a global scoring (0 to 10) | Mean ± SD 8.9 ± 1.3 vs 8.8 ± 1.5 |
| Manning 2007 | Level of satisfaction using 5‐point Likert scale (5 = highest): "How satisfied were you with the form you received from the nurse when she/he was talking to you about your medications?" | Mean ± SD 4.24 ± 0.6986 vs 4.26 ± 0.8768 |
| Naunton 2003 | Survey of intervention group only | 94% very satisfied; 84% stated information they were given 'helped a great deal' |
| Nazareth 2001 | Validated patient satisfaction questionnaire, each item scored 1 to 4, mean score per item calculated (higher = better) | Mean ± SD 3.4 ± 0.6 (n = 62) vs 3.2 ± 0.6 (n = 61) |
| Taylor 2003 | Mean ± SD number of patients with pharmacy‐related satisfaction (details unclear) | Mean ± SD 81.9 ± 4.8 (n = 33) vs 89.0 ± 6.2 (n = 36) |
| Volume 2001 | Satisfaction with pharmacy services using 34‐item instrument and 7‐point Likert scale (lower scores = better). General satisfaction extracted | Mean ± SD 1.53 ± 0.77 vs 1.62 ± 0.88 |
| Willeboordse 2017 | Medication satisfaction questionnaire assessed on a 7‐point Likert scale | B (95% CI): 0.11 (‐0.08 to 0.30) (P = 0.25) |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
CI: confidence interval.
IQR: interquartile ratio.
SD: standard deviation.
Educational interventions
Educational interventions were identified in five studies, four of which reported no impact on satisfaction (George 2016;Hanlon 1996;Volume 2001;Willeboordse 2017). Bond 2007, which involved community pharmacists providing consultations on medications, medication adherence, lifestyle, and social support, reported that participants in the intervention group had greater satisfaction compared to those in the usual care group; however, the effect size was small.
In summary, educational interventions may have little or no impact on consumer satisfaction (5 studies; low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains including sequence generation, allocation, incomplete outcome reporting, and other sources of bias and (by ‐1) for inconsistency (variations in settings, providers, duration, and outcome measures).
Behavioural interventions
We identified no behavioural interventions that included a measure of satisfaction.
Mixed educational and behavioural interventions
Mixed educational and behavioural interventions were identified in eight studies. Three studies ‐ Holland 2007; Lingler 2016; Naunton 2003 ‐ measured participant satisfaction only in the intervention group, with satisfaction levels ranging from 64% in Holland 2007 to 94% in Naunton 2003. Five studies measured participant satisfaction in both intervention and usual care groups; between‐group differences were non‐significant in four studies (Bernsten 2001; Lopez Cabezas 2006; Manning 2007; Nazareth 2001), and one study reported slightly higher pharmacy‐related satisfaction in the group receiving usual care (mean 89.0, SD 6.2) compared with the intervention (mean 81.9, SD 4.8); however, the satisfaction measure used was poorly described (Taylor 2003).
Overall, we are uncertain of the effects of mixed interventions on consumer satisfaction (8 studies; very low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains including sequence generation, allocation, and incomplete outcome reporting, (by ‐1) for inconsistency (variations in interventions, settings, providers, duration, and outcome measures), and (by ‐1) due to imprecision (three of eight studies reported satisfaction only in the intervention group).
Secondary outcome ‐ HRQoL
Fourteen studies included a measure of HRQoL (Table 5). Meta‐analysis was not possible due to differences in scales used and differences in reporting of results.
| Study | Measure | Time point | Outcome |
|---|---|---|---|
| Bernsten 2001 | SF‐36 | 18 months | Change: GH: +0.28 vs ‐0.66, MH: ‐0.80 vs ‐1.34, PF: ‐0.95 vs‐0.68 |
| Bond 2007 | SF‐36 | 12 months | Med (IQR): GH: 52 (35 to 65) vs 50 (35 to 70), MH: 80 (64 to 88) vs 80 (64 to 88), PF: 60 (35 to 80) vs 65 (35 to 85) |
| EQ‐5D | 12 months | Med (IQR): 0.73 (0.7 to 0.9) vs 0.73 (0.7 to 0.9) | |
| Cohen 2011 | VR‐36 | 6 months | Change: Med (IQR): MH: 0.48 (‐3.37 to 4.32), C: 0.78 (‐2.67 to 4.23), PF: 1.65 (‐5.21 to 1.31), C: ‐1.95 (‐5.21 to 1.31) |
| Hale 2016 | MLHFQ | 90 days | Mean ± SD: 62.2 ± 20.6 vs 28.2 ± 22.3 |
| Hanlon 1996 | SF‐36 | 12 months | Mean ± SD: GH: 37.4 ± 1.6 vs 35.2 ± 1.7, MH: 61.1 ± 1.8 vs 60.4 ± 1.8, PF: 44.1 ± 2.0 vs 42.2 ± 2.0 |
| Holland 2007 | EQ‐5D, VAS | 6 months | Mean ± SD: EQ‐5D: 0.58 ± 0.29 vs 0.52 ± 0.34, VAS: 58.2 ± 19.6 vs 58.6 ± 19.8 |
| MLHFQ | Mean ± SD: 47.7 ± 26.3 vs 44.5 ± 27.9 | ||
| Khdour 2009 | SGRQ | 12 months | Mean (confidence interval): 61.8 (57.9 to 65.6) vs 65.3 (61.0 to 69.6) |
| Krska 2001 | SF‐36 | 3 months | No significant differences ‐ values not reported |
| Lopez Cabezas 2006 | EQ‐5D (Spanish | 12 months | Mean ± SD: 64 ± 15.4 vs 60.6 ± 17.8, subgroup > 70 years: 63.8 ± 15.3 vs 58.4 ± 15.9 |
| Marek 2013 | SF‐36 | 12 months | Comparison 1 ‐ Mean (confidence interval): planner (intervention) vs control (usual care) = PCS: 1.390 (0.816 to 1.963), MCS: 1.686 (0.949 to 2.423) |
| Comparison 2 ‐ Mean (confidence interval): MD.2 (intervention 1) vs planner (intervention 2) = PCS: 0.095 (‐0.450 to 0.640), MCS: 0.241 (‐0.459 to 0.940) | |||
| Muth 2016 | EQ‐5D | 12 weeks | Mean ± SD: change: ‐0.6 ± 19.61 vs ‐1.0 ± 13.66 |
| Taylor 2003 | SF‐36 | 12 months | Mean SD: GH: 57.0 ± 19.6 vs 50.1 ± 15.9, MH: 73.1 ± 21.2 vs 72.3 ± 17.1, PF: 68.6 ± 24.0 vs 56.1 ± 27.5 |
| Volume 2001 | SF‐36 | 12 to 13 months | Mean ± SD: MCS: 56.14 ± 8.30 vs 54.55 ± 8.65, PCS: 36.87 ± 11.62 vs 38.39 ± 11.44 |
| Willeboordse 2017 | SF‐12 | 6 months | Regression coefficients adjusted for baseline: PCS: ‐0.06 (‐3.19 to 3.06), MCS: 0.16 (‐2.89 to 3.22) |
| EQ‐5D‐3L | Regression coefficients adjusted for baseline: utility: 0.02 (‐0.02 to 0.05), VAS: 2.30 (‐0.16 to 4.76) |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
C: control.
EQ‐5D: EuroQoL Group Quality of Life Questionnaire based on 5 dimensions.
GH: general health.
I: intervention.
MH: mental health.
MCS: mental components summary.
MLHFQ: Minnesota Living With Heart Failure Questionnaire (21 items, coded 0 to 5; higher scores indicate adverse impact on life).
PCS: physical components summary.
PF: physical function.
SD: standard deviation.
SF‐36: Short Form‐36 Health Survey.
SGRQ: St. George's Respiratory Questionnaire (76 items, total score 100; higher = better).
VAS: visual analogue scale.
Educational interventions
Educational interventions were identified in six studies (Bond 2007;Hanlon 1996;Krska 2001;Muth 2016;Volume 2001;Willeboordse 2017); none of these had a significant impact on HRQoL.
Overall, educational interventions probably have little or no effect on HRQoL (6 studies; moderate‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains.
Behavioural interventions
Behavioural interventions were identified in only one study (Hale 2016). The intervention involved a remotely monitored electronic medication dose administration aid with alerts and follow‐up calls if medications were missed. HRQoL was measured via the MLHFQ, with results showing that the intervention group had worse HRQoL (higher MLHFQ scores) both at baseline and at 90‐day follow‐up compared to the usual care group (baseline: mean ± SD: 43.7 ± 25.9 versus 26.2 ± 23.1; 90 days: mean ± SD: 62.2 ± 20.6 versus 28.2 ± 22.3).
Overall, we are uncertain of the effects of behavioural interventions on HRQoL (1 study; very low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains, (by ‐1) for indirectness (MLHFQ specific to heart failure populations), and (by ‐1) for imprecision (low participant numbers).
Mixed educational and behavioural interventions
Mixed educational and behavioural interventions were identified in seven studies, with six showing no significant impact on HRQoL (Bernsten 2001;Cohen 2011;Holland 2007;Khdour 2009;Lopez Cabezas 2006;Taylor 2003). Marek 2013 showed that nurse education, follow‐up, and weekly nurse‐filled DAAs resulted in improved physical and mental summary scores on Short Form (SF)‐36 at 12 months compared to usual care (mean change: physical 1.390, 95% CI 0.816 to 1.963; mental 1.686, 95% CI 0.949 to 2.423; P < 0.0001).
Overall, mixed interventions may have little or no effect on HRQoL (7 studies; low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains and (by ‐1) for inconsistency (variations in interventions, outcome measures, settings, duration, etc.).
