Patients' nutritional status is not recorded

used in the multivariate analysis.For instance, could social class factors (smoking history, body mass index) or other factors (drug treatment) have varied among the hospital populations and affected the results?Similarly, could the large observed differences between the sexes indicate particular comorbidities that were not measured?The proportion of the total variability in 90 day mortality that was due to the casemix factors used in the model is not stated.If this proportion was small there may have been other casemix factors that were not taken into consideration; the modelling process used by the authors could be better judged if all the factors considered were stated and the extent of the total variability in 90 day mortality explained by the model was recorded.It seems implausible that treatment in hospital 6 should have exerted such a pronounced protective effect.Appreciable variations in survival due to differences in the processes of care among hospitals are difficult to identify after adjustment for casemix.2Adequate adjustment for casemix is difficult, and many studies tend to collect few data with which to attempt such adjustments.This study may have considered too few casemix factors for any reliable comment to be made about the effect of the processes of care in the different hospitals.