Types of studies

We will include randomised trials comparing the intervention to either usual care or other interventions to improve professional practice.

We will include full text studies, conference abstracts, and unpublished data.

Types of participants

We will include practicing healthcare professionals and those in postgraduate training responsible for patient care. We will exclude undergraduate students or non‐professional (lay) healthcare workers.

Types of interventions

Examples of patient‐mediated interventions are:

• patient information where patients are informed about recommended care

• patient education/training/counselling to increase patients' knowledge about their condition

• patient feedback about clinical practice (collecting information from patients and giving it to professionals before, during or after an encounter)

• patient decision aids to ensure that the choices about treatment and management reflect recommended care and the patients' values and preferences

• patients being members of a committee or board

• patient‐led training or education of healthcare professionals

We will include trials comparing patient‐mediated interventions with common practice/usual care, no patient‐mediated interventions, or other interventions to improve professional practice.

We will exclude interventions that are not solely patient‐mediated and that do not include authentic patients (such as standardised or simulated patients).

Types of outcome measures

Electronic searches

We will search the following electronic databases for primary studies without any language or time limits.

• The Cochrane Central Register of Controlled Trials (CENTRAL), part of the Cochrane Library (www.cochranelibrary.com) and including the Cochrane Effective Practice and Organisation of Care (EPOC) Group Specialised Register

• MEDLINE OvidSP

• Embase OvidSP

As CENTRAL should provide all randomised trials from MEDLINE prior to 2005, we will search MEDLINE from 2005 and onwards.

To test whether or not to search Embase, we will search Embase and MEDLINE for the phrase 'patient mediated' in title and abstract. We will screen all records that are unique to Embase, and will only do a systematic search of Embase if any of these records are eligible for inclusion.

See Appendix 1 for the MEDLINE strategy, which has been peer reviewed using the Peer Review of Electronic Search Strategies (PRESS) checklist (Sampson 2008).

Searching other resources

Selection of studies

Two authors (MSF and TKD) will screen titles and abstracts independently to assess which studies meet the inclusion criteria. We will retrieve full‐text copies of all papers that are potentially relevant, including those where the description of the population, intervention, comparison or outcomes is insufficient in the abstract to make a decision about inclusion. Two authors (MSF and TKD or AF) will independently assess the full‐text copies of the papers for relevance. We will resolve any disagreements by discussion and consensus with a third author (MJ, SAF or HS). We will keep a log of the selection process to complete a PRISMA flow diagram (Moher 2009). We will describe studies that initially appear to meet the inclusion criteria but are later excluded with reasons in the 'Characteristics of excluded studies' table.

Data extraction and management

At least two authors (MSF and TKD, AF or MJ) will independently extract data from each included study using a modified version of the EPOC Data Collection Checklist (EPOC 2013a). We will resolve any disagreements by discussion and by consensus, if needed with a third author. Data from a study that is missing or not clear in a published paper will be marked clearly on the data collection form. Missing data will be sought from the corresponding author of a published paper or principal investigator of an ongoing study.

Assessment of risk of bias in included studies

Two authors (MSF and TKD or AF) will independently assess the risk of bias in accordance with the Cochrane Handbook for Systematic Reviews of Interventions and in line with the EPOC suggested risk of bias criteria (EPOC 2015; Higgins 2011). We will resolve any discrepancies in quality ratings through discussion.

We will assess the risk of bias according to the following domains:

• random sequence generation

• allocation concealment

• blinding of participants and personnel

• blinding of outcome assessment

• incomplete outcome data

• selective outcome reporting

• baseline outcomes measurement

• baseline characteristics

• other bias

We will judge each potential source of bias as high, low, or unclear and provide a quote from the study report together with a justification for our judgement in the 'Risk of bias' table.

Measures of treatment effect

We will include dichotomous or continuous healthcare professional performance outcomes. For dichotomous outcomes, we will analyse data based on the number of events and the number of people assessed in the intervention and comparison groups. We will use these to calculate the risk ratio (RR) and 95% confidence interval (CI). For continuous outcomes, we will analyse data based on the mean, standard deviation (SD) and number of people assessed for both the intervention and comparison groups to calculate mean difference (MD) and 95% CI.

All outcomes reported in the studies will be collected along with how they were measured (self‐report, chart‐abstraction, other objective primary or secondary outcome). For included outcomes, we will extract the intervention effect estimates reported by the investigators of the study along with its confidence interval, and the method of statistical analysis used to calculate it. We will extract data from all time points and categorise them into one of three follow‐up time intervals (0 to 3 months, more than 3 months to 12 months, more than 12 months). If one study reports multiple data within one interval, we will report the data with the longest follow‐up within that interval.

If the same study reports more than one adherence outcome, we will use the primary outcome as defined by the study authors. If a primary outcome is not clearly defined, we will calculate and use the median value from all relevant outcomes.

We will extract data for different socioeconomic groups, if this is reported.