Secondary outcome ‐ adverse clinical health outcomes
Twenty‐eight studies included a measure of adverse clinical health outcomes (Table 6).
| Study | Time point | ED/Hospital admissions | Mortality | Adverse drug reactions | GP visits |
|---|---|---|---|---|---|
| Al‐Rashed 2002 | 3 months | Patients re‐admitted to hospital: 8/43 v 28/40 | Total unplanned visits: 43 (n = 43) vs 59 (n = 40) | ||
| Bernsten 2001 | 18 months | Self‐reported: 35.6% vs 40.4%; n values unclear | |||
| Chrischilles 2014 | 3 months | Self‐reported: 100/802 (12.9%) vs 33/273 (12.2%) | |||
| Cossette 2015 | 30 days | ED visits: 18% (n = 108) vs 20% (n = 95) | |||
| Haag 2016 | 30 days | ED or hospital re‐admission: 2/11 (18%) vs 1/11 (9%) | |||
| Hale 2016 | 90 days | No. participants: ED: 3/11 (27%) vs 6/14 (43%); hospitalisation: 1/11 (9%) vs 7/14 (50%); total no.: ED 4 vs 7, hospital 2 vs 8 | |||
| Hanlon 1996 | 12 months | Self‐reported: 30.2% (n = 86) vs 40% (n = 83) (P = 0.19) | |||
| Holland 2007 | 6 months | Total number of ED admissions (not number of participants admitted): 134 (n = 148) vs 112 (n = 143) | 30/148 vs 24/143 | ||
| Khdour 2009 | 12 months | n = 71 and 72; ED: 40 vs 80, hospital: 26 vs 64; total hospital days: 164 vs 466 | Unscheduled: 28 (n = 71) vs 47 (n = 72) | ||
| Lim 2004 | 2 months | Self‐reported and assessed by physician, total ADRs at 2 months: 13 vs 6; residual ADRs from baseline: 4/13 vs 4/8 | |||
| Lipton 1994 | 6 months | Total days in hospital: mean ± SD 2.29 ± 5.96, n = 350 vs 2.02 ± 5.83, n = 356 | |||
| Lopez‐Cabezas 2016 | 12 months | Patients with re‐admission: 23/70 (32.9%) vs 31/64 (48.4%) | > 70 years subgroup: 7/53 (13.2%) vs 13/50 (26.0%) | ||
| Marusic 2013 | 30 days | Re‐admission or ED: 20/80 (25%) vs 27/80 (33.8%) | ? Self‐reported: 24/80 (30%) vs 30/80 (37.5%) (P = 0.315) | ||
| Messerli 2016 | 28 weeks | Self‐reported unplanned GP visit or hospitalisation: total during study: 110 vs 99, n unclear? ‐ 181 vs 191 | * | ||
| Murray 1993 | 6 months | Self‐reported side effects, ill effects, or other problems with medication: C1: 2/12, C2: 2/10, Int: 1/9 | |||
| Muth 2016 | 12 weeks | Days in hospital: mean ± SD (T1 + T2) ‐ T0 = ‐0.4 ± 0.73 vs ‐0.2 ± 0.69, n unclear | |||
| Naunton 2003 | 90 days | 1 or more unplanned re‐admissions: 16/57 (28%) vs 29/64 (45%) | 3/57 (5%) vs 5/64 (8%) | ||
| Nazareth 2001 | 6 months | Re‐admissions: 38/136 (27.9%) vs 43/151 (28.4%) Outpatient department: 39/137 vs 40/151 | 22/137(16.1%) vs 19/151 (12.6%) | 76/107 vs 82/116 | |
| Olesen 2014 | 24 months | Unplanned admissions: 77/253 (30%) vs 73/264 (28%) | 19/253 (7.5%) vs 14/264 (5%) | ||
| Rich 1996 | 90 days | Re‐admissions: 18/80 (22.5%) vs 22/76 (28.9%) | |||
| Saez de la Fuente 2011 | 50 days | Total re‐admissions: 5 (n = 26) vs 7 (n = 24); ED: 7 (n = 26) vs 9 (n = 24) (note percentages listed in paper do not match n values) | 2 (? n = 26) vs 1 (? n = 24) | ||
| Shively 2013 | 6 months | Hospital: mean (SD): 0.21 (0.409) vs 0.32 (0.475) n = 39 vs 37 | |||
| Taylor 2003 | 12 months | Hospital: 2/33 vs 11/36; ED: 4/33 and 6/36 | |||
| Vinluan 2015 | 90 days | Hospital admissions: 2/7 vs 2/7 | 0/7 vs 2/7 | ||
| Willeboordse 2017 | 6 months | DRPs: baseline: 4.4 ± 1.9 vs 3.7 ± 1.7 % solved: 20.2 (12.2 to 28.1) | |||
| Winland‐Brown 2000 | 6 months | Hospitalisations G1: 4/16, G2: 3/24, C: 12/21 | Physician visits: G1: 1.5/month, G2: 1/month, C: 1/month | ||
| Wu 2006 | 2 years | Med (IQR): n = 219 vs 223 | 25/219 vs 38/223 | ||
| Young 2016 | 180 days | Hospital: 18/51 (35.3%) vs 20/49 (40.8%); ED visits: 12/51 (23.5%) vs 11/49 (22.4%) |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
ADR: adverse drug reaction.
C: control.
C1: control group 1.
C2: control group 2.
DRP: drug‐related problem.
ED: emergency department.
G1: group 1.
G2: group 2.
GP: general practitioner.
Int: intervention.
SD: standard deviation.
Emergency department (ED)/Hospital admissions
ED/Hospital admissions were measured in 23 studies, and 16 studies were included in the meta‐analysis (Analysis 1.9). Four studies reported hospital and ED admissions separately, but only hospital admissions were chosen to be included in the meta‐analysis to avoid duplication of participants in the analysis (Hale 2016;Khdour 2009;Taylor 2003;Young 2016).
Educational interventions
Educational interventions had no significant impact on admissions (RR 1.02, 95% CI 0.71 to 1.48) (Haag 2016; Marusic 2013; Messerli 2016). One additional study that could not be included in meta‐analysis reported mean number of days in hospital but reported no effect of an educational intervention (Muth 2016).
Overall, educational interventions probably have little or no effect on ED/hospital admissions (RR 1.02, 95% CI 0.71 to 1.48; moderate‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains.
Behavioural interventions
Behavioural interventions reduced the risk of admissions (RR 0.21, 95% CI 0.08 to 0.55) (Hale 2016;Winland‐Brown 2000), although overall we are uncertain about the effects of behavioural interventions on admissions because of the very low quality of evidence for this outcome. We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains and (by ‐2) for imprecision (low participant numbers and low numbers of events).
Mixed educational and behavioural interventions
Meta‐analysis of 11 studies shows that mixed interventions reduced the risk of admissions (RR 0.67, 95% CI 0.50 to 0.90); however the level of heterogeneity was high (I² = 73%) (Al‐Rashed 2002;Cossette 2015;Khdour 2009;Lopez Cabezas 2006;Naunton 2003;Nazareth 2001;Olesen 2014;Rich 1996;Taylor 2003;Vinluan 2015;Young 2016).
A further six studies were not included in the meta‐analysis (Bernsten 2001;Holland 2007;Lipton 1994;Saez de la Fuente 2011;Shively 2013;Wu 2006). Two studies had unclear participant numbers (Bernsten 2001;Saez de la Fuente 2011), one reported only total admissions and not the number of participants admitted (Holland 2007), one reported the mean (SD) number of days in hospital (Lipton 1994), one reported mean (SD) admissions (Shively 2013), and one reported median (IQR) hospital visits (Wu 2006). None of these studies reported a difference between intervention and usual care groups for ED/hospital admissions.
Overall, mixed interventions may reduce the number of ED/hospital admissions (RR 0.67, 95% CI 0.50 to 0.90; low‐quality evidence). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains and (by ‐1) for inconsistency (variations in interventions, outcome measures, settings, duration, and effect estimates across studies).
In total, four out of 23 individual studies had results favouring the intervention ‐ one behavioural ‐ Winland‐Brown 2000 ‐ and three mixed interventions (Al‐Rashed 2002;Khdour 2009;Taylor 2003). Two studies involved pharmacist‐led education and medication review with ‐ Khdour 2009 ‐ or without ‐ Taylor 2003 ‐ motivational interviewing conducted in an outpatient clinic ‐ Khdour 2009 ‐ or a medical clinic ‐ Taylor 2003. One study involved an automated medication dispenser with audible adherence reminders in the participant's home (Winland‐Brown 2000). Another study involved pharmacist pre‐discharge counselling and provision of a medicine reminder card in hospital (Al‐Rashed 2002).
Mortality
Mortality was reported in eight studies, all involving mixed behavioural and educational interventions led by pharmacists. No studies were identified that evaluated the impact of educational or behavioural interventions alone on mortality. Meta‐analysis involving seven studies shows that we are uncertain of the impact of mixed interventions on mortality (Analysis 1.10; RR 0.93, 95% CI 0.67 to 1.30; very low‐quality evidence) (Holland 2007;Lopez Cabezas 2006;Naunton 2003;Nazareth 2001;Olesen 2014;Vinluan 2015;Wu 2006). No individual study had a significant impact on mortality; three appeared to favour the intervention, and four appeared to favour usual care. We excluded one further study the meta‐analysis due to unclear participant numbers (Saez de la Fuente 2011). We downgraded the evidence (by ‐1) due to high or unclear risk of bias across multiple domains, (by ‐1) for inconsistency (variations in interventions, outcome measures, settings, duration, etc.), and (by ‐1) for imprecision (limits of 95% confidence intervals include both potential benefit and potential harm).
Adverse drug reactions (ADRs)
Adverse drug reactions (ADRs) were reported in six studies.
Educational interventions
Three studies providing educational interventions included a measure of ADRs; two reported no differences between groups (Hanlon 1996; Marusic 2013), and one found that the percentage of solved medication‐related problems (including ADRs) was significantly higher in the intervention group (regression coefficient 22.6, 95% CI 14.1 to 31.1; P < 0.001) (Willeboordse 2017).
Behavioural interventions
One study providing behavioural interventions was identified but reported no differences between groups in self‐reported ADRs (Murray 1993).
Mixed educational and behavioural interventions
Two studies providing mixed educational and behavioural interventions were identified. One study reported no differences between intervention and usual care groups (Chrischilles 2014), and the other found that total ADRs were higher at two months following an intervention involving education, medication review, and regimen simplification (total 13 versus 6) but that residual ADRs from baseline were lower (4/13 versus 4/8) (Lim 2004).
Physician visits
Physician visits were reported in four studies.
Mixed educational and behavioural interventions
Two studies, both involving mixed educational and behavioural interventions, reported reductions in the number of unplanned physician visits with the intervention over usual care (43 versus 59 total visits and 39% versus 65% of participants, respectively) (Al‐Rashed 2002; Khdour 2009).
Two studies ‐ one behavioural ‐ Winland‐Brown 2000 ‐ and one mixed ‐ Nazareth 2001 ‐ reported no between‐group differences for the number of physician visits.
Secondary outcome ‐ condition‐specific outcomes
Seven studies included a condition‐specific outcome measure (Table 7). Meta‐analysis was not possible due to variations in outcome measures.
| Study | Measure | Outcome |
|---|---|---|
| Blalock 2010 | Falls (self‐reported) in 12 months (ITT analysis) | ≥ 1 fall: 53/93 vs 52/93 |
| Bond 2007 | Total score (/8) for reaching targets at 12 months | 4.6 ± 1.2 vs 4.6 ± 1.1 |
| Cohen 2011 | Percentage achieving targets at 6 months (SBP < 130, LDL < 100, HbA1c < 7%) | 16% (n = 50) vs 4.1% (n = 49) |
| Lee 2006 | Systolic and diastolic blood pressure (mmHg) and LDL‐cholesterol (mg/dL) at 6 months post phase 1 | SBP: 124.4 ± 14.0 vs 133.3 ± 21.5 LDL: 87.5 ± 24.2 vs 88.4 ± 21.0 |
| Nascimento 2016 | Fasting blood glucose and HbA1c at 6 months | FBG: 117.3 ± 26.8 vs 142.2 ± 32.9 HbA1C: 7.7 ± 0.8 vs 7.99 ± 0.67 |
| Taylor 2003 | Number of people reaching goal level at 12 months | BP: 22 (91.7%) vs 8 (27.6%) INR: 4 (100%) vs 1 (16.7%) (Note: calculated mean across all 4 measures: 92% vs 19%) |
| Williams 2012 | Blood pressure, HbA1c, eGFR, and creatinine levels at 12 months (9 months post intervention) | SBP: mean (CI) ‐6.9 (‐13.8 to 0.02) vs ‐3.0 (‐8.4 to 2.4) HbA1c: med (IQR): 7 (7 to 9) vs 8 (7 to 9) eGFR: med (IQR): 48 (38 to 76) vs 46 (32 to 72) Creatinine: med (IQR): 117 (82 to 144) vs 108 (89 to 171) |
Outcome results presented as intervention group vs usual care group unless otherwise stated and presented as mean ± SD unless otherwise stated.
