Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Protocol - Intervention

Computerised decision support systems to promote appropriate use of blood products

Authors' declarations of interest

Version published: 03 February 2017 Version history

https://doi.org/10.1002/14651858.CD012545
Collapse all Expand all

Abstract

This is a protocol for a Cochrane Review (Intervention). The objectives are as follows:

To assess the effect of computerised decision support systems (DSSs) on transfusion practice.

Background

Description of the condition

Blood transfusions have long been associated with significant short‐ and long‐term risks in patients (Bolton‐Maggs 2012; Bolton‐Maggs 2014; Vamvakas 2009). One approach to minimise the use of blood involves applying a restrictive transfusion policy, whereby blood transfusions are given only when the potential benefits are deemed to outweigh the potential risks.

A Cochrane review compared restrictive versus liberal red blood cell (RBC) transfusion strategies and found reduced in‐hospital mortality associated with a restrictive transfusion policy (Carson 2012a). Recent systematic reviews and meta‐analyses have also shown reduced cardiac events, re‐bleeding, bacterial infections and mortality, and healthcare‐associated infection, in individuals treated according to a restrictive compared with liberal transfusion strategy (Holst 2015; Rohde 2014; Salpeter 2014. A more liberal strategy is recommended for people with cardiovascular disease or acute coronary syndrome (Docherty 2016; NICE 2015).

Cochrane systematic reviews that compared a lower (10 x 109/L) versus a higher (20 to 30 x 109/L) platelet count threshold (Estcourt 2015a), or low dose versus high dose platelet transfusions (Estcourt 2015b), showed no evidence of a difference for bleeding outcomes and a significant reduction in platelet component usage for the more restrictive transfusion policies.

Current guidelines recommend these restrictive policies (Carson 2012b; Kaufman 2015; NICE 2015). However, national audits of blood use in the UK have consistently shown that around 20% of blood product usage is outside of guideline recommendations, which results in risks to patients and unnecessary costs (Estcourt 2012; NCABT 2011; NCABT 2015).

Description of the intervention

A decision support system (DSS) is defined as "any software designed to directly aid in clinical decision‐making in which characteristics of individual patients are matched to a computerised knowledge base for the purpose of generating patient‐specific assessments or recommendations that are then presented to clinicians for consideration" (Hunt 1998). Computerised DSSs have been introduced across a range of healthcare settings as an aid to both clinicians and patients in making healthcare decisions based on individualised patient characteristics (Cresswell 2012). Two large systematic reviews assessed the effect of DSSs across a range of clinical fields and demonstrated that the introduction of a DSS improved clinical practice in around two‐thirds of included studies (Garg 2005; Kawamoto 2005).

A recent systematic review of DSS applied to transfusion practice found that implementation of a DSS improves RBC usage in many studies (Hibbs 2015). DSSs have been shown to be effective in intensive care unit (ICU) (Pentti 2003), cardiothoracic surgery (Razavi 2014), haematology settings (Butler 2015), and in paediatrics (Adams 2011; Baer 2011; McCrory 2014). DSSs can improve practice in recipients of red cells (Kassakian 2016), platelets (Butler 2015; Collins 2015; Pentti 2003; Lin 2010), and plasma (Pentti 2003). However, most of these studies were uncontrolled before‐and‐after studies that relied on historical controls as their comparator group, which led to potential bias in their conclusions (Hibbs 2015).

How the intervention might work

Blood transfusion guidelines state that clinicians must take into account individual patient variables before implementing transfusion. One method of implementing a restrictive transfusion strategy is through the use of computerised DSSs, incorporating individual patient clinical characteristics and laboratory values, primarily data from blood counts, in order to promote appropriate use of blood transfusion and improve patient outcome.

DSSs can be used as a stand‐alone resource (e.g. as a piece of computer software, internet resource, or smart phone application) or can be integrated into a computerised blood component request system. For example, a DSS could require the requester to choose the reason for the transfusion from a list of common transfusion indications. Based on the selected indication and the individual's most recent laboratory test results, the DSS could provide advice on whether the request is within the recommended guidelines and also provide an alert or recommendation for modifying the order if it is inappropriate.

