BackgroundWe have reported the results of a cluster randomized trial of rural Kenyan hospitals evaluating the effects of an intervention to introduce care based on best-practice guidelines. In parallel work we described the context of the study, explored the process and perceptions of the intervention, and undertook a discrete study on health worker motivation because this was felt likely to be an important contributor to poor performance in Kenyan public sector hospitals. Here, we use data from these multiple studies and insights gained from being participants in and observers of the intervention process to provide our explanation of how intervention effects were achieved as part of an effort to better understand implementation in low-income hospital settings.MethodsInitial hypotheses were generated to explain the variation in intervention effects across place, time, and effect measure (indicator) based on our understanding of theory and informed by our implementation experience and participant observations. All data sources available for hospitals considered as cases for study were then examined to determine if hypotheses were supported, rejected, or required modification. Data included transcriptions of interviews and group discussions, field notes and that from the detailed longitudinal quantitative investigation. Potentially useful explanatory themes were identified, discussed by the implementing and research team, revised, and merged as part of an iterative process aimed at building more generic explanatory theory. At the end of this process, findings were mapped against a recently reported comprehensive framework for implementation research.ResultsA normative re-educative intervention approach evolved that sought to reset norms and values concerning good practice and promote 'grass-roots' participation to improve delivery of correct care. Maximal effects were achieved when this strategy and external support supervision helped create a soft-contract with senior managers clarifying roles and expectations around desired performance. This, combined with the support of facilitators acting as an expert resource and 'shop-floor' change agent, led to improvements in leadership, accountability, and resource allocation that enhanced workers' commitment and capacity and improved clinical microsystems. Provision of correct care was then particularly likely if tasks were simple and a good fit to existing professional routines. Our findings were in broad agreement with those defined as part of recent work articulating a comprehensive framework for implementation research.ConclusionsUsing data from multiple studies can provide valuable insight into how an intervention is working and what factors may explain variability in effects. Findings clearly suggest that major intervention strategies aimed at improving child and newborn survival in low-income settings should go well beyond the fixed inputs (training, guidelines, and job aides) that are typical of many major programmes. Strategies required to deliver good care in lo...