BMI: body mass index.
BP: blood pressure.
DBP: diastolic blood pressure.
eGFR: estimated glomerular filtration rate.
FBG: fasting blood glucose.
HbA1c: glycosylated haemoglobin.
INR: international normalised ratio.
IQR: interquartile ratio.
ITT: intention‐to‐treat.
LDL: low‐density lipoprotein.
SBP: systolic blood pressure.
Educational interventions
Three studies involved educational interventions: Blalock 2010 reported no difference between groups in terms of the number of participants experiencing one or more falls; Bond 2007 found no between‐group differences in the number of participants reaching health targets (total score for eight targets; e.g. physical activity, diet, weight); and Nascimento 2016 reported greater reductions in fasting blood glucose and glycosylated haemoglobin (HbA1c) levels in intervention participants compared with those receiving usual care.
Behavioural interventions
No studies that involved only behavioural interventions were identified.
Mixed educational and behavioural interventions
Four studies involved mixed interventions. Two studies involving multidisciplinary ‐ Cohen 2011 ‐ or pharmacist ‐ Taylor 2003 ‐ education in a medical clinic with follow‐up reported higher numbers of people reaching goal levels for blood pressure, HbA1c, and low‐density lipoprotein (LDL) cholesterol with the intervention over usual care (16% versus 4% and 92% versus 19%, respectively). Two additional studies measured multiple health targets including blood pressure, reporting that although results favoured the intervention, they were not significantly different from those attained with usual care (Lee 2006;Williams 2012).
Secondary outcome ‐ cost‐effectiveness
Four studies ‐ one providing educational intervention ‐ Bond 2007 ‐ and three providing mixed interventions ‐ Bernsten 2001;Lipton 1994;Lopez Cabezas 2006 ‐ included a measure of cost‐effectiveness (see Table 8).
| Study | Measure of costs | Outcome |
|---|---|---|
| Bernsten 2001 | Direct costs of the study, including additional time spent by pharmacists, costs associated with contacts with other health professionals, costs of hospitalisation and drugs | Average cost per patient (saving): Denmark: 1298.13 vs 1419.88 (+121.75) Germany: 2992.25 vs 3167.25 (+175.00) Northern Ireland: 735.22 vs 750.01 (+14.79) Sweden: 1266.76 vs 1250.34 (‐16.42) |
| Bond 2007 | Total NHS‐related study costs, including costs of intervention and other treatment (e.g. medicines, hospital, other health consultations) | Median cost (IQR): 970.5 (667.0 to 1489.0) vs 835.2 (534.4 to 1396.3) Median (IQR) cost of intervention alone (pharmacist time and training): 90 (60 to 118) |
| Lipton 1994 | Medicare Part B charges, total hospital inpatient charges | Total charges: mean ± SD 2769 ± 4789 vs 2598 ± 3722 Inpatient charges: mean ± SD 5472 ± 10904 vs 5263 ± 11478 |
| Lopez Cabezas 2006 | Hospitalisation costs, adding in intervention direct costs, delivered materials and time spent by the pharmacist | Average cost per patient: 997 vs 1575 |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
IQR: interquartile ratio.
NHS: National Health Service.
SD: standard deviation.
Educational interventions
In one study (Bond 2007), the educational intervention involved a one‐off pharmacist medication review; total National Health Service (NHS)‐related study costs (medicines plus NHS visits plus intervention costs) were higher in the intervention group compared with the usual care group (median 970.5 versus 835.2), although the main difference was the cost of the intervention itself (median 90, IQR 60 to 118).
Mixed interventions
Two mixed interventions involving pharmacist medication review with repeated/continuous follow‐up for 18 months, either in the pharmacy ‐ Bernsten 2001 ‐ or in hospital with home follow‐up ‐ Lopez Cabezas 2006 ‐ reported that the intervention resulted in a reduction in mean/median costs per patient, although cost savings were variable. The third mixed intervention ‐ Lipton 1994 ‐ involving one‐off face‐to‐face medication review and education in hospital with telephone follow‐up post discharge for three months shows that intervention patients had higher Medicare Part B charges and total hospital inpatient charges as measured at six months compared to usual care patients.
Secondary outcome ‐ other
Four studies included other outcome measures potentially related to measures of medication‐taking ability, medication adherence, or adverse drug events (see Table 9).
| Study | Measure | Outcome (Intervention vs usual care) |
|---|---|---|
| Chrischilles 2014 | Mean (SD) number of medication management problems from a list of 8 problems, including questions on multiple prescribers, multiple pharmacies, mail order prescriptions, confusion whether medication was taken, taking medication without knowing indication, problems affording medications, feeling that medications are not working, and feeling that medications are not doing what they were intended to do | Mean ± SD 1.4 ± 1.4 vs 1.6 ± 1.5 |
| Lingler 2016 | Medication deficiency checklist: a 15‐item, investigator‐developed instrument that uses caregiver interviews to assess for the presence of errors and problems (e.g. incorrectly chewing pills or capsules, taking at the wrong time, repeating doses, patient refuses/unco‐operative) | Mean ± SD 2.19 ± 1.52 vs 2.36 ± 1.51 |
| Moral 2015 | Average number of medication errors, defined as both patient errors (e.g. omission of dose) and prescriber errors (e.g. dose too high or too low, duplicate therapy) (as reported in Perula de Torres 2014 paper) | Mean 0.429 vs 1.145 |
| Taylor 2013 | Number of participants with at least 1 medication misadventure (defined as medication errors, adverse drug events, and/or adverse drug reactions) | 2.8% (n = 33) vs 3.0% (n = 36) |
Outcome results presented as intervention group vs usual care group unless otherwise stated.
SD: standard deviation.
Behavioural interventions
One study, involving a behavioural intervention comprising motivational interviewing and follow‐up by physicians and nurses, found that the average number of medication errors, defined as both patient errors (e.g. omission of dose) and prescriber errors (e.g. dose too high or too low, duplicate therapy), was significantly lower in the intervention group compared to the usual care group (0.429 vs 1.145; P = 0.047) (Moral 2015).
Mixed interventions
Three studies involving mixed interventions reported no differences between intervention and usual care groups in medication management problems (Chrischilles 2014), medication errors and problems (Lingler 2016), and medication errors and adverse drug events or reactions (Taylor 2003).
COMPARISON 2. Intervention versus intervention
Six studies involved comparison between interventions as part of a three‐arm RCT design (Begley 1997;Cargill 1992;Marek 2013;Murray 1993;Olesen 2014;Winland‐Brown 2000).
Primary outcome ‐ medication‐taking ability
Two studies included measures of medication‐taking ability.
Cargill 1992 compared a mixed intervention versus an educational intervention. In this comparison, both groups received a 20‐minute nurse education session and review of medications (educational intervention); one of the groups also received follow‐up telephone calls (mixed intervention). No difference in medication‐taking behaviour between groups was reported at four to six weeks (score out of 100 read from a graph: mean 84 versus 86).
Begley 1997 compared a mixed intervention involving pharmacist home interview with counselling versus a behavioural intervention of home interview only without counselling (modified usual care). This study reported no difference in dexterity between groups at 12 months (data were not reported).
Primary outcome ‐ adherence
Six studies involved a measure of medication adherence.
Two studies compared two behavioural interventions. In Murray 1993, pill count adherence was slightly higher in the group that received regimen simplification and unit of dose medication packaging than in the group given regimen simplification alone (92.6% versus 82.6%; P = 0.02). In Winland‐Brown 2000, nurse‐filled pillbox DAAs were associated with a higher number of missed pills as measured via pill count than via an automated dispenser with audible adherence reminders (mean 15.1 versus 1.7; P < 0.001); however the time interval is unclear.
Cargill 1992 compared a mixed intervention versus an educational intervention and found no difference in pill count adherence between the group that received nurse teaching with additional telephone follow‐ups versus the group receiving a nurse teaching session without follow‐up (mean 76% versus 74%).
Two studies compared mixed interventions versus behavioural interventions. Begley 1997 found that pharmacist home interview with counselling had slightly greater impact on pill count adherence than the home interview alone (mean ± SD percentage 86 ± 19 versus 75 ± 21). Olesen 2014 measured adherence using different methods in different groups, and thus data are not comparable.
Marek 2013 compared two mixed interventions and found that pill count adherence was similar across both medication dispensing machine and simple medication box groups (98.8% versus 97.4%).
Secondary outcome ‐ medication knowledge
Begley 1997 compared a mixed intervention involving pharmacist home interview with counselling versus a behavioural intervention of home interview only without counselling (modified usual care). This study reported no difference in medication knowledge between groups at 12 months (mean 70% versus 68%), as measured by comparison of patient answers to hospital discharge and general practitioner (GP) instructions regarding medication name, purpose, dose, dosage frequency, and side effects.
Secondary outcome ‐ satisfaction
No studies were identified.
Secondary outcome ‐ HRQoL
Marek 2013 compared two mixed interventions ‐ a medication‐dispensing machine with audio and visual prompts for adherence versus nurse‐filled simple weekly medication boxes. There was no difference in improvement in participant physical or mental summary scores between the two groups as measured via SF‐36 (mean physical 0.095, 95% CI ‐0.450 to 0.640; mean mental 0.241, 95% CI ‐0.459 to 0.940).
Secondary outcome ‐ adverse clinical health outcomes effects
Three studies reported measures of adverse clinical health outcomes.
Two studies compared two behavioural interventions. In Winland‐Brown 2000, hospitalisations (4/16 (25%) versus 3/24 (12.5%) patients) and mean number of physician visits (1.5/month versus 1/month) were higher in the weekly pre‐filled pill box DAA group than in the group using an automated dispenser with audible reminders. In Murray 1993, the main intervention group (regimen simplification and unit of use DAA packages) and the modified usual care group (group C2; regimen simplification without unit of use medication DAA packages) had similar numbers of self‐reported side effects over six months of follow‐up (1/9 versus 2/10).
Olesen 2014 compared a mixed intervention versus a behavioural intervention (electronic reminder device), but study authors were contacted and reported that data on unplanned admissions and mortality were not collected for the electronic reminder device group.
Secondary outcome ‐ condition‐specific outcomes
No studies were identified.
Secondary outcome ‐ cost‐effectiveness
No studies were identified.