Why it is important to do this review

Unnecessary transfusions not only expose individuals to the risk of serious adverse transfusion reactions and transfusion‐transmitted infections, but also reduce the availability of blood products for other people who need them (WHO 2015). The World Health Organization (WHO) recommends an integrated strategy for ensuring blood safety and availability. This includes the "rational use of blood and blood products to reduce unnecessary transfusions and minimize the risks associated with transfusion, the use of alternatives to transfusion, where possible, and safe and good clinical transfusion practices, including patient blood management" (WHO 2015).

Other systematic reviews have evaluated the effectiveness of interventions (such as educational initiatives, clinician feedback, and audits) intended to improve compliance with local transfusion guidelines and promote appropriate blood usage (Tinmouth 2005; Wilson 2002).

This Cochrane review will assess the use of DSSs to change the behaviour of clinical staff and reduce the number of unnecessary transfusions. Although most DSSs integrated into a computerised blood component request system are only likely to be relevant to high‐income countries (Hibbs 2015), stand‐alone DSSs on smart phones may be accessible to a much wider proportion of the world's population. In fact, smart phone ownership rates have risen dramatically in low‐ and middle‐income countries from a median of 21% in 2013 to 37% in 2015 (Pew Research Center 2016).

Objectives

To assess the effect of computerised decision support systems (DSSs) on transfusion practice.

Methods

Criteria for considering studies for this review

Types of studies

As a review of the effects of a health system strategy, we expect most studies to be non‐randomised, or to have compared outcomes before and after the introduction of a DSS. We will therefore follow the suggestions of the Cochrane Effective Practice and Organisation of Care (EPOC) Group (Cochrane EPOC 2013a), and will include the following.

  • Randomised controlled trials (RCTs)

  • Non‐randomised controlled trials (NRCTs)

  • Controlled before‐and‐after (CBA) studies

  • Interrupted time series (ITS) and repeated measures studies with a clearly defined time when the intervention occurred and at least three data points before and three after the intervention.

We will exclude uncontrolled studies, cross‐sectional studies, and case‐control studies.

We will exclude cluster‐RCTs, non‐randomised cluster trials, and CBAs with fewer than two intervention sites and two control sites. In studies with only one intervention or control site, the intervention (or comparison) is completely confounded by study site which makes it difficult to attribute any observed differences to the intervention rather than to other site‐specific variables.

If there are sufficient data to answer this review's questions using only data from RCTs we will only report data from RCTs.

Types of participants

We will include all people (adults and children) who are considered for transfusion of red blood cells (RBCs), platelets, plasma, cryoprecipitate, or granulocytes in any clinical setting.

We will exclude people who receive other blood products e.g. intravenous immunoglobulin, factor VIII.

Types of interventions

Any electronic/computerised DSS that provides clinicians with recommendations on RBC, platelet, plasma, cryoprecipitate, or granulocyte ordering at the time the decision to order a transfusion is being made based on individual patient characteristics.

The comparator in the control group (for controlled studies) or prior to introduction of the intervention (for ITS studies) will be no DSS.

Types of outcome measures

The primary and secondary outcomes of this review are outcomes of interest and we will not use them as inclusion criteria for the assessment of studies.

We will categorise all outcomes according to short‐, medium‐, and long‐term outcomes. We will report the exact definition of these time frames over time periods that are common to as many studies as possible (e.g., up to three months, three to 12 months, and greater than 12 months from the start of the study). Included studies will typically involve a change in service for all people treated at a particular hospital with the aim of improving practice of the hospital as a whole; thus patient‐reported quality of life outcomes are unlikely to be measured and we will not assess them in this review.

