Advancing digital health equity through implementation science: A multi-method evaluation of drivers of implementation outcomes in a community-based text-messaging intervention for neonatal health in western Kenya

Abstract

Sub-Saharan Africa bears the highest burden of neonatal mortality, and despite the availability of effective essential newborn care interventions, progress has been hindered by persistent barriers to timely care, including limited access to health services, health system constraints, and socioeconomic inequities. Digital health interventions integrated with community health worker (CHW) programs offer a promising approach to extend care beyond health facilities, improve communication, and strengthen continuity of care during the critical neonatal period. However, these interventions can risk widening existing health inequities if differences in digital access, skills, infrastructure, and implementation capacity are not addressed. Although implementation science provides tools to evaluate how interventions are delivered in real-world settings, less attention has been paid to equity of implementation of digital health interventions in low-resource contexts. This dissertation examined the equitable implementation of CHV-NEO, a community-based two-way text messaging intervention for peripartum women integrated within Kenya’s electronic Community Health Information System (eCHIS). Through three complementary studies, this dissertation provides evidence to inform the equitable design, implementation, and scale-up of digital health interventions in resource-constrained settings. In the first study, we explored multilevel determinants of equitable implementation of CHV-NEO through eight focus group discussions (FGDs) with 62 CHWs and peripartum clients across four intervention facilities, guided by the Health Equity Implementation Framework (HEIF). Findings showed that while CHV-NEO was broadly acceptable, equitable engagement was shaped by determinants operating across household, organizational, and societal levels. Phone sharing and spousal control emerged as household-level barriers reflecting a gendered digital divide, with poor connectivity, unreliable electricity, limited digital infrastructure and health system resource constraints reducing equitable feasibility and engagement. Participants recommended multimodal communication, greater male engagement, and investments in digital infrastructure and women’s empowerment to support more equitable implementation. The second study evaluated the implementation outcomes of CHV-NEO and their determinants using data from 405 CHWs and 1,154 women across 10 intervention facilities. Acceptability, feasibility, adoption, fidelity, and penetration data were assessed descriptively using validated measures. Determinants were identified using multilevel regression models. Overall, CHV-NEO demonstrated high acceptability and feasibility among CHWs and clients (mean scores: 4.6-4.9/5), high adoption (median: 81%), and high penetration, with 85% of eligible women consenting and 78% receiving at least one automated message. However, median response fidelity was low (11%). Phone sharing consistently predicted lower acceptability, feasibility, and likelihood of consenting to receive messages, while women with below-secondary education were more likely to consent and primigravidae women were less likely to receive automated messages. No individual CHW characteristics were associated with implementation outcomes, but substantial between-facility variation in fidelity suggested that organizational context influenced implementation more than provider characteristics. These findings help identify populations and settings requiring additional implementation support. In the third study, we used an explanatory sequential mixed methods design to examine behavioral determinants of CHW fidelity and the mechanisms through which implementation strategies influenced performance. Routine eCHIS data from 264 CHWs were used to measure fidelity before and after the introduction of nine implementation strategies. In-depth interviews were then conducted with 24 purposively selected CHWs representing four fidelity trajectories guided by the Theory of Planned Behavior Framework. Fidelity improved by an average of 20.5% following strategy implementation, although substantial differences persisted between high- and low-fidelity CHWs. While positive attitudes toward messaging and supportive social norms were shared across all CHWs, higher fidelity was distinguished by greater adaptive capacity, including routinized workflows, clinical confidence, and integration of decision-support tools into practice. Most implementation strategies reinforced similar motivational mechanisms across fidelity groups but did not overcome persistent capacity- and context-related barriers among lower-performing CHWs. These findings suggest that fidelity is primarily a capacity rather than a motivational outcome and that equitable implementation requires tailored implementation support, ongoing mentoring, and investments in supportive health system infrastructure beyond introduction of universal implementation strategies. Together this dissertation demonstrates that successful implementation of digital health interventions cannot be judged by average implementation performance alone. While CHV-NEO achieved high implementation across multiple outcomes, important inequities persisted at the household, organizational, and health system levels. Equitable implementation requires addressing structural barriers to digital access, strengthening facility-level capacity and supportive supervision, tailoring implementation strategies to the needs of lower-performing providers, and monitoring fidelity alongside adoption to identify implementation gaps. These findings advance implementation science by extending understanding of equity determinants in digital health implementation and provide practical, policy, and research recommendations to support the equitable scale-up of community-based digital health interventions in low-resource settings.