Discussion
Summary of main results
Our systematic review identified a range of simple to complex interventions for improving medication‐taking ability and medication adherence in older adults prescribed multiple medications. All included studies evaluated interventions versus usual care.
No studies of educational only or behavioural only interventions for improving medication‐taking ability were identified. We were unable to determine the impact of mixed interventions on medication‐taking ability due to the small number of eligible studies, variations in effects of the interventions across studies, and large variations in design and quality of those studies.
Our findings suggest that interventions with behavioural components alone and mixed educational and behavioural interventions compared with usual care may improve the proportion of people who are adherent to their prescribed medication. However, we are uncertain of the effects of educational only interventions on the proportion of people who are adherent to their prescribed medication.
When adherence is measured as a continuous variable (e.g. percentage of pills taken), our findings suggest that educational only and mixed interventions may have little or no impact on adherence, and we are uncertain of the effects of behavioural only interventions.
These results must be interpreted with a degree of caution, given the variations in intervention design, duration, follow‐up, and risk of bias of included studies, along with the overall low or very low quality rating of evidence for these outcomes.
Within mixed interventions for improving medication adherence, all of the individual studies that had a significant impact on medication adherence (11 out of 27 studies) were conducted at the hospital‐community/primary care interface (e.g. in hospital, at discharge, post discharge, at outpatient clinics). Three formal subgroup analyses performed on studies involving mixed interventions found no significant differences between subgroups based on intervention duration, type of outcome measure, or health provider delivering the intervention. Studies involving behavioural only interventions were too few to enable subgroup analyses.
Several secondary outcomes were evaluated, but heterogeneity and concerns regarding risk of bias limited our ability to draw firm conclusions. Large variations in outcome measures also limited our ability to pool results via meta‐analyses for most secondary outcomes such as medication knowledge and consumer satisfaction, and many secondary outcomes were reported in only a limited number of studies (e.g. condition‐specific outcomes ‐ 7 studies, adverse drug reactions (ADRs) ‐ 6 studies, and cost‐effectiveness of interventions ‐ 4 studies).
Overall, we found that mixed educational and behavioural interventions may improve medication knowledge but to a variable degree across studies. Pooled results suggest that mixed interventions may reduce the number of emergency department (ED)/hospital admissions compared with usual care, although studies unable to be included in the meta‐analysis suggest that the interventions may have little or no effect on these outcomes. Mixed interventions may lead to little or no change in health‐related quality of life (HRQoL), and we are uncertain about effects on mortality, consumer satisfaction, and other secondary outcomes such as ADRs, physician visits, and costs for these interventions.
Educational interventions delivered alone and compared with usual care may have little or no effect on medication knowledge and probably have little or no impact on most secondary outcomes, including HRQoL and ED/ hospital admissions. We are uncertain of the effects of behavioural interventions delivered alone on the above outcomes (HRQoL, ED/hospital admissions, knowledge) and of the effects of educational or behavioural interventions on a range of other secondary outcomes including satisfaction, ADRs, physician visits, and costs. We identified no studies evaluating the effects of educational or behavioural interventions delivered alone on mortality.
Six studies reported a comparison between two interventions as part of a three‐arm randomised controlled trial (RCT) design; however, due to the limited number of studies assessing the same types of interventions and comparisons, we were unable to draw firm conclusions for any primary or secondary outcomes.
Overall completeness and applicability of evidence
Most of the studies included in this review are relatively new, with almost half (24/50) published since 2010. This trend most likely reflects the increasing prevalence of multiple medication use in older adults and increasing efforts to improve medication adherence. In contrast, three of the five interventions evaluating medication‐taking ability were published last century.
Studies from four continents were identified. Most studies were from high‐income countries, and the greatest proportions emanated from the USA (21), the UK (8), and Canada (5). The results of this review may be more applicable to older adults residing in developed countries, mostly Western countries, with only two studies identified in non‐Western countries (one each from China and Singapore).
Although we attempted to pool interventions under three broad categories (educational only, behavioural only, or mixed), a large degree of heterogeneity remained within each category, which impacted both our confidence in, and the potential applicability of, our findings. For example, among the mixed interventions that showed positive effects on medication adherence, interventions varied from one‐off to two years in duration, were delivered by various health professionals face‐to‐face and/or via telephone, and involved one or more behavioural components (e.g. regimen simplification, motivational interviewing, adherence aids, reminder cards). Furthermore, although interventions were primarily compared with usual care, variation in definitions of 'usual care' likely influenced the size of the effect for some interventions. Interventions that were compared to usual care that involved some form of medication counselling, education, or monitoring may have been less likely to impact medication‐taking ability and medication adherence than interventions that were compared to usual care that more closely resembled a pure control (no intervention) group. However, the often poor description of usual care made this difficult to assess. Our review found only a limited number of studies comparing one intervention versus another intervention. Further research assessing this comparison in different ways may help to identify the most clinically effective and cost‐effective interventions, without the potential ethical dilemmas inherent when health‐related interventions are compared to a pure control consisting of minimal or no intervention.
Reporting of medication use was generally poor and inconsistent. This may have resulted in exclusion of potentially eligible studies for which authors did not collect and/or clearly report the number of medications participants were taking. We also noted variation in the types of medications reported across studies, with some reporting only prescription medication, some reporting only regular medication, and some reporting all medications (e.g. regular, when required, prescription, non‐prescription medications). It was also unclear at times whether the interventions targeted all medications taken by participants, and whether assessment of medication‐taking ability or medication adherence applied to all medications or only to a subset. There is a need for clearer reporting of medication use in intervention studies targeting medication‐taking ability or medication adherence.
Reporting of risk factor(s) for poor medication‐taking ability and/or medication adherence was also inconsistent or non‐existent. Common risk factors such as number of medications, frailty, and cognitive impairment were extracted when possible, but subgroup analysis based on such factors was not possible. There is a need for clearer identification and reporting of patient, therapy, condition, system, and environmental factors that influence both medication‐taking ability and medication adherence, and that may be targeted by interventions to improve medication‐taking ability and adherence. Clearer reporting may also highlight and clarify the sometimes inter‐related nature of medication‐taking ability and medication adherence, as many of the interventions described within this review likely targeted aspects of medication‐taking ability (e.g. regimen simplification) in an attempt to improve medication adherence but did not directly measure medication‐taking ability as an outcome.
A limited number of studies evaluated clinical health outcomes, including condition‐specific outcomes and adverse events. Clinical health outcomes should be measured for any intervention that may result in changes in a person's medication intake, including any changes (decrease or increase) in medication adherence. ED and hospital admissions were reported in 23 studies; however, it is often unclear how many were unplanned admissions and/or medication‐related admissions. Only six studies reported the number of participants experiencing ADRs, and four studies reported the number of primary care physician visits, which may be a surrogate for minor to moderate medication‐related problems. Condition‐specific outcomes such as blood pressure or blood glucose were reported in seven studies, meaning the clinical impact of changes in medication‐taking ability or medication adherence was omitted from most studies. Improving clinical outcomes and preventing adverse health outcomes should be the priority of any intervention aimed at improving medication use. This is particularly important among older adults prescribed multiple medications, as this population is at increased risk of adverse events. Evidence of clinical impact and reduction in adverse health outcomes is also necessary to drive translation of research into clinical practice and policy.
Studies that evaluated the cost‐effectiveness of interventions to improve medication‐taking ability or medication adherence are scarce. Future studies should include appropriate measures of cost‐effectiveness of interventions to assist decision‐makers in allocating healthcare resources efficiently.
Quality of the evidence
We evaluated the certainty of the body of evidence using the GRADE approach for the two primary outcomes ‐ medication‐taking ability and medication adherence ‐ and for three key secondary health outcomes ‐ HRQoL, ED/hospital admissions, and mortality. Overall there were serious concerns related to risk of bias and inconsistencies, resulting in low‐ or very low‐quality evidence for most outcomes, except for the effects of educational interventions on HRQoL and ED/hospital admissions, for which evidence was considered of moderate quality.
Most studies had unclear or high risk of bias across multiple domains, particularly related to random sequence generation and allocation concealment. Nearly all studies (98%) also had high risk of performance bias, but we acknowledge that blinding of participants and clinicians (providers of the intervention) is often impossible to achieve in pragmatic health services research. More than half of the included studies were also rated as having unclear or high risk of detection, attrition, and/or reporting biases. Studies were commonly given an ‘unclear’ risk of bias rating due to poorly described methods or results. Future studies should strongly consider prospective registration of the trials in registries, publication of study protocols with detailed methods and description of outcome measures, and more rigorous study methods and reporting.
Serious concerns related to inconsistencies were identified due to the heterogeneous and complex nature of the interventions (components, providers, settings, duration) and variations in outcome measures. Although interventions were broadly grouped as educational, behavioural, or mixed interventions, moderate to high heterogeneity was evident in meta‐analyses and in most cases contributed to our limited certainty in the results.
Concerns related to imprecision were also present for behavioural interventions, for which participant numbers and event rates were often low and/or confidence intervals were wide.
Potential biases in the review process
Differing terminologies for medication‐taking ability and medication adherence may have limited the number of studies found, despite our use of broad search terms and our search of grey literature and the reference lists of included studies. As mentioned previously, poor reporting of medication use may have caused us to miss potentially eligible studies.
Agreements and disagreements with other studies or reviews
Our review is the first to evaluate interventions for improving medication‐taking ability in older adults prescribed multiple medications. Reviews of instruments to assess medication‐taking ability ‐ Elliott 2009 ‐ and self‐efficacy for medication management ‐ Lamarche 2018 ‐ have been published, but no reviews of interventions utilising these instruments have been published to date.
Several reviews have investigated medication adherence, but only four reviews to date have evaluated interventions for improving medication adherence in older people taking multiple medications. Two of these reviews were conducted over a decade ago (George 2008;Williams 2008), and two were published more recently ‐ one focusing on theory‐based interventions only (Patton 2017), and the other including both cross‐sectional analyses of prevalence of medication adherence and clinical trials/systematic reviews of interventions targeting adherence (Zelko 2016). Both Patton 2017 and Zelko 2016 included far fewer intervention studies than are included in our systematic review because of their different inclusion criteria and search strategies. Our findings are consistent with the findings of these four previous reviews, all of which concluded that high‐quality evidence is scarce, and that inconsistencies in methods, interventions, and outcome measures have made it difficult to draw firm conclusions regarding the most effective interventions for improving medication adherence.
Our concerns regarding risk of bias of included studies were also consistent with a previous Cochrane systematic review investigating interventions for enhancing medication adherence (Nieuwlaat 2014), which evaluated all RCTs of interventions to improve adherence with prescribed medications (i.e. not restricted to older adults or multiple medications) and found that only 17 of 182 included studies had the lowest risk of bias for study design features and their primary clinical outcome.