Primary outcomes

  • Proportion of participants who receive transfusions

  • Amount of blood product used per participant (number of units in adults and volume in mL in infants and children)

  • Serious adverse events

    • Transfusion‐related (including transfusion‐related acute lung injury (TRALI), transfusion‐transmitted infection, transfusion‐associated circulatory overload (TACO), transfusion‐associated dyspnoea (TAD), acute transfusion reactions)

    • Bleeding (including World Health Organization (WHO) grade 3 or 4, or equivalent, or bleeding that requires an operation)

    • Infection

    • Arterial or venous thromboembolism (including deep vein thrombosis; pulmonary embolism; stroke; myocardial infarction)

Secondary outcomes

  • Number of transfusions compliant with institutional transfusion guidelines

  • Blood count or coagulation parameter (e.g. haematocrit, haemoglobin, prothrombin time, partial thromboplastin time, or platelet count) preceding and after the transfusion

  • Length of participant stay (in‐hospital)

  • Length of participant stay (intensive care unit (ICU))

  • All‐cause mortality

  • Clinician workflow (additional time per intervention implemented)

Search methods for identification of studies

The Systematic Review Initiative’s Information Specialist (CD) will formulate the search strategies in collaboration with the Cochrane Haematological Malignancies Group.

Electronic searches

We will search the following electronic databases for eligible studies from 1980 to present.

  • Cochrane Central Register of Controlled Trials (CENTRAL, the Cochrane Library, current issue) (www.cochranelibrary.com/)

  • MEDLINE (OvidSP, Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE Daily and Ovid MEDLINE, 1946 onwards)

  • PubMed (Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, for recent records not yet added to MEDLINE) (www.ncbi.nlm.nih.gov/sites/entrez)

  • Embase (OvidSP, 1974 onwards)

  • CINAHL (EBSCOHost, 1937 onwards)

  • Proquest Dissertations and Thesis Global (ProQuest, 1861 onwards)

  • Transfusion Evidence Library (1950 onwards) (www.transfusionevidencelibrary.com)

  • Web of Science Science & Social Sciences Conference Proceedings Indexes (CPSI‐S & CPSSI, 1990 onwards)

We will search the following databases for ongoing trials.

We will apply the Cochrane RCT search filters (Lefebvre 2011) and the SIGN systematic review and observational studies filters to the MEDLINE and Embase search strategies; and the SIGN RCT, systematic review and observational studies filters (www.sign.ac.uk/methodology/filters.html) to the searches in CINAHL. We will limit searches to 1980 onwards. We will not apply any restrictions on language or publication status. The search strategies can be found in Appendix 1.

Once we identify studies for inclusion, we will search MEDLINE (Ovid) for errata or retraction statements for the reports of these studies.

Searching other resources

We will handsearch reference lists of included studies in order to identify further relevant studies. We will contact the lead authors of the included studies to identify any unpublished material, missing data, or information regarding ongoing studies.

Data collection and analysis

Selection of studies

We will select studies for inclusion according to Chapter 7 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). The Systematic Review Initiative’s Information Specialist (CD) will initially screen all search hits for relevance against the eligibility criteria and discard all those that are clearly irrelevant. Thereafter, two review authors (SF, AD) will independently screen all the remaining references for relevance against the full eligibility criteria.

We will retrieve full‐text papers of all references for which we cannot decide on eligibility based on the title and abstract alone. We will assess study design features against the inclusion criteria. We will request additional information from study authors, as necessary, to assess the eligibility for inclusion of individual studies. The two review authors will discuss the results of study selection and try to resolve any discrepancies between themselves. In the event that this is not possible, they will consult a third review author (LE). We will report the results of the study selection process using a PRISMA flow diagram (Moher 2009).

We will record the reasons for exclusion of studies after full‐text assessment and will add those to the 'Characteristics of excluded studies' table.

We will collate multiple reports of one study so that the study, and not the report, is the unit of analysis.

Data extraction and management

As recommended in the Cochrane Handbook for Systematic Reviews of Interventions, two review authors (SF, AD) will independently extract data onto standardised forms and perform a cross‐check (Higgins 2011a). We will pilot the data extraction form on two included studies (if available, one controlled study and one ITS study). The review authors will come to a consensus on the required changes. If they cannot reach an agreement, they will consult a third author (LE). The review authors will not be blinded to the names of authors, institutions, journals, or the study outcomes. We will extract the following information from each included study.

