In a cluster randomised controlled trial of a policy to provide community breastfeeding support groups in Scotland, breastfeeding rates declined in 3 of 7 intervention localities. From a preliminary study, we expected breastfeeding outcomes to vary and we prospectively used qualitative and quantitative methods to ask why. Ethnographic in-depth interviews, focus groups, observations and survey data were analysed to build seven embedded case studies. A pyramidal model of how primary health service organisations implemented the policy was constructed prior to knowing trial outcomes to minimise bias. Informed by a realist approach, the model explained variation in (a) policy implementation (b) the breastfeeding outcomes, whereas the quantity of intervention delivered did not. In the three localities where breastfeeding rates declined, negative aspects of place including deprivation, unsuitable premises and geographical barriers to inter-professional communication; personnel resources including staff shortages, high workload and low morale; and organisational change predominated (the base model tiers). Managers focused on solving these problems rather than delivering the policy and evidence of progress to the higher model tiers was weak. In contrast, where breastfeeding rates increased the base tiers of the model were less problematic, there was more evidence of leadership, focus on the policy, multi-disciplinary partnership working and reflective action cycles (the higher model tiers). We advocate an ethnographic approach to the design and evaluation of complex intervention trials and illustrate how this can assist in developing an explanatory model. More attention should be given to the complex systems within which policies and interventions occur, to identify and understand the favourable conditions necessary for a successful intervention.
2009 Elsevier Ltd. All rights reserved.