Study flow diagram.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Comparison 1: Interventions versus usual care, Outcome 1: Primary outcome: adherence, grouped by types of interventions (dichotomous)
Comparison 1: Interventions versus usual care, Outcome 2: Primary outcome: adherence, grouped by types of interventions (continuous)
Comparison 1: Interventions versus usual care, Outcome 3: Primary outcome: adherence, mixed interventions, grouped by intervention duration (dichotomous)
Comparison 1: Interventions versus usual care, Outcome 4: Primary outcome: adherence, mixed interventions, grouped by intervention duration (continuous)
Comparison 1: Interventions versus usual care, Outcome 5: Primary outcome: adherence, mixed interventions, grouped by subjective or objective outcome measures (dichotomous)
Comparison 1: Interventions versus usual care, Outcome 6: Primary outcome: adherence, mixed interventions, grouped by subjective or objective outcome measure (continuous)
Comparison 1: Interventions versus usual care, Outcome 7: Primary outcome: adherence, mixed interventions, grouped by provider (dichotomous)
Comparison 1: Interventions versus usual care, Outcome 8: Primary outcome: adherence, mixed interventions, grouped by provider (continuous)
Comparison 1: Interventions versus usual care, Outcome 9: Secondary outcome: ED/Hospital admissions, grouped by type of intervention (dichotomous)
Comparison 1: Interventions versus usual care, Outcome 10: Secondary outcome: mortality, mixed interventions
| Mixed educational and behavioural interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving both educational and behavioural components Comparison: usual care | |||
| Outcomes | Impacts | No of Studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: 2 weeks to 12 months | The effects of mixed interventions on medication‐taking ability were unable to be determined. Meta‐analysis was not possible due to all 5 studies using different outcome measures. Of the 5 studies, 1 demonstrated significant improvement in medication‐taking ability, 2 showed no significant impact, 1 did not test for differences between groups, and 1 did not report results | 5 | Lowa,b |
| Medication adherence (dichotomous) Follow‐up: 1 to 18 months | Mixed interventions may improve the proportion of people who are adherent (dichotomous adherence outcome) Twelve studies (3147 participants) were included in a meta‐analysis. Risk ratio was 1.22 (95% CI 1.08 to 1.37), indicating interventions increased the absolute number of adherent participants by 12.8% (4.6% to 21.5%) Two studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study reported the intervention increased the number of medications taken correctly; 1 study showed no differences between groups | 14 | Lowa,b |
| Medication adherence (continuous) Follow‐up: 1 to 12 months | Mixed interventions may have little or no impact on medication adherence measured by continuous adherence outcomes (e.g. proportion of pills dispensed or taken) Seven studies (1825 participants) were included in a meta‐analysis. Standardised mean difference was 0.47 (95% CI ‐0.08 to 1.02), indicating that the mean adherence score in the intervention group was 0.47 standard deviations higher (0.08 lower to 1.02 higher) than in the usual care group Four studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study showed fewer medication errors as a proportion of total doses with the intervention; 3 studies showed no significant effect on adherence. Two additional studies were excluded due to unclear reporting of results | 13 | Lowb,c |
| Health‐related quality of life Follow‐up: 6 to 18 months | Mixed interventions may lead to little or no change in health‐related quality of life. Six of 7 studies showed no significant impact on this outcome. One study reported the intervention may improve both physical and mental summary scores on the SF‐36 at 12 months. Meta‐analysis was not possible due to differences in scales used and differences in reporting of results | 7 | Lowa,b |
| Emergency department (ED)/Hospital admissions Follow‐up: 1 to 24 months | Mixed interventions may reduce the number of emergency department (ED) and/or hospital admissions. Eleven studies (1827 participants) were included in meta‐analysis. Risk ratio was 0.67 (95% CI 0.50 to 0.90), indicating mixed interventions may reduce the absolute number of patients admitted to ED/hospital by 12.3% (18.7% to 3.7% fewer). Six studies were excluded from the meta‐analysis due to alternate reporting of outcome data; none of these studies reported differences between groups in ED/hospital admissions | 17 | Lowa,b |
| Mortality Follow‐up: 3 to 24 months | We are uncertain of the effects of mixed interventions on mortality. Seven studies (1776 participants) were included in a meta‐analysis. Risk ratio was 0.93 (95% CI 0.67 to 1.30), with an anticipated absolute effect of 0.9% fewer deaths (4.1% fewer to 3.8% more). One study was excluded from meta‐analysis due to incomplete information | 8 | Very lowa,b,d |
| CI: confidence interval; SF‐36: Short Form Health Survey‐36. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures, and because of high levels of heterogeneity in results. cOne mark deducted due to high or unclear risk of bias across multiple domains and inclusion of studies at risk of attrition bias in meta‐analysis. dOne mark deducted due to imprecision with limits of confidence intervals including both substantial potential benefit and harm. | |||
| Educational interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving educational components only Comparison: usual care | |||
| Outcomes | Impacts | No of studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: N/A | No studies that evaluated medication‐taking ability were found | ‐ | ‐ |
| Medication adherence (dichotomous) Follow‐up: 1 to 6 months | We are uncertain of the effects of educational interventions on the proportion of people who are adherent Two studies (182 participants) using dichotomous measures of adherence were included in a meta‐analysis. Risk ratio was 1.66 (95% CI 1.33 to 2.06), indicating that educational interventions increased the absolute number of adherent participants by 31.1% (15.6% to 50.1% more) Three studies were excluded from the meta‐analysis due to alternate reporting of outcome data: 1 study reported that the intervention increased the number of resolved medication issues (including non‐adherence); 2 studies reported no significant effect on adherence | 5 | Very lowa,b,c |
| Medication adherence (continuous) Follow‐up: 1 to 12 months | Educational interventions may have little or no impact on medication adherence measured by continuous adherence outcomes (e.g. proportion of pills dispensed or taken) Five studies (1165 participants) using continuous measures of adherence were included in a meta‐analysis. Standardised mean difference was 0.16 (95% CI ‐0.12 to 0.43), indicating that the mean adherence score in the intervention group was 0.16 standard deviations higher (0.12 lower to 0.43 higher) than in the usual care group Four studies were excluded from the meta‐analysis: 2 due to alternate reporting of outcome data (neither showed a difference between groups); 2 did not report results | 9 | Lowa,b |
| Health‐related quality of life Follow‐up: 3 to 12 months | Educational interventions probably have little or no effect on health‐related quality of life, with all 6 studies reporting no differences between groups. Meta‐analysis was not possible due to differences in scales used and differences in reporting of results | 6 | Moderatea |
| ED/Hospital admissions Follow‐up: 4 to 28 weeks | Educational interventions probably have little or no effect on ED/hospital admissions. Three studies (554 participants) were included in a meta‐analysis. Risk ratio was 1.02 (95% CI 0.71 to 1.48), indicating no change in the number of patients admitted to ED/hospital. One further study not included in meta‐analysis, reporting mean number of days in hospital, found no differences between groups | 4 | Moderatea |
| Mortality Follow‐up: N/A | No studies that evaluated the effects of educational interventions on mortality were found | ‐ | ‐ |
| CI: confidence interval; ED: emergency department. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures, and because of high levels of heterogeneity in results. cOne mark deducted due to imprecision ‐ small total number of participants and only two studies in meta‐analysis (one had very wide confidence interval and low events) plus the number of adherent patients (i.e. events) were not clearly reported in the two studies excluded from meta‐analysis. | |||
| Behavioural interventions aimed at improving medication‐taking ability and/or medication adherence compared with usual care for older community‐dwelling patients taking multiple medications | |||
| Patient or population: older patients using at least 4 regular prescription medications (and/or their carers) Settings: community setting (including discharge from a hospital or other healthcare facility to the community) Intervention: interventions involving behavioural components only Comparison: usual care | |||
| Outcomes | Impacts | No of studies | Quality of the evidence |
|---|---|---|---|
| Medication‐taking ability Follow‐up: N/A | No studies that evaluated medication‐taking ability were found | ‐ | ‐ |
| Medication adherence (dichotomous) Follow‐up: 3 to 18 months | Behavioural interventions may improve the proportion of people who are adherent (dichotomous adherence outcome) Four studies (528 participants) were included in a meta‐analysis. Risk ratio was 1.22 (95% CI 1.07 to 1.38), indicating behavioural interventions increased the absolute number of adherent participants by 10.5% (3.3% to 18.1% more) | 4 | Lowa,b |
| Medication adherence (continuous) Follow‐up: 6 to 12 months | We are uncertain of the effects of behavioural interventions on medication adherence when continuous measures of adherence are used Three studies were identified, but results could not be pooled in a meta‐analysis due to differences in reporting. All 3 reported significant impact on medication adherence, 2 showed large effects on adherence based on pill count, and 1 showed moderate improvement in self‐reported adherence using daily logbooks to calculate percentage of days adherent | 3 | Very lowa,b,c |