  • Source: study ID; report ID; review author ID; date of extraction; ID of author checking extracted data; citation of paper; contact authors details

  • General study information: publication type; study objectives; funding source; conflict of interest declared; other relevant study publication reviewed

  • Study details: location; country; setting; number of centres; inclusion and exclusion criteria; total study duration; recruitment dates; defined primary and secondary outcomes; follow‐up time points; power calculations

  • Study methodology: design (RCT, NRCT, CBA, ITS), primary analysis (and definition); stopping rules; method of sequence generation and allocation concealment (controlled studies), blinding (of clinicians, participants, and outcome assessors) (controlled studies); any concerns regarding bias

  • Characteristics of interventions: description of experimental arms with defined criteria for transfusion (e.g. laboratory values compared with institutional threshold, bleeding status of patient, presence of cardiac ischaemia or early septic shock); time of intervention (CBA and ITS studies); cost of intervention

  • Characteristics of participants: age; gender; primary diagnosis etc.

  • Participant flow: total number of participants screened for inclusion; total number recruited; total number excluded; total number allocated to each group (controlled studies); total number analysed (for review outcomes); number of allocated patients who received planned treatment; number of drop‐outs with reasons (percentage in each arm); protocol violations; missing data

  • Outcomes: proportion of patients receiving transfusions; amount of blood product used per participant; serious adverse events (transfusion‐related, bleeding, infection, arterial, or venous thromboembolism); number of transfusions compliant with institutional transfusion guidelines; mean blood count/coagulation parameter preceding transfusion; length of patient stay (in‐hospital; ICU); number of deaths, and clinician workflow (additional time per intervention implemented)

  • For interventional cohort and pre‐post single arm or multiple arms studies we will also collect data if available on: confounding factors, the comparability of groups on confounding factors; methods used to control for confounding and on multiple effect estimates (both unadjusted and adjusted estimates) as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Reeves 2011)

  • For time series data reported graphically, we will extract crude data values for each time point from graphs by using PlotDigitizer (PlotDigitizer 2015).

Assessment of risk of bias in included studies

RCTs

We will assess the risk of bias for all included RCTs using the Cochrane 'Risk of bias' tool according to chapter eight of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). Two review authors (SF, AD) will independently assess each element of potential bias listed below as 'high', 'low', or 'unclear' risk of bias. We will provide a brief description of the judgement statements upon which the authors assess potential bias in the 'Characteristics of included studies' table. We will reach a consensus on the degree of risk of bias through comparison of the review authors statements and, where necessary, by consulting a third review author (LE). We will use the 'Risk of bias' assessment to explore statistical heterogeneity in each included study and to perform sensitivity analyses. We will use the Cochrane 'Risk of bias' assessment tool, which includes following domains.

Selection bias

We will describe for each included study if and how the allocation sequence was generated and if allocation was adequately concealed prior to assignment. We will also describe the method used to conceal the allocation sequence in detail and determine if intervention allocation could have been foreseen in advance of, or during recruitment, or changed after assignment.

Performance bias

We will describe for each included study, where possible, if the study participants and personnel were adequately blinded from knowledge of which intervention a participant received. We will judge studies as low risk of bias if they were blinded, or if we judge that lack of blinding could not have affected the results.

Detection bias

Was blinding of the outcome assessors effective in preventing systematic differences in the way in which the outcomes were determined?

Attrition bias

We will describe for each included study the attrition bias due to amount, nature, or handling of incomplete outcome data. We will also try to evaluate whether intention‐to‐treat analysis has been performed or could be performed from published information.

Reporting bias:

We will describe for each included study the possibility of selective outcome reporting bias.

Other issues

Was the study apparently free of other problems that could put it at risk of bias?

We will summarise the risk of bias for each key outcome for each included study. We will judge studies with at least one domain of high risk to be at high risk of bias overall.

Non‐randomised studies

We will use ROBINS‐I tool (formerly known as ACROBAT‐NRSI) to rate the quality of non‐randomised controlled trials (non‐RCTs), CBAs, and ITS studies (Sterne 2014). This tool is based on the Cochrane 'Risk of bias' tool for rating the quality of RCTs (Higgins 2011c). The tool covers seven domains and the quality of evidence is rated as either low, moderate, serious, critical, or no information (see Appendix 2 for a copy of the tool), and uses signalling questions for the assessment of the following.

