Community and health systems learning: critical realist evaluation of the VAPAR ‘learning platform’ in rural South Africa 2015-25 [version 3; peer review: 3 approved, 1 approved with reservations, 1 not approved]

Abstract

Background Learning platforms can strengthen primary healthcare (PHC) by integrating community knowledge with system decision-making, but evidence on how they work in low-resource settings is limited. This study presents a realist evaluation of the Verbal Autopsy with Participatory Action Research (VAPAR) learning platform in rural Mpumalanga, South Africa (2015–25). VAPAR aimed to embed participatory evidence generation and shared learning within routine district processes to support more equitable, community-linked PHC. Methods A realist design was used to synthesise data from five action-learning cycles (2017–23), a preceding pilot (2015–16), and an engagement and uptake phase (2023–25). Data included cycle reports, participatory outputs, verbal autopsy (VA) analyses, 22 endline interviews, policy, strategy and planning documents. Using a co-developed theory of change, qualitative data were coded to examine context-mechanism-outcome patterns. Mechanisms were identified and refined through cross-cycle comparison, triangulation, and stakeholder validation. Results VAPAR was contextually responsive, adapting to shocks such as COVID-19 and progressively embedding within the district health system. Through regular dialogue, the platform activated generative mechanisms of trust-building, role clarity and recognition, collective sense-making, and strengthened agency, particularly among Community Health Workers (CHWs), whose skills, confidence and legitimacy expanded. These mechanisms operated within an enabling structural context shaped by PHC reforms that strengthened the District Health System and Ward-Based Primary Health Care Outreach Teams, alongside trade-union action for CHW absorption into public service. Institutionalisation followed through Mpumalanga’s revitalised Health Promotion Programme, with adaptation to additional provinces and for outbreak response and emergency obstetric care. Outcomes were interpreted through context-mechanism-outcome patterns, illustrating how participatory learning becomes embedded in decentralised health systems. Conclusions Over a decade, VAPAR demonstrated how structured, participatory learning can reshape relationships, strengthen community-linked PHC, and support institutionalisation of routine, evidence-informed practice in decentralised health systems. The findings offer transferable lessons for sustaining learning platforms in resource-constrained settings.