| Health‐related quality of life Follow‐up: 3 months | We are uncertain of the effects of behavioural interventions on health‐related quality of life. Only 1 study was identified, which found that the intervention resulted in worsening quality of life using the Minnesota Living With Heart Failure Questionnaire | 1 | Very lowa,d,e |
| ED/Hospital admissions Follow‐up: 3 to 6 months | We are uncertain of the effects of behavioural interventions on ED/hospital admissions. Two studies (70 participants) were included in a meta‐analysis. Risk ratio was 0.21 (95% CI 0.08 to 0.55), indicating behavioural interventions may reduce the absolute number of patients admitted to ED/hospital by 42.9% (49.9% to 24.4% fewer) | 2 | Very lowa,f |
| Mortality Follow‐up: N/A | No studies that evaluated the effects of behavioural interventions on mortality were found | ‐ | ‐ |
| CI: confidence interval; ED: emergency department. | |||
| GRADE Working Group grades of evidence. | |||
| aOne mark deducted due to high or unclear risk of bias across multiple domains including sequence generation and allocation concealment. bOne mark deducted due to variations in intervention, provider, setting, duration, and outcome measures. cOne mark deducted due to low participant numbers. dOne mark deducted due to indirectness of evidence as the Minnesota Living With Heart Failure Questionnaire is specific for heart failure populations and results may not be generalisable to general population of older people. eOne mark deducted due to low participant numbers from a single small study. fTwo marks deducted due to low participant numbers and low number of events. | |||
| Study | Measure of medication‐taking ability | Outcome |
|---|---|---|
| Begley 1997 | Objective measure: 5‐task dexterity test (e.g. opening child‐resistant closure), 1 point awarded for each successfully completed activity. Note: no difference across groups at baseline ‐ mean (SD) group A: 7.8 (1.3), group B: 7.5 (1.5), group C: 8.0 (1.4) | Objective measure: follow‐up results not reported |
| Cargill 1992 | Objective measure: behaviour score/100 for congruency between supply of medications on hand and prescribed medications (/40), verbalising correct regimen (/30), maintaining each prescribed med (/20), appropriate use of OTC (/10). Points deducted for sequestering old scripts, inappropriate use of alternative medications, or mixing medications together | Mean read from graph: Control: 74 vs intervention (group 3); 86 vs intervention (group 2); 84 vs intervention (group 3): 86 |
| Lingler 2016 | Objective measure: Medication Management Instrument for Deficiencies in the Elderly (MedMaIDE). MedMaiDE uses interview and observation to assess ability to self‐administer medications in 3 areas: knowledge of medications, how to take medications, and how to procure medications. Each medication is reviewed during administration. Scores 0 to 13, max total deficiency score is 13 | Baseline: mean ± SD intervention 0.833 ± 0.745 vs control 0.692 ± 0.768 Unpublished follow‐up results: mean ± SD: intervention 0.595 ± 0.725 vs control 0.297 ± 0.777; both groups showed significant decreases in number of medication management problems at 2 months (P < 0.01) |
| Manning 2007 | Subjective measure: self‐reported safety. Since discharge, how many mistakes have you made taking your medications (score 0 to 4)? | Mean ± SD: intervention 0.78 ± 0.4187 (n = 72) vs control 0.79 ± 0.4113 (n = 57) |
| Pereles 1996 | Objective measure: assessed differently for each group: intervention = pharmacist assessment with input from other team members, primarily based on having made 2 or fewer errors at stage 2 of the inpatient self‐medication programme ‐ considered able to self‐medicate at discharge. Control = pharmacist assessment with input from other team members at time of discharge counselling. YES/NO ‐ self‐medicating at discharge (note: there could be reasons other than failing the SMP that might explain why they were not self‐medicating at discharge, such as patient preference) | n (%): intervention 39 (76.5%) vs control 39 (69.6%) |
| OTC: over‐the‐counter. SD: standard deviation. | ||
| Study | Measure of adherence | Outcome |
|---|---|---|
| Al‐Rashed 2002 | Objective measure: percentage compliance using home medicines stocks and refill prescriptions between visits 1 and 2 | Intervention: 70% (n = 342 medications) |
| Blalock 2010 | Subjective measure: Brief Medication Questionnaire (5‐item regimen screen that assesses how medication is used) | Not reported |
| Bond 2007 | Subjective measure: Extended Medication Adherence Report Scale (MARS) questionnaire (12 statements about medicine‐taking; score range 12 to 60) | Med (IQR): intervention 59 (57 to 60) vs control 59 (57 to 60) |
| Cargill 1992 | Objective measure: pill count, percentage of pills taken compared to those prescribed | Mean scores: control: 74.5; intervention (group 3): 76.2; intervention (group 2): 74; intervention (group 3): 76.2 |
| Cohen 2011 | Objective measure: medication possession ratios | Not clearly reported |
| Hanlon 1996 | Subjective measure: self‐report proportion of medications for which the patient's response agreed with the directions for use on the action profile | Intervention 77.4% (n = 86 people) vs usual care 76.1% (n = 83 people) |
| Holland 2007 | Subjective measure: Medication Adherence Report Scale (MARS) scores from 5 (very poor adherence) to 25 (perfect adherence) | Mean (median): 23.74 (25), n = 101 vs 23.55 (25), n = 103 |
| Krska 2001 | Subjective measure: pharmaceutical care issues including potential or actual compliance issues, number of baseline issues resolved at 3 months | 51 of 74 issues resolved (n = 168) vs 21 of 69 issues resolved (n = 164) |
| Lim 2004 | Subjective measure: self‐reported; patients asked if they 'forgot to take medication as directed'. Then categorised as least compliant (compliant base, not at 2 months), not compliant (not compliant at base or at 2 months), compliant (compliant at 2 months) | Not clearly reported; unadjusted OR 1.50, 90% CI 0.73 to 3.08 |
| Marek 2013 | Objective measure: machine recorded or nurse pill count, average percentage of correct doses per month | Not reported for control group MD.2: 98.8% (SD 0.32), planner: 97.4% (SD 5.19) |
| Pandey 2017 | Subjective measure: participants used a logbook to record name and timing of medications taken on a daily basis. Absolute medication adherence calculated as percentage of total prescribed doses that were actually taken each month. 12 months adherence calculated as the mean of each of the 12 monthly measurements. Adherence outcome is % of days covered | Intervention: 91% (n = 9), control: 73% (n = 8) |
| Pereles 1996 | Objective measure: patients discharged with 40 days worth of medication, pill count conducted in home at 40 days. Number of medication errors as a proportion of the total doses administered | Not clearly reported After controlling for age and MMSE ‐ I: 0.045, C: 0.086; P < 0.001 |
| Shimp 2012 | Objective measure: medication possession ratios defined as sum of all days of medication supply received during year divided by total numbers of days supply needed ‐ calculated for top 8 drug classes for chronic conditions | Not reported ‐ MPRs were very high for both groups (range 0.84 to 0.96), and no clinically meaningful changes were observed over time for either group. Fewer patients reported missed doses after the intervention |
| Taylor 2003 | Subjective measure: self‐reported number of medication doses missed. Presented as % adherence | Intervention mean 100 vs control mean 88.9 (± SD 6.3) |
| Volume 2001 | Subjective measure: Morisky Adherence, scores 0 to 4; lower scores = better adherence | Mean SD: 0.56 ± 0.75 vs 0.47 ± 0.69; number of participants in each group unclear |
| Willeboordse 2017 | Subjective measure: self‐reported adherence problems | Persistence of adherence problems = OR 0.83 (0.54 to 1.27) (P = 0.38) (unpublished = adherence worsened or persisted: 65 vs 54; adherence improved or remained the same: 143 vs 144) |
| Winland‐Brown 2000 | Objective measure: pill count, average number of missed doses (unclear over what time period). Please note: group 2 vs group 3 was used for comparison of intervention vs usual care; group 1 vs group 2 was used for comparison of intervention vs intervention | Mean: group 1 = 15.1, group 2 = 1.7, control = 19.7 |
| Outcome results presented as intervention group vs usual care group unless otherwise stated. C: control. CI: confidence interval. I: intervention. IQR: interquartile ratio. MMSE: Mini‐Mental State Examination. MPR: medication possession ratio. OR: odds ratio. SD: standard deviation. | ||
| Study | Measure of medication knowledge | Outcome |
|---|---|---|
| Al‐Rashed 2002 | Pharmacist‐delivered questionnaire; percentage scores for correct answers (drug use, dose, dosage interval) | Drug use: 97.4% vs 69.5%; dosage interval: 97.4% vs 86.0%; dose: 98.5% vs 91.5% |
| Begley 1997 | Patients asked about name, purpose, dose, dosage frequency, and side effects. Reported as percentage of correct answers. Accuracy compared to hospital discharge or GP instructions | Group A 70%, Group B 68%, Group C 66% (usual care) |
| Bernsten 2001 | Interview‐based questionnaire calculating percentage correct (looking at 4 areas: indication, number of dosage units taken per dose, number of doses per day, and awareness of potential adverse effects). Higher scores = better knowledge | Mean ± SD change at 18 months: +3.19 ± 15.18 (n = 704) vs +3.16 ± 16.19 (n = 636) |
| Bond 2007 | Patients were asked whether they "knew more about their medicines compared with a year ago" on 5‐point Likert scale. Those who said agree/strongly agree | Trial report: 73% vs 65% |
| Grymonpre 2001 | Knows purpose of prescribed drugs (yes/no), expressed as number and percentage of drugs correct | 304/327 (93%) vs 335/373 (90%) |