  • Bias due to confounding

  • Bias in the selection of participants

  • Bias in measurement of interventions

  • Bias due to departure from intended interventions

  • Bias due to missing data

  • Bias in measurement of outcomes

  • Bias in the selection of the reported result

We will resolve disagreements on the assessment of quality of an included trial by discussion until we reach consensus or, failing that, we will consult a third review author (LE).

We have pre‐specified the main potential confounding factors.

  • Primary diagnosis of participant (e.g. liver disease; critical illness; pregnancy)

  • Age: variability in the age of participants included, e.g. infant (zero to one year); paediatric (one year to 16 years) versus adult (greater than 16 years) versus older adult (greater than 60 years)

  • Gender: male to female ratio

  • Medications: use of anticoagulants or anti‐platelet agents

Measures of treatment effect

RCTs

For continuous outcomes we will extract and report the mean or mean change from baseline, standard deviation (SD), and total number of participants in both the treatment and control groups. For dichotomous outcomes we will record the number of events and the total number of participants in both the treatment and control groups.

For continuous outcomes that use the same scale, we will perform analyses using the mean difference (MD) with 95% confidence intervals (CIs). If continuous outcomes are reported using different scales we will use standardised mean difference.

If available, we will extract and report hazard ratios (HRs) for time‐to‐event‐data (mortality or time in hospital) data. If HRs are not available, we will estimate as accurately as possible the HR by using the available data and a purpose built method based on the approaches of Parmar 1998 and Tierney 2007. If sufficient studies provide HRs, we will use HRs in favour of risk ratios (RRs) or MDs in a meta‐analysis, but for completeness we will also perform a separate meta‐analysis of data from studies providing only RRs or MDs for the same outcome.

For dichotomous outcomes we will extract and report the risk ratio (RR) with a 95% CI (Deeks 2011). Where the number of observed events is small (less than 5% of sample per group), and where trials have balanced treatment groups, we will report the Peto’s Odds Ratio (OR) with 95% CI (Deeks 2011).

If data allow, we will undertake quantitative assessments using Review Manager 5 (RevMan 5) (Review Manager 5).

Non‐randomised studies with a control group (including CBAs)

For continuous outcomes we will record the mean, SD, and total number of participants in both the treatment and control groups. For dichotomous outcomes we will record the number of events and the total number of participants in both the treatment and control groups.

For continuous variables we will extract and report the absolute change from a statistical analysis adjusting for baseline differences (such as regression models, mixed models, or hierarchical models) or the relative change adjusted for baseline differences in the outcome measures (i.e. the absolute post‐intervention difference between the intervention and control groups, as well as the absolute pre‐intervention difference between the intervention and control groups/the post‐intervention level in the control group) (Cochrane EPOC 2015).

For dichotomous outcomes we will extract and report the RR with a 95% CI from statistical analyses adjusting for baseline differences (such as Poisson regressions or logistic regressions) or the ratio of RRs (i.e. the RR post‐intervention/RR pre‐intervention).

If data allow, we will undertake quantitative assessments using RevMan 5 (Review Manager 5).

ITS studies

For ITS studies, in order to obtain comparable effect size measures, we will re‐analyse extracted data using segmented time‐series regression analysis according to recommended methods for ITS designs as described (Cochrane EPOC 2013b), to obtain two standardised effect sizes for each study: i) change in level (the difference between the observed level at the first intervention time point and that predicted by the pre‐intervention time trend) and ii) change in slope (the difference between post‐ and pre‐intervention slopes) of the regression lines before and after the intervention (Ramsay 2003).

All studies

Where appropriate, we will report the number needed to treat for an additional beneficial outcome (NNTB) and the number needed to treat for an additional harmful outcome (NNTH) with 95% CIs.

If we cannot report the available data in any of the formats described above, we will perform a narrative report and, if appropriate, we will present the data in tables.

Unit of analysis issues

We do not expect to encounter unit of analysis issues as we are unlikely to include cluster‐randomised trials, cross‐over studies, and multiple observations for the same outcome in this Cochrane review. Should any studies of these designs arise, we will treat these in accordance with the advice given in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011c).

Dealing with missing data

Where we identify that data are missing or unclear in the published literature, we will contact the study authors directly. We will record the number of participants lost to follow‐up for each included study. Where possible, we will analyse the data on an intention‐to‐treat (ITT) basis, but if insufficient data are available, we will present per protocol (PP) analyses (Higgins 2011c).

