ABSTRACT
BACKGROUND: Quality improvement collaboratives are a common approach to improving quality of care. They rely on collaboration across and within health facilities to enable and accelerate quality improvement. Originating in high-income settings, little is known about how collaboration transfers to low-income settings, despite the widespread use of these collaboratives. METHOD: We explored collaboration within quality improvement collaboratives in Ethiopia through 42 in-depth interviews with staff of two hospitals and four health centers and three with quality improvement mentors. Data were analysed thematically using a deductive and inductive approach. RESULTS: There was collaboration at learning sessions though experience sharing, co-learning and peer pressure. Respondents were used to a blaming environment, which they contrasted to the open and non-blaming environment at the learning sessions. Respondents formed new relationships that led to across facility practical support. Within facilities, those in the quality improvement team continued to collaborate through the plan-do-study-act cycles, although this required high engagement and support from mentors. Few staff were able to attend learning sessions and within facility transfer of quality improvement knowledge was rare. This affected broader participation and led to some resentment and resistance. Improved teamwork skills and behaviors occurred at individual rather than facility or systems level, with implications for sustainability. Challenges to collaboration included unequal participation, lack of knowledge transfer, high workloads, staff turnover and a culture of dependency. CONCLUSION: We conclude that collaboration can occur and is valued within a traditionally hierarchical system, but may require explicit support at learning sessions and by mentors. More emphasis is needed on ensuring quality improvement knowledge transfer, buy-in and system level change. This could include a modified collaborative design to provide facility-level support for spread.