| Hanlon 1996 | Self‐report knowledge of 'how they took each analysed medication and what the medication was for'; percentage of correct responses | 89.4% (n = 86) vs 90.6% (n = 83) |
| Khdour 2009 | COPD knowledge questionnaire (validated) ‐ effectiveness of education in helping persons with COPD. 16 T/F questions, correct response = 1, range 0 to 16, higher score = better knowledge | Median (IQR): 75.0 (32.0) vs 59.3 (33.0) |
| Lim 2004 | Composite knowledge of dose (D), frequency (F), and indication (I), percentage correct | Not reported |
| Manning 2007 | Assessment of knowledge of indication, dosage frequency, and special comments or cautions. 0 (for no correct responses) to 3 (all correct responses) | Mean ± SD: 1.96 ± 0.7561 vs 1.66 ± 0.6851 |
| Messerli 2016 | Knowledge of medicines and daily use ‐ phone questionnaire. 58 questions ‐ included assessing knowledge | Not reported |
| Nazareth 2001 | Prescription medicine interview ‐ patient's knowledge of prescribed drugs. Validated self‐report semi‐structured interview (knowledge score is out of 1, with 1 being 'total/highest' knowledge). Mean (SD) out of 1 | Mean ± SD: 0.69 ± 0.35 (n = 65) vs 0.68 ± 0.32 (n = 68) |
| Pereles 1996 | "Short medication knowledge questionnaire" = Patients asked to name and describe appearance and purpose of their medication, to describe their regimen and any potential side effects or drug interactions. Percentage of correct responses in each knowledge category | Discharge: name: 69% vs 55%; appearance: 77% vs 66%; times: 80% vs 69%; purpose: 77% vs 72%; side effects: 6% vs 4% |
| Taylor 2003 | Self‐reports used to assess medication knowledge. Score determined by dividing the number of medications for which a patient reported the correct name, purpose, dose, and frequency by the total number of medications and multiplying by 100 | Mean ± SD: 92.6 ± 3.4 vs 42.9 ± 12.8 |
| Outcome results presented as intervention group vs usual care group unless otherwise stated. COPD: chronic obstructive pulmonary disease. GP: general practitioner. IQR: interquartile ratio. SD: standard deviation. T/F: true/false. | ||
| Study | Measure of satisfaction | Outcome (intervention vs usual care) |
|---|---|---|
| Bernsten 2001 | Self‐reported rating of services provided, satisfaction with services, and general opinion of pharmaceutical care. Questionnaire administered by pharmacist. Results presented as percentage who agree/mainly agree | Rating of services as excellent: 73.8% vs 64.6%; satisfaction with services: 93.9% vs > 90%; general opinion: 77% (intervention group only) |
| Bond 2007 | Overall score on 15 positive and negative statements of most recent pharmacy visit (total score 15 to 75, higher scores better) | Median (IQR): 46 (40 to 55) vs 43 (38 to 49) |
| George 2016 | User satisfaction regarding use of the computer programme questionnaire (USUCPQ): an 8‐item measure based on 7‐point Likert score (max score 56, higher scores better) | Mean ± SD total satisfaction: 45.33 ± 7.81 vs 44.68 ± 6.75 |
| Hanlon 1996 | Health Care Attitude Questionnaire: 3 questions on pharmacy‐related healthcare satisfaction (directions received, explanation of SES, numbers/types of drugs) based on 5‐point Likert scale (lower scores better) | Mean ± SD total score: 5.2 ± 1.5 vs 5.4 ± 1.7 |
| Holland 2007 | Satisfaction questionnaire; usefulness of community pharmacist visits | 75 (64%) considered the visits to have been extremely or very useful |
| Lingler 2016 | Acceptability of the intervention using a set of Likert scale questions and eliciting open‐ended comments | 88% of caregivers reported intervention topics useful and relevant; 92% reported that the intervention was helpful for managing the patient's treatment plan |
| Lopez Cabezas 2006 | Catalan Health Department satisfaction survey, asking participants about the care and information received and asking them to provide a global scoring (0 to 10) | Mean ± SD 8.9 ± 1.3 vs 8.8 ± 1.5 |
| Manning 2007 | Level of satisfaction using 5‐point Likert scale (5 = highest): "How satisfied were you with the form you received from the nurse when she/he was talking to you about your medications?" | Mean ± SD 4.24 ± 0.6986 vs 4.26 ± 0.8768 |
| Naunton 2003 | Survey of intervention group only | 94% very satisfied; 84% stated information they were given 'helped a great deal' |
| Nazareth 2001 | Validated patient satisfaction questionnaire, each item scored 1 to 4, mean score per item calculated (higher = better) | Mean ± SD 3.4 ± 0.6 (n = 62) vs 3.2 ± 0.6 (n = 61) |
| Taylor 2003 | Mean ± SD number of patients with pharmacy‐related satisfaction (details unclear) | Mean ± SD 81.9 ± 4.8 (n = 33) vs 89.0 ± 6.2 (n = 36) |
| Volume 2001 | Satisfaction with pharmacy services using 34‐item instrument and 7‐point Likert scale (lower scores = better). General satisfaction extracted | Mean ± SD 1.53 ± 0.77 vs 1.62 ± 0.88 |
| Willeboordse 2017 | Medication satisfaction questionnaire assessed on a 7‐point Likert scale | B (95% CI): 0.11 (‐0.08 to 0.30) (P = 0.25) |
| Outcome results presented as intervention group vs usual care group unless otherwise stated. CI: confidence interval. IQR: interquartile ratio. SD: standard deviation. | ||
| Study | Measure | Time point | Outcome |
|---|---|---|---|
| Bernsten 2001 | SF‐36 | 18 months | Change: GH: +0.28 vs ‐0.66, MH: ‐0.80 vs ‐1.34, PF: ‐0.95 vs‐0.68 |
| Bond 2007 | SF‐36 | 12 months | Med (IQR): GH: 52 (35 to 65) vs 50 (35 to 70), MH: 80 (64 to 88) vs 80 (64 to 88), PF: 60 (35 to 80) vs 65 (35 to 85) |
| EQ‐5D | 12 months | Med (IQR): 0.73 (0.7 to 0.9) vs 0.73 (0.7 to 0.9) | |
| Cohen 2011 | VR‐36 | 6 months | Change: Med (IQR): MH: 0.48 (‐3.37 to 4.32), C: 0.78 (‐2.67 to 4.23), PF: 1.65 (‐5.21 to 1.31), C: ‐1.95 (‐5.21 to 1.31) |
| Hale 2016 | MLHFQ | 90 days | Mean ± SD: 62.2 ± 20.6 vs 28.2 ± 22.3 |
| Hanlon 1996 | SF‐36 | 12 months | Mean ± SD: GH: 37.4 ± 1.6 vs 35.2 ± 1.7, MH: 61.1 ± 1.8 vs 60.4 ± 1.8, PF: 44.1 ± 2.0 vs 42.2 ± 2.0 |
| Holland 2007 | EQ‐5D, VAS | 6 months | Mean ± SD: EQ‐5D: 0.58 ± 0.29 vs 0.52 ± 0.34, VAS: 58.2 ± 19.6 vs 58.6 ± 19.8 |
| MLHFQ | Mean ± SD: 47.7 ± 26.3 vs 44.5 ± 27.9 | ||
| Khdour 2009 | SGRQ | 12 months | Mean (confidence interval): 61.8 (57.9 to 65.6) vs 65.3 (61.0 to 69.6) |
| Krska 2001 | SF‐36 | 3 months | No significant differences ‐ values not reported |
| Lopez Cabezas 2006 | EQ‐5D (Spanish | 12 months | Mean ± SD: 64 ± 15.4 vs 60.6 ± 17.8, subgroup > 70 years: 63.8 ± 15.3 vs 58.4 ± 15.9 |
| Marek 2013 | SF‐36 | 12 months | Comparison 1 ‐ Mean (confidence interval): planner (intervention) vs control (usual care) = PCS: 1.390 (0.816 to 1.963), MCS: 1.686 (0.949 to 2.423) |
| Comparison 2 ‐ Mean (confidence interval): MD.2 (intervention 1) vs planner (intervention 2) = PCS: 0.095 (‐0.450 to 0.640), MCS: 0.241 (‐0.459 to 0.940) | |||
| Muth 2016 | EQ‐5D | 12 weeks | Mean ± SD: change: ‐0.6 ± 19.61 vs ‐1.0 ± 13.66 |
| Taylor 2003 | SF‐36 | 12 months | Mean SD: GH: 57.0 ± 19.6 vs 50.1 ± 15.9, MH: 73.1 ± 21.2 vs 72.3 ± 17.1, PF: 68.6 ± 24.0 vs 56.1 ± 27.5 |
| Volume 2001 | SF‐36 | 12 to 13 months | Mean ± SD: MCS: 56.14 ± 8.30 vs 54.55 ± 8.65, PCS: 36.87 ± 11.62 vs 38.39 ± 11.44 |
| Willeboordse 2017 | SF‐12 | 6 months | Regression coefficients adjusted for baseline: PCS: ‐0.06 (‐3.19 to 3.06), MCS: 0.16 (‐2.89 to 3.22) |
| EQ‐5D‐3L | Regression coefficients adjusted for baseline: utility: 0.02 (‐0.02 to 0.05), VAS: 2.30 (‐0.16 to 4.76) | ||
| Outcome results presented as intervention group vs usual care group unless otherwise stated. C: control. EQ‐5D: EuroQoL Group Quality of Life Questionnaire based on 5 dimensions. GH: general health. I: intervention. MH: mental health. MCS: mental components summary. MLHFQ: Minnesota Living With Heart Failure Questionnaire (21 items, coded 0 to 5; higher scores indicate adverse impact on life). PCS: physical components summary. PF: physical function. SD: standard deviation. SF‐36: Short Form‐36 Health Survey. SGRQ: St. George's Respiratory Questionnaire (76 items, total score 100; higher = better). VAS: visual analogue scale. | |||
| Study | Time point | ED/Hospital admissions | Mortality | Adverse drug reactions | GP visits |
|---|---|---|---|---|---|
| Al‐Rashed 2002 | 3 months | Patients re‐admitted to hospital: 8/43 v 28/40 | Total unplanned visits: 43 (n = 43) vs 59 (n = 40) | ||
| Bernsten 2001 | 18 months | Self‐reported: 35.6% vs 40.4%; n values unclear | |||
| Chrischilles 2014 | 3 months | Self‐reported: 100/802 (12.9%) vs 33/273 (12.2%) | |||
| Cossette 2015 | 30 days | ED visits: 18% (n = 108) vs 20% (n = 95) | |||
| Haag 2016 | 30 days | ED or hospital re‐admission: 2/11 (18%) vs 1/11 (9%) | |||
| Hale 2016 | 90 days | No. participants: ED: 3/11 (27%) vs 6/14 (43%); hospitalisation: 1/11 (9%) vs 7/14 (50%); total no.: ED 4 vs 7, hospital 2 vs 8 | |||
| Hanlon 1996 | 12 months | Self‐reported: 30.2% (n = 86) vs 40% (n = 83) (P = 0.19) | |||
| Holland 2007 | 6 months | Total number of ED admissions (not number of participants admitted): 134 (n = 148) vs 112 (n = 143) | 30/148 vs 24/143 | ||
| Khdour 2009 | 12 months | n = 71 and 72; ED: 40 vs 80, hospital: 26 vs 64; total hospital days: 164 vs 466 | Unscheduled: 28 (n = 71) vs 47 (n = 72) | ||
| Lim 2004 | 2 months | Self‐reported and assessed by physician, total ADRs at 2 months: 13 vs 6; residual ADRs from baseline: 4/13 vs 4/8 | |||
| Lipton 1994 | 6 months | Total days in hospital: mean ± SD 2.29 ± 5.96, n = 350 vs 2.02 ± 5.83, n = 356 | |||
| Lopez‐Cabezas 2016 | 12 months | Patients with re‐admission: 23/70 (32.9%) vs 31/64 (48.4%) | > 70 years subgroup: 7/53 (13.2%) vs 13/50 (26.0%) | ||
| Marusic 2013 | 30 days | Re‐admission or ED: 20/80 (25%) vs 27/80 (33.8%) | ? Self‐reported: 24/80 (30%) vs 30/80 (37.5%) (P = 0.315) | ||
| Messerli 2016 | 28 weeks | Self‐reported unplanned GP visit or hospitalisation: total during study: 110 vs 99, n unclear? ‐ 181 vs 191 | * | ||
| Murray 1993 | 6 months | Self‐reported side effects, ill effects, or other problems with medication: C1: 2/12, C2: 2/10, Int: 1/9 | |||
| Muth 2016 | 12 weeks | Days in hospital: mean ± SD (T1 + T2) ‐ T0 = ‐0.4 ± 0.73 vs ‐0.2 ± 0.69, n unclear | |||
| Naunton 2003 | 90 days | 1 or more unplanned re‐admissions: 16/57 (28%) vs 29/64 (45%) | 3/57 (5%) vs 5/64 (8%) | ||