Assessment of heterogeneity

If the clinical and methodological characteristics of individual studies are sufficiently homogeneous, we will combine the data to perform a meta‐analysis. We will assess statistical heterogeneity of treatment effects between studies using a Chi2 test with a significance level of P < 0.1. We will use the I2 statistic to quantify the degree of potential heterogeneity and classify it as moderate if the I2 statistic value is greater than 50% or considerable if the I2 statistic value is greater than 80%. We anticipate that we will identify at least moderate clinical and methodological heterogeneity within the included studies; therefore we will use a random‐effects model. If we identify a cause for the heterogeneity, we will explore this with by sensitivity and subgroup analyses (Deeks 2011). If statistical heterogeneity is considerable and we cannot find a cause for the heterogeneity, we will not perform a meta‐analysis but will comment on the results narratively and present the results from all studies in tables.

Assessment of reporting biases

Where at least ten studies are identified for inclusion in a meta‐analysis, we will explore potential publication bias (small trial bias) by generating a funnel plot and using a linear regression test. We will consider a P value of less than 0.1 as significant for this test (Lau 2006; Sterne 2011).

Data synthesis

If studies are sufficiently homogenous in their study design, we will conduct a meta‐analysis according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). For statistical analysis, we will enter data into the Cochrane statistical package RevMan 5 (Review Manager 5). One review author (SF) will enter the data into the software. A second review author (AD) will then check for accuracy. We will conduct separate analyses for each study design (RCTs, NRCTs, CBA studies, ITS studies). We will not conduct meta‐analyses that include both RCTs and non‐RCTs.

For RCTs where meta‐analysis is feasible, we will use the random‐effects model to pool data. For binary outcomes we will base the estimation of the between‐study variance on the Mantel‐Haenszel estimator. We will use the inverse‐variance method for continuous outcomes, outcomes that include data from cluster‐RCTs, or outcomes where HRs are available. If we find that the heterogeneity is above 80% and we identify a cause for the heterogeneity, we will explore this with subgroup analyses. If we cannot find a cause for the heterogeneity then we will not perform a meta‐analysis, but we will comment on the results narratively and will present the results from all studies in tables.

If meta‐analysis is feasible for non‐RCTs or CBA studies, we will analyse these separately. We will only analyse outcomes with adjusted effect estimates if these are adjusted for the same factors using the inverse‐variance method as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Reeves 2011).

If meta‐analysis is feasible for ITS studies, we will use the effect sizes if reported in the included studies or are obtained (as described above) and pool them using the generic inverse variance method in RevMan 5 (Review Manager 5).

Where data do not allow quantitative assessment (either through insufficient studies of similar design or due to considerable heterogeneity), we will present outcome data individually per study and comment on any trends in the data.

We will document and summarise the reporting of clinician workflow but we will not perform any formal analysis of these outcomes.

'Summary of findings' table

We will use the GRADE system to build a 'Summary of findings' table using GRADEpro GDT 2014 software, as outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2011a; Schünemann 2011b). We have listed the outcomes that we will include when we compare DSS with no DSS.

  • Proportion of participants who receive transfusions

  • Total number of units of blood product used

  • Mean dose of blood product received per transfused participant

  • All‐cause mortality

  • Length of participant stay (in‐hospital)

  • Length of participant stay (ICU)

Subgroup analysis and investigation of heterogeneity

If adequate data are available, we will perform subgroup analyses according to each of the following types of DSS.

  • Mandatory or non‐mandatory DSS

  • Single laboratory value‐based DSS or DSS incorporating patient clinical characteristics

  • DSS trigger with recommendation on amount to transfuse or DSS with trigger alone

  • DSS with ongoing additional interventions (e.g. education, training, other complex intervention) or DSS only

Sensitivity analysis

We will assess the robustness of our findings by performing the following sensitivity analyses where appropriate.

  • We will only include studies with a ‘low risk of bias’ (e.g. RCTs with methods assessed as low risk for random sequence generation and concealment of treatment allocation)

  • We will only include studies with less than a 20% dropout rate