We attempted to undertake a cluster-randomised trial to study the effect of a simple one day newborn resuscitation training on health worker practices. However, our criteria for randomisation, aiming to ensure health workers were present to be observed in a defined period, resulted in few staff being eligible. We cannot therefore discount the possibility of bias in group allocation although we feel this is unlikely. The training intervention significantly improved the performance of initial resuscitation steps, with 66% initial practices being adequate in the intervention group compared with 27% in the control group. In addition, there were significant reductions in the frequency of inappropriate and potentially harmful practices and improvements in overall resuscitation scores. There was no obvious effect of training on mortality of babies resuscitated, no obvious decline in asphyxia admission rates and no overall decline in newborn mortality in the hospital as the number of trained providers increased. However, this study was neither specifically designed nor powered with mortality as the primary outcome and our mortality results are best used to inform the design of future studies. In addition, appropriate initial resuscitation is clearly only the first stage in a continuum of effective care, not addressed by this intervention, that is likely to be required to prevent many adverse outcomes from severe asphyxia.
We are not aware of any previous randomised controlled studies examining the effect of resuscitation training on provider practices in a true clinical setting. The majority of studies on newborn resuscitation have focussed on less direct outcomes such as participants' knowledge and skills 
. Such surrogate outcomes may not necessarily reflect practice changes, a more useful and direct way of measuring the effectiveness of resuscitation training programmes 
. Although our primary study outcome was only able to capture the initial steps in effective practice we believe it does indicate an important behaviour change effect, especially if considered together with the reduction in unnecessary / potentially harmful practices and an improvement in overall resuscitation scores.
Previous studies and our control group data demonstrate that both resuscitation skills and knowledge are poor despite frequent exposure to situations in which both are needed 
Internationally, there is now considerable consensus on how newborn resuscitation should be provided 
and it is believed that in 95% cases when it is required resuscitation should be possible with only a minimum of equipment and without access to intensive care skills or facilities 
. Recent research findings have strengthened this opinion demonstrating that suction in the presence of meconium and the use of oxygen are in most newborns unnecessary 
. These findings have relevance to our study as the failure to provide suction to a non-breathing baby born through meconium as the first step was a major reason for failing to achieve a ‘perfect’ classification in our primary outcome. If, as seems likely, there is little value of suction in these babies then a substantial clinical impact from our intervention, 66% of adequate appropriate practices in trained providers, might be a more reasonable interpretation than the modest impact suggested by only 25% of initial practices in trained providers being perfect.
Our data add to a body of knowledge suggesting some improvement in clinical outcomes 
or in acquisition of knowledge and skills of providers following resuscitation training 
. In a systematic review on the effectiveness of all types of life support courses all the three mortality and morbidity studies indicated a positive impact, with an overall odds ratio of 0.28 (95% CI 0.22–0.37). However, no net increase in scores in 5/8 studies of retention of knowledge and in 8/9 studies of skills retention were apparent, although all the studies assessing behavioral outcomes were reported to be methodologically weak 
Similarly, our study has limitations. Attempts to randomise health workers had limited success. We cannot exclude the possibility of cross-group contamination, although this would tend to reduce the apparent effect of the intervention. In contrast it is likely that the difficulty in maintaining observer blinding could bias the results in favour of an intervention effect. If the observers, even unintentionally, were more likely to view the practices of a provider they came to know was trained as correct this would bias our results despite our efforts in training to limit this effect. We also only observed practitioners for a short period after training and are unable to provide any information on the duration of the training effect. In the few studies assessing the duration of effect a rapid and linear decay in cardio-pulmonary (CPR) skills -from as early as two weeks after training, with skills deteriorating to pre-training levels by one year, have been reported 
For low-income countries Life Support Courses are associated with relatively high direct and opportunity costs (learners'/instructors' time, equipment purchase, etc). While there is increasing pressure to implement such courses it is important that their true effects on actual health worker performance and ideally morbidity and mortality are established. Such studies need to be based in typical, low-income settings where supervision and opportunities for continuous learning or ongoing mentorship and resources for post-resuscitation care may be limited. In addition, they should perhaps consider a range of possible training delivery mechanisms, be embedded in local health systems to promote sustainability, assess impact over the long term and consider costs and cost effectiveness to optimise appropriate health policy decisions. Clearly such studies will require appropriate levels of funding.
In conclusion, our findings suggest that implementation of a simple one day newborn resuscitation training can be followed by significant, short-term improvement in health workers' practices. To ensure a high proportion of all resuscitation episodes are appropriately managed clearly a large majority of providers must be trained. Evidence on effects on long term performance or clinical outcomes, however, remain inconclusive and can only be established by larger trials. The availability, accessibility and correct functioning of basic resuscitation equipment is still a missing essential pre-requisite for the success of training and resuscitation itself in many settings