Analysis and synthesis

We anticipated substantial heterogeneity in study design and intervention [29]; meta-analysis was therefore not appropriate and a narrative synthesis was undertaken. Asthma and diabetes papers were analysed separately. We classified components of the interventions (e.g. whether the financial incentive was paid to the individual (self-employed) healthcare professional or to an organisation); whether payment was for achieving process standards (e.g. attendance at a diabetes course) or health outcomes (e.g. reduced unscheduled care). A matrix of interventions was developed with the interventions being shown to be effective or ineffective under the headings of: “organisational process”; “measure of disease control” and “individual behaviour”. We used Adams’ 2013 framework, which has been specifically designed for documenting financial incentive interventions [30]. The framework contains nine domains which we used to identify the features and describe the schemes in detail. These domains are: direction (positive reward or avoidance of penalty), form (cash or healthcare costs), magnitude (total value of incentive available to participant), certainty (certainty of receiving payment if behaviour is successfully changed: certain, certain chance or uncertain chance), target behaviour (process, intermediate or outcome), frequency of reward (all or some instances incentivised), immediacy (time between behaviour and payment), schedule (fixed or variable), and recipient(s) of incentives (clinicians).

We synthesised our results in the form of Harvest Plots, as they are a useful method for illustrating the different effects of interventions because they represent all relevant data in one plot [31]. In a Harvest plot, each bar represents an individual study, the bar colour indicates the study design, the bar height reflects the number of participants in the study and the number reflects the Downs and Black quality score.