The systematic review will be performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (11, 12). The protocol was registered in the International Prospective Register of Systematic Reviews (number CRD42020167674) (13).
A comprehensive literature search of published and unpublished studies (“grey literature”) in electronic databases will be conducted. For unpublished studies we will contact partners (e.g. UK government departments, agencies and public bodies such as Public Health England; the Behavioural Insights Team). The search will be conducted in five electronic databases: Medline, EMBASE.com, PsycINFO, Cochrane Library and CINAHL. We will conduct supplementary searches in Google Scholar, hand search relevant journals, and conduct backward and forward citation searching of included studies and relevant reviews.
A search strategy (Additional file 1) following PICOS will be adapted (population, intervention, comparison, outcome and study design), including Medical Subject Headings (MeSH) terms and relevant key words (14, 15). The PICOS process facilitates a more evidence-based approach to literature searching and helps to rapidly and accurately locate the best available scientific information, avoiding unnecessary searching (15).
Eligibility criteria
All studies written in English language published until March 2020 will be considered. The specific inclusion criteria of the study will be the following:
- Types of studies to be included: We will include published and unpublished (grey literature) RCTs, which excludes non-randomized study designs with a predefined control group or comparator group (excluding also before-and-after studies and ‘natural experiments’). Due to our close relationships with the Department of Health and Social Care (DHSC) and behavioural science units across Europe as a Policy Research Unit, we are in a strong position to identify additional grey literature not (yet) published in peer reviewed journals.
- Population: We will include HCPs who are involved in providing direct patient care (this includes HCPs in training). No restrictions were placed on geographical region.
- Interventions: We will be looking for interventions focused on letters sent from healthcare organisations to HCPs with the intent to alter their professional practice. We will also consider emails, but classifying them separately in our analysis (including interventions in which the letter is the content of an email or an attachment to an email).
- Comparisons: Relevant comparisons will include no intervention control arms, non-personalised information such as circulars or newsletters, letters with different format or content or letters with additional materials (such as leaflets) or reminders (by phone, text or letter).
- Primary Outcomes: Measures of health professional clinical behaviour such as rates of performing prevention, diagnosis, and treatment behaviours (e.g. immunisation, blood pressure measurement, prescription, referral); measures of health professional non-clinical behaviour such as rates of performing specified non-clinical behaviours (for example achieve recruitment targets for clinical trials). No secondary outcomes will be included.
- Study designs: We will focus on RCT designs.
Patient and public involvement (PPI)
PPI involvement has the ability to empower people with respect to health and social care services, influencing change and improvement upon those people the services most concern (16). A core component of PRU (Policy Research Unit) policy is to maximise public involvement in all our research activities (17). Thus, the PRU PPI panel has commented on the protocol and will receive regular updates on the review and comment on outputs.
Selection of studies
All titles and abstracts retrieved by electronic searching will be downloaded to the reference management database End Note and uploaded to Covidence, one of Cochrane’s recommended tools. This web-based software platform has been designed to support more efficient management of systematic reviews and can be used from the beginning of title/abstract screening through the beginning of meta-analysis (18). Duplicates will be removed and two review authors independently will examine the remaining references. The abstracts will be included if they meet the inclusion criteria. Subsequently, two reviewers will independently determine the eligibility of studies on the basis of a review of the full texts. Differences in judgment will be resolved through discussion and inclusion of a third rater. The selection process will be recorded and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram will be completed (11).
Data extraction
The process of data extraction will involve one or two reviewers who generate a data extraction form. The data extraction form, which will be reviewed and refined by the reviewers, will include variables as follows:
- characteristics of the receiver
- characteristics of the sender
- intervention features including
- Template for Intervention Description and Replication (TIDieR) checklist: (brief name, why, what (materials), what (procedure), who provided, how, where, when and how much, tailoring, modifications, how well (planned), how well (actual) (19).
- MINDSPACE checklist of behavioural economic techniques: the nine most robust effects on behaviour (Messenger, Incentives, Norms, Defaults, Salience, Priming, Affect, Commitment and Ego) (9, 20).
- BCTs (observable and replicable components designed to change behaviour) (7, 21), intervention functions (education, persuasion, incentivisation, coercion, training, restriction, environmental restructuring, modelling, enablement) and policy categories (communication/ marketing, guidelines, fiscal, regulation, legislation, environmental/ social planning, service provision), which are all part of the Behaviour Change Wheel framework (7, 22).
- Primary outcome(s): Measures of health professional clinical and non-clinical behaviour. Where more than one reported outcome is provided, we will use the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) approach to assess the certainty of evidence (separated into those that are critical, important and not important) (23). Outcomes measured at multiple time points will be categorized as following: immediate (within 2 weeks of the intervention delivery), short-term (2–13 weeks after intervention delivery), medium-term (14–50 weeks after intervention delivery) and long-term effects (51 or more weeks after intervention delivery). We will present multiple time points only for critical outcomes.
Quality assessment
An assessment of the methodological quality of included studies will be conducted using the Cochrane Collaboration Risk of Bias Tool (CCRBT) (24). CCRBT is a two-part tool, addressing seven evidence-based domains, namely random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective outcome reporting (reporting bias) and other sources of bias (other bias). The first part of the tool describes a sufficiently detailed support for judgment about the risk of bias, ensuring its transparency. The second part assigns a judgment relating to the risk of bias for each domain. This is achieved by assigning a judgment of ‘Low risk’ of bias (+), ‘High risk’ of bias (-), or ‘Unclear risk’ of bias (?). According to the Cochrane Collaboration’s recommendations, the studies, whose all domains will be rated positively, will be judged as low risk of bias, while the studies with one or more unclear domains will be judged as having unclear risk of bias. Furthermore, the studies, where one or more domains will be rated negatively, will be judged as high risk of bias(25, 26). Two review authors will independently evaluate the methodological quality of each included study using the assessment tools. Discrepancies will be resolved through a consensus procedure.
Data analysis and synthesis
We will compare the reported intervention features against a control condition which does not contain those features. This will provide information useful for explaining why interventions were effective or ineffective. The analyses of those comparisons will be conducted and reported separately according to the participants’ characteristics and characteristics of the sender.
BCTs will be double-coded by trained coders using the Behaviour Change Techniques Taxonomy v1 (7) using a BCT extraction form. BCTs will be coded separately for intervention and control groups. The reliability of coding of BCTs, MINDSPACE checklist, and TiDIER checklist (namely the presence or absence of codes within each intervention) will be assessed using the prevalence and bias-adjusted kappa (PABAK) statistic (27). PABAK was used because adjusts for shared bias in the coders’ use of categories and high prevalence of negative agreement (i.e. when both coders agree that codes are absent). A random-effects meta-analysis with the extracted BCTs will be conducted, if the data permits. The meta-analysis will be conducted when a group of studies will be sufficiently homogeneous in terms of subjects involved, interventions, and outcomes to provide a meaningful summary. Although a meta-analysis can be conducted with minimum 2 studies, Valentine, Pigott (28) suggest that the combination of very few studies with heterogeneous characteristics makes any kind of synthesis untenable in most cases, while parameter estimation (e.g., the random effects variance component) will likely be poor, rendering conclusions that are highly uncertain.