The study design aimed to balance intervention and control group hospitals on the basis of size, presence of a paediatrician and medical officer interns, and some basic characteristics of the geographic location [5
] (Additional File 1
, Figure ). It can be seen this also resulted in reasonable balance with respect to many gross structural attributes of hospital care, including organizational aspects of care, availability of human resources, equipment, and drugs at baseline. Intervention group hospitals, however, tended to have higher inpatient paediatric mortality. For indicators related to the process of hospital care, intervention group hospitals and control group hospitals fared equally badly in general at baseline. These baseline data also indicate that little progress had been made in improving paediatric care, or in implementing available WHO and national treatment recommendations in the four years between the baseline surveys reported and similar surveys in 2002 involving seven of these hospitals [3
Although the baseline cross-sectional data provide some reassurance that the process of randomization helped achieve group balance, the highly dynamic reality of hospitals evidenced by the prospectively organized approach to description underscores the need for caution when interpreting the results of the intervention in the future. The data presented are, we hope, an aide to those interested to consider for themselves the plausibility of any cause and effect relationship attributed to the intervention. National level developments, such as improved health spending or introduction of new management approaches, both of which occurred during the intervention period, would be expected to affect all hospitals in a similar way and no specific regional initiatives were encountered. However, we cannot discount the possibility that national directives are differentially applied and/or significantly affected by a hospital's local administration and management, potentially affecting uptake of new hospital and health worker practices. More obviously, it is clear that hospitals work with a range of partners and initiatives at a local level. None during our observations targeted improvements in child and newborn health care broadly other than the planned intervention. However, some may have influenced the quality of service provision for specific aspects of care, such as the provision of incubators for the newborn, effects that might be attributed to our broad intervention unless documented. Alternatively, the intervention's effectiveness might be negatively affected by prioritization of other areas, and in this regard it is interesting to note that seven of eight hospitals were working with non-governmental partners supporting HIV-related activities often bringing considerable resources.
Such rich contextual data have a number of implications. Firstly, the diversity encompassed by the simple term 'hospital', even in a sample of only eight in one low-income setting, is striking. This is rarely considered in national or international debate, or in interpretation of results of research or evaluation. Secondly, hospital management and staffing are clearly likely to be poorly described by a single round of data collection. In the light of our study, it should also be clear, despite some reassurance provided by randomization, that there is considerable scope for residual confounding and bias to influence the direction of results, both of which may be time-varying in direction or magnitude. Such careful descriptions of the type we have attempted may allow the plausibility of any causal relationship between intervention and response to be scrutinized and debated, but do not overcome these potential problems of bias and hidden confounding. While very large randomized controlled trials might be expected to provide a solution, it is questionable whether they are feasible and even if performed it would appear prudent that they still be accompanied by detailed description.
Perhaps the most striking finding resulting from our attempts to track changes in hospital contexts is the rapidity of turnover in senior hospital management, senior departmental nursing staff, and clinical service providers relevant to delivery of paediatric and newborn services. Such turnover was apparent in all hospitals and resulted from staff transfers between hospitals, locally controlled internal staff rotation of nurses, scheduled rotation of clinical staff linked to training requirements and, where clinical staff were few, reallocation of clinical staff away from paediatric and newborn areas that were considered a low priority. Thus any intervention aimed at changing service provision must transcend these staff dynamics to be successful in changing practice over the long term. A factor that encouraged us to explore the role of a local facilitator as part of the intervention, alternatively, or in addition, hospitals or implementers interested in achieving long-term change may need to develop strategies for expert staff retention. While this might encompass incentives to retain staff in rural or underserved areas, thought should also be given to revising routine staff rotation policies.