More than 7 million children die each year worldwide[1], most from preventable causes and almost all are in poor countries[1, 2]. Of these, an estimated 3 million babies die in the first 4 weeks of life, the majority in the first week, accounting for 41% of mortality under 5 years [1, 2]. The Millennium Development Goal 4 for child survival (MDG-4) is therefore unlikely to be met in many countries without a substantial reduction in neonatal mortality [1, 2]. Several interventions to help reduce neonatal mortality should be delivered or supported by rural hospitals. These include keeping babies warm, provision of oxygen, assisted feeding and administration of antibiotics [3–5]. From a system perspective, these elements of care have much in common with those required for another highly vulnerable patient population, those with severe acute malnutrition, whose care is organized by the World Health Organization (WHO) into 10 steps [6, 7]. However, in reports from African hospital settings mortality in both these vulnerable groups is often at least 20% [8–10]. At least for malnutrition much of this high inpatient mortality has been attributed to provision of inappropriate and poor quality clinical care [8, 11].

Given the high mortality of these conditions and the development of packages of interventions that provide similar challenges for service delivery mechanisms we set out to explore how well such services are provided in Kenya, conducting a descriptive study while undertaking a parent multi-centre intervention study [12–14]. Our aim, using a simple set of indicators, was to provide basic data on quality of care helping to define the level of need for improvement and to inform design of future intervention studies.

We and others have previously reported poor quality of care for children with common, severe illnesses in low-income country hospitals [17, 19–21]. Others have reported high specific case fatality rates in African hospitals for sick newborns and children with severe malnutrition [8–10]. Our baseline data indicate, despite these reports and the production of international guidelines aimed at reducing mortality, that care for such vulnerable groups remains poor, at least in rural Kenyan hospitals. The inadequacies in care, particularly that for neonates, at levels of the health system expected to provide expertise, supervision and leadership in support of primary care are of major concern and threaten progress towards the 4^th Millennium Development Goal.

Inadequacies or errors in care spanned use of basic antibiotics including penicillin and gentamicin. These drugs have been recommended for neonatal sepsis and severe malnutrition for decades and are used by health workers in probably tens of thousands of doses annually for neonatal sepsis in Kenya alone. Gentamicin is also being considered internationally for use by community health workers [22–24]. In the post-baseline period, after provision of guidelines, penicillin dosing improved. However, errors remained common for gentamicin prescriptions. Often it seemed errors were because prescribers were unaware of, or failed to consider, the specific dosing requirements for newborns in the first week of life. Poor prescribing was not confined to neonatal care however. Vitamin A, that can prevent blindness in children with severe malnutrition, was not prescribed to children at baseline although there was improvement in the post-baseline period. The results we report are a major concern given the critical importance now given to providing essential packages of care [5, 25–28].

A report from these study hospitals at baseline on their readiness to provide newborn care indicates that they were ill prepared [15] and this may explain why care for newborns and those with malnutrition seemed to improve relatively little. However, in a recently published paper by Ayieko and colleagues[16] evaluating the parent study in these hospitals it was demonstrated that face to face feedback of performance, supportive supervision and provision of a local facilitator but no additional material resources did result in improvements of paediatric care and the readiness for providing care over the same period. Although the data for newborn care and care of those with malnutrition is poorer quality, the findings presented and experience during the intervention, suggest that a lower emphasis placed on supervision and feedback in these clinical areas may have contributed to their generally poorer performance improvement. In addition our experience suggests that health workers, who often receive as little as 2 weeks training in neonatal care, had limited knowledge and skills on the use of essential drugs or appropriate supportive care including feeding in vulnerable, often preterm, babies and children with severe malnutrition. Such shortcomings emanate from limitations in basic training for many health workers.

The other striking finding of our study is that it would appear almost impossible to make any reasonable assessment of neonatal care based on routine hospital records. Although we targeted a total of 960 records (i.e. 30 case records per site per survey) for each patient group record retrieval was as low as 7 newborn records per site per 6 months inter-survey period, a rate clearly inconsistent with likely workloads. Unfortunately even if records are present clinical documentation is often so poor that it is impossible to assess the adequacy of practice with even mortality data often missing. Such failings indicate an inability to assess quality of care, adequacy of intervention coverage and outcomes for newborns, all of which are essential to understanding health system performance.

Consistent with findings from other studies [29–36] and in the context of an intervention programme, there was marked improvement in case-specific documentation if a structured form was used, however it is difficult to quantify the extent to which this represented actual improvement of care - and this problem is an inevitable consequence of assessing quality from routine records in any setting. However, we are confident that changes in drug prescribing would lead to changes in actual care delivered and the changes observed in this area are similar to those in other areas [16].

Although standardized admission record forms may have the drawback of taking up more clinician time while staff become familiar with them the direct benefits of comprehensive evaluation, that may prompt more complete problem identification and consequently better treatment, would appear to outweigh this problem, particularly as there may be indirect benefits of decreased cost to the health system due to better use of resources. More efficient methods of clinical data collection could result from introduction of electronic medical records systems which have the further advantage of employing inbuilt clinical decision support systems and periodic reports. However such interventions will be limited by the availability of funds, expertise and electricity in many resource limited settings.

The data we report have several limitations. These include: a relatively small number of hospitals studied; the potential for selection bias at the point of record retrieval; possible ascertainment bias at the point of data abstraction from written records; misclassification of cases when populating quality indicators; and, the implicit assumption that the written records reflect actual practice. Despite this we feel our data still provide useful insight into the process and quality of care provided to newborns and severely malnourished children in Kenyan rural hospitals. Data of this type are rarely reported and we suspect, but cannot confirm, that data from Kenya may be indicative of problems present much more widely in low-income African settings given reports of poor quality of care for children in the region [37].

In conclusion, our preliminary data indicate that many health workers at hospital level are unable to provide appropriately planned care for sick newborns or children with severe malnutrition. Inadequacies are seen in key tasks such as prescription of antibiotics and feeds even when resources are available and after provision of guidelines. Further, without improvements in hospital based information systems it will be impossible to know whether the frequent calls for deployment of and investment in interventions to improve care are being heeded. Problems exist despite a long history of development and formulation of international, and adapted national, guidelines. Future research should examine therefore how best to implement better care in rural hospitals, enabling them to serve their expected role in global efforts to reduce neonatal and child mortality in low-income settings such as Kenya.