Unit of analysis issues

We may find eligible studies with cluster designs (studies in which the unit of allocation is not a person, but for instance a group of people). Studies in which comparisons are allocated as groups of people (for instance at a clinic) should account for clustering in their analysis. Standard statistical methods assume independence of observation, and for cluster‐design studies the use of these will generally result in artificially small P values and overly narrow confidence intervals for the effect estimates (Ukoumunne 1999), if analysed at the individual level rather than on the cluster level.

We will attempt to re‐analyse studies with potential unit of analysis errors by using the information on the size number of clusters and the value of the intra‐cluster correlation coefficient (ICC). If no ICC is reported, we will utilise the median ICC value from similar studies found in the University of Edinburgh's Database of ICCs (ABDN 2015).

If studies have used an inappropriate statistical method, we will not present confidence intervals or P values unless we are able to re‐analyse the data.

Dealing with missing data

We will attempt to contact study authors in order to verify key study characteristics and to obtain missing numerical outcome data where possible. If this is unsuccessful, we will report the data as missing and will not attempt to impute missing values. The potential impact of the missing data will be explored in the 'Assessment of risk of bias' and 'Discussion' sections of the review. However, if we are able to secure all the missing data we will undertake the analyses on a full intention‐to‐treat basis, meaning we will include in our analyses all participants randomised to each group and analyse data according to this group allocation irrespective of whether or not they received, or adhered to, the intervention (Higgins 2011).

Assessment of heterogeneity

By examining study populations, interventions and outcomes, we will consider if the studies are similar enough to be pooled in a meta‐analysis. We will assess the degree of statistical heterogeneity by visual examination of the scatter of effect estimates on forest plots and by using the Chi² and I² statistics to assess whether observed differences across the studies might be due to chance (Higgins 2003). Where we detect substantial clinical or statistical heterogeneity across included studies, we will use a narrative approach to present data.

Assessment of reporting biases

The tendency for inconclusive results to remain unpublished may impact the findings of a systematic review. We will therefore investigate publication bias when generating pooled effects. We will assess reporting bias qualitatively based on the characteristics of the included studies (e.g. if only small studies that indicate positive findings are identified for inclusion), and if information that we obtain from contacting experts and investigators or studies suggests that there are relevant unpublished studies. If we identify at least 10 studies reporting the same outcome for inclusion in the review, we will construct a funnel plot to investigate small‐study effects, which may indicate the presence of publication bias (Higgins 2011). We will use the test that Egger 1997 proposed for continuous outcomes and that Harbord 2006 proposed for dichotomous outcomes. If we detect asymmetries, we will discuss possible explanations and, in light of those, consider performing sensitivity analyses.

Data synthesis

We will initially group interventions into the six categories listed under Types of interventions, modifying the categories as needed to include all of the patient‐mediated interventions evaluated in the included studies. We will then prepare tables summarising the results of studies for each type of intervention, separating comparisons to usual care from comparisons to other interventions.

We aim at carrying out meta‐analyses to provide an overall effect estimate when two or more comparisons of interventions are sufficiently similar, using Review Manager 5 (RevMan 2014). We will use random‐effects meta‐analysis for combining data, as we anticipate that there may be natural heterogeneity between studies attributable to the variation across similar interventions, populations and implementation strategies. For continuous variables, we will use the inverse‐variance method while for dichotomous variables we will use the method proposed by Mantel‐Haenszel. If cluster‐randomised trials are included or studies report results based on statistical methods that adjust for confounders or design specific variables, we will use the generic inverse variance in RevMan 2014.

If trial authors report the MD without individual group data, we will use this to report the study results.

If we include studies with three or more arms, we need to either exclude or combine one or more arms if we are conducting meta‐analyses with pair‐wise comparisons. Based on our judgment, we will combine groups when appropriate (e.g. if there are two different but still quite similar patient‐mediated interventions, they can be treated as one). We will exclude data from arms deemed irrelevant from the analyses.

Subgroup analysis and investigation of heterogeneity

If the effect estimates vary considerably across studies of similar types of patient‐mediated interventions, we will explore whether the following factors can explain the observed variation:

• direction of change required (increase current behaviour, decrease current behaviour, mix, or unclear). Hypothesis: effect on increasing a behaviour is larger than on decreasing behaviour

• recipient (physician; other healthcare professionals). Hypothesis: clinical practice is more difficult to change among physicians than among non‐physicians

• risk of bias (high; unclear; low). Hypothesis: effect sizes are smaller when risk of bias is low

• baseline clinical performance (continuous measure of healthcare professionals' compliance with recommended practice or clinical guidelines). Hypothesis: when baseline clinical performance is low, effect sizes are larger

Sensitivity analysis

If appropriate, we will conduct sensitivity analyses where we:

• exclude data from studies assessed as being at high or unclear risk of bias

• use different ICCs to re‐analyse data from cluster‐RCTs