| Nazareth 2001 | 6 months | Re‐admissions: 38/136 (27.9%) vs 43/151 (28.4%) Outpatient department: 39/137 vs 40/151 | 22/137(16.1%) vs 19/151 (12.6%) | 76/107 vs 82/116 | |
| Olesen 2014 | 24 months | Unplanned admissions: 77/253 (30%) vs 73/264 (28%) | 19/253 (7.5%) vs 14/264 (5%) | ||
| Rich 1996 | 90 days | Re‐admissions: 18/80 (22.5%) vs 22/76 (28.9%) | |||
| Saez de la Fuente 2011 | 50 days | Total re‐admissions: 5 (n = 26) vs 7 (n = 24); ED: 7 (n = 26) vs 9 (n = 24) (note percentages listed in paper do not match n values) | 2 (? n = 26) vs 1 (? n = 24) | ||
| Shively 2013 | 6 months | Hospital: mean (SD): 0.21 (0.409) vs 0.32 (0.475) n = 39 vs 37 | |||
| Taylor 2003 | 12 months | Hospital: 2/33 vs 11/36; ED: 4/33 and 6/36 | |||
| Vinluan 2015 | 90 days | Hospital admissions: 2/7 vs 2/7 | 0/7 vs 2/7 | ||
| Willeboordse 2017 | 6 months | DRPs: baseline: 4.4 ± 1.9 vs 3.7 ± 1.7 % solved: 20.2 (12.2 to 28.1) | |||
| Winland‐Brown 2000 | 6 months | Hospitalisations G1: 4/16, G2: 3/24, C: 12/21 | Physician visits: G1: 1.5/month, G2: 1/month, C: 1/month | ||
| Wu 2006 | 2 years | Med (IQR): n = 219 vs 223 | 25/219 vs 38/223 | ||
| Young 2016 | 180 days | Hospital: 18/51 (35.3%) vs 20/49 (40.8%); ED visits: 12/51 (23.5%) vs 11/49 (22.4%) | |||
| Outcome results presented as intervention group vs usual care group unless otherwise stated. ADR: adverse drug reaction. C: control. C1: control group 1. C2: control group 2. DRP: drug‐related problem. ED: emergency department. G1: group 1. G2: group 2. GP: general practitioner. Int: intervention. SD: standard deviation. | |||||
| Study | Measure | Outcome |
|---|---|---|
| Blalock 2010 | Falls (self‐reported) in 12 months (ITT analysis) | ≥ 1 fall: 53/93 vs 52/93 |
| Bond 2007 | Total score (/8) for reaching targets at 12 months | 4.6 ± 1.2 vs 4.6 ± 1.1 |
| Cohen 2011 | Percentage achieving targets at 6 months (SBP < 130, LDL < 100, HbA1c < 7%) | 16% (n = 50) vs 4.1% (n = 49) |
| Lee 2006 | Systolic and diastolic blood pressure (mmHg) and LDL‐cholesterol (mg/dL) at 6 months post phase 1 | SBP: 124.4 ± 14.0 vs 133.3 ± 21.5 LDL: 87.5 ± 24.2 vs 88.4 ± 21.0 |
| Nascimento 2016 | Fasting blood glucose and HbA1c at 6 months | FBG: 117.3 ± 26.8 vs 142.2 ± 32.9 HbA1C: 7.7 ± 0.8 vs 7.99 ± 0.67 |
| Taylor 2003 | Number of people reaching goal level at 12 months | BP: 22 (91.7%) vs 8 (27.6%) INR: 4 (100%) vs 1 (16.7%) (Note: calculated mean across all 4 measures: 92% vs 19%) |
| Williams 2012 | Blood pressure, HbA1c, eGFR, and creatinine levels at 12 months (9 months post intervention) | SBP: mean (CI) ‐6.9 (‐13.8 to 0.02) vs ‐3.0 (‐8.4 to 2.4) HbA1c: med (IQR): 7 (7 to 9) vs 8 (7 to 9) eGFR: med (IQR): 48 (38 to 76) vs 46 (32 to 72) Creatinine: med (IQR): 117 (82 to 144) vs 108 (89 to 171) |
| Outcome results presented as intervention group vs usual care group unless otherwise stated and presented as mean ± SD unless otherwise stated. BMI: body mass index. BP: blood pressure. DBP: diastolic blood pressure. eGFR: estimated glomerular filtration rate. FBG: fasting blood glucose. HbA1c: glycosylated haemoglobin. INR: international normalised ratio. IQR: interquartile ratio. ITT: intention‐to‐treat. LDL: low‐density lipoprotein. SBP: systolic blood pressure. | ||
| Study | Measure of costs | Outcome |
|---|---|---|
| Bernsten 2001 | Direct costs of the study, including additional time spent by pharmacists, costs associated with contacts with other health professionals, costs of hospitalisation and drugs | Average cost per patient (saving): Denmark: 1298.13 vs 1419.88 (+121.75) Germany: 2992.25 vs 3167.25 (+175.00) Northern Ireland: 735.22 vs 750.01 (+14.79) Sweden: 1266.76 vs 1250.34 (‐16.42) |
| Bond 2007 | Total NHS‐related study costs, including costs of intervention and other treatment (e.g. medicines, hospital, other health consultations) | Median cost (IQR): 970.5 (667.0 to 1489.0) vs 835.2 (534.4 to 1396.3) Median (IQR) cost of intervention alone (pharmacist time and training): 90 (60 to 118) |
| Lipton 1994 | Medicare Part B charges, total hospital inpatient charges | Total charges: mean ± SD 2769 ± 4789 vs 2598 ± 3722 Inpatient charges: mean ± SD 5472 ± 10904 vs 5263 ± 11478 |
| Lopez Cabezas 2006 | Hospitalisation costs, adding in intervention direct costs, delivered materials and time spent by the pharmacist | Average cost per patient: 997 vs 1575 |
| Outcome results presented as intervention group vs usual care group unless otherwise stated. IQR: interquartile ratio. NHS: National Health Service. SD: standard deviation. | ||
| Study | Measure | Outcome (Intervention vs usual care) |
|---|---|---|
| Chrischilles 2014 | Mean (SD) number of medication management problems from a list of 8 problems, including questions on multiple prescribers, multiple pharmacies, mail order prescriptions, confusion whether medication was taken, taking medication without knowing indication, problems affording medications, feeling that medications are not working, and feeling that medications are not doing what they were intended to do | Mean ± SD 1.4 ± 1.4 vs 1.6 ± 1.5 |
| Lingler 2016 | Medication deficiency checklist: a 15‐item, investigator‐developed instrument that uses caregiver interviews to assess for the presence of errors and problems (e.g. incorrectly chewing pills or capsules, taking at the wrong time, repeating doses, patient refuses/unco‐operative) | Mean ± SD 2.19 ± 1.52 vs 2.36 ± 1.51 |
| Moral 2015 | Average number of medication errors, defined as both patient errors (e.g. omission of dose) and prescriber errors (e.g. dose too high or too low, duplicate therapy) (as reported in Perula de Torres 2014 paper) | Mean 0.429 vs 1.145 |
| Taylor 2013 | Number of participants with at least 1 medication misadventure (defined as medication errors, adverse drug events, and/or adverse drug reactions) | 2.8% (n = 33) vs 3.0% (n = 36) |
| Outcome results presented as intervention group vs usual care group unless otherwise stated. SD: standard deviation. | ||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1.1 Primary outcome: adherence, grouped by types of interventions (dichotomous) Show forest plot | 18 | Risk Ratio (M‐H, Random, 95% CI) | Subtotals only | |
| 1.1.1 Educational interventions | 2 | 182 | Risk Ratio (M‐H, Random, 95% CI) | 1.66 [1.33, 2.06] |
| 1.1.2 Behavioural interventions | 4 | 528 | Risk Ratio (M‐H, Random, 95% CI) | 1.22 [1.07, 1.38] |
| 1.1.3 Mixed educational and behavioural interventions | 12 | 3147 | Risk Ratio (M‐H, Random, 95% CI) | 1.22 [1.08, 1.37] |
| 1.2 Primary outcome: adherence, grouped by types of interventions (continuous) Show forest plot | 12 | Std. Mean Difference (IV, Random, 95% CI) | Subtotals only | |
| 1.2.1 Educational interventions | 5 | 1165 | Std. Mean Difference (IV, Random, 95% CI) | 0.16 [‐0.12, 0.43] |
| 1.2.3 Mixed educational and behavioural interventions | 7 | 1825 | Std. Mean Difference (IV, Random, 95% CI) | 0.47 [‐0.08, 1.02] |
| 1.3 Primary outcome: adherence, mixed interventions, grouped by intervention duration (dichotomous) Show forest plot | 11 | Risk Ratio (M‐H, Random, 95% CI) | Subtotals only | |
| 1.3.1 Short duration (≤ 3 months) | 6 | 486 | Risk Ratio (M‐H, Random, 95% CI) | 1.40 [1.13, 1.74] |
| 1.3.2 Long duration (> 3 months) | 5 | 2505 | Risk Ratio (M‐H, Random, 95% CI) | 1.11 [0.97, 1.27] |
| 1.4 Primary outcome: adherence, mixed interventions, grouped by intervention duration (continuous) Show forest plot | 7 | Std. Mean Difference (IV, Random, 95% CI) | Subtotals only | |
| 1.4.1 Short duration (≤ 3 months) | 3 | 398 | Std. Mean Difference (IV, Random, 95% CI) | 0.18 [‐0.52, 0.88] |
| 1.4.2 Long duration (> 3 months) | 4 | 1427 | Std. Mean Difference (IV, Random, 95% CI) | 0.70 [‐0.25, 1.65] |
| 1.5 Primary outcome: adherence, mixed interventions, grouped by subjective or objective outcome measures (dichotomous) Show forest plot | 12 | Risk Ratio (M‐H, Random, 95% CI) | Subtotals only | |
| 1.5.1 Objective outcome measure | 5 | 762 | Risk Ratio (M‐H, Random, 95% CI) | 1.13 [0.93, 1.38] |
| 1.5.2 Subjective outcome measure | 7 | 2385 | Risk Ratio (M‐H, Random, 95% CI) | 1.26 [1.09, 1.46] |
| 1.6 Primary outcome: adherence, mixed interventions, grouped by subjective or objective outcome measure (continuous) Show forest plot | 7 | Std. Mean Difference (IV, Random, 95% CI) | Subtotals only | |
| 1.6.1 Objective outcome measure | 3 | 360 | Std. Mean Difference (IV, Random, 95% CI) | 0.82 [‐0.49, 2.13] |
| 1.6.2 Subjective outcome measure | 4 | 1465 | Std. Mean Difference (IV, Random, 95% CI) | 0.21 [‐0.23, 0.66] |
| 1.7 Primary outcome: adherence, mixed interventions, grouped by provider (dichotomous) Show forest plot | 12 | Risk Ratio (M‐H, Random, 95% CI) | Subtotals only | |
| 1.7.1 Provider: pharmacist | 8 | 2672 | Risk Ratio (M‐H, Random, 95% CI) | 1.21 [1.04, 1.41] |
| 1.7.2 Provider: nurse | 2 | 297 | Risk Ratio (M‐H, Random, 95% CI) | 1.19 [1.02, 1.38] |
| 1.7.3 Provider: 2 or more health professionals | 2 | 178 | Risk Ratio (M‐H, Random, 95% CI) | 1.38 [0.88, 2.16] |
| 1.8 Primary outcome: adherence, mixed interventions, grouped by provider (continuous) Show forest plot | 6 | Std. Mean Difference (IV, Random, 95% CI) | Subtotals only | |
| 1.8.1 Provider: pharmacist | 2 | 286 | Std. Mean Difference (IV, Random, 95% CI) | 1.38 [0.01, 2.75] |
| 1.8.2 Provider: nurse | 2 | 140 | Std. Mean Difference (IV, Random, 95% CI) | ‐0.13 [‐0.53, 0.27] |
| 1.8.3 Provider: 2 or more health professionals | 2 | 324 | Std. Mean Difference (IV, Random, 95% CI) | 0.42 [‐0.38, 1.21] |
| 1.9 Secondary outcome: ED/Hospital admissions, grouped by type of intervention (dichotomous) Show forest plot | 16 | Risk Ratio (M‐H, Random, 95% CI) | Subtotals only | |
| 1.9.1 Educational interventions | 3 | 554 | Risk Ratio (M‐H, Random, 95% CI) | 1.02 [0.71, 1.48] |
| 1.9.2 Behavioural interventions | 2 | 70 | Risk Ratio (M‐H, Random, 95% CI) | 0.21 [0.08, 0.55] |
| 1.9.3 Mixed educational and behavioural interventions | 11 | 1827 | Risk Ratio (M‐H, Random, 95% CI) | 0.67 [0.50, 0.90] |
| 1.10 Secondary outcome: mortality, mixed interventions Show forest plot | 7 | 1776 | Risk Ratio (M‐H, Random, 95% CI) | 0.93 [0.67, 1.30] |
