Regarding topic areas, future research should focus on: • Replicating studies of promising strategies tested with few studies (e.g., audit with in-person feedback plus peer review) • Head-to-head comparisons of key supervision strategies (e.g., routine supervision versus audit with feedback), strategy combinations (e.g., audit with feedback plus peer review versus audit with feedback alone) and supervision attributes (e.g., different supervision methods, such as involving supervisors in group problem-solving with HCPs, supervision of supervisors, and frequencies). Understanding the optimal frequency or dose of supervision in different contexts is an especially critical topic • Rigorous studies of supervision strategies to improve the practices of lay or community health workers • Better quantitative and qualitative understanding of how context influences strategy effectiveness Regarding methods, future research should: • Use standardized methods, especially for outcomes, strategy description, implementation (including dose and fidelity), and characterization of study context • Prioritize head-to-head studies, which provide stronger evidence for comparing different supervision approaches • Have rigorous study designs, such as interrupted time series with a randomized comparison group, which reduce bias and show how effectiveness changes over time • Have follow-up periods that match the timeframe that programs require for improvements to be meaningful (e.g., at least 12 months) and include multiple measures of effect so changes (reductions or further improvements) in effectiveness over time can be quantified • Include assessments of strategy cost and cost-effectiveness • Be designed to better contribute to filling gaps in the evidence base about strategy choice and combinations of componentsa

^aStudies directly comparing two supervision approaches without other components are the easiest to interpret. However, given the generally moderate effect of supervision as a sole strategy, studies should include other enhancing components in both study arms (e.g., supervision approach A + training versus supervision approach B + training)