Evidence-based policymaking when evidence is incomplete: The case of HIV programme integration

• Sustainable Development Goal 3 aims to "ensure healthy lives and promote well-being for all at all ages" and has set a target of achieving global universal health coverage, representing a major policy shift away from mostly disease-specific "vertical programmes". • While health service integration can be a promising strategy to improve healthcare coverage, health outcomes, and efficiency, the exact impact of integration in different settings is hard to predict, and policy makers need to choose from a large variety of integration strategies and opportunities with varying levels of scientific evidence. • Using the case of health service integration for HIV in low- and middle-income countries, we outline implementation strategies for integration opportunities with lacking or scarce high-level causal evidence, based on existing frameworks and methodologies from within and beyond healthcare and implementation science. • Proper use of scientific evidence in other contexts requires adequate and systematic assessments of the transportability of an intervention. Several methods exist that allow for judging transferability and comprehensively identifying key context-specific indicators across studies that can affect the reported impact of interventions. • When (transferable) evidence is absent, we propose that by drawing on well-established design and implementation methodologies - underpinned by ongoing learning and iterative improvement of local service delivery strategies - countries could substantially improve decision-making even in the absence of scientific evidence. • Reaching the goal of AU: Anabbreviationlisthasbeencompiledforthoseusedthroughoutthetext:Pleaseverifythatallentriesarecorrectmaking the HIV response an integral part of a larger, universal, people- centred health system that meets the needs and requirements of citizens can be facilitated by applying lessons learned from implementation science and novel design methodologies.