Reply to the Letter

Abstract

We acknowledge that multilevel analysis represent a useful tool in the assessment of health care performance across hospitals. Even if some studies as those quoted by Merlo, used multilevel model to take into account intra-cluster correlation, the majority of studies on the issue of recourse to c-section recurred to traditional logistic regression models to produce leagues table.In our case we supposed that the variability across structures could be attributed to well-known hospital factors as dimension and administrative type as well as staff attitude. In fact in our logistic regression analysis the risk adjusted OR of 8 out if 9 completely private units were in the 4th quartile of the odds ratios distribution.However we agree with Merlo that further enquiry should be made on this issue. Therefore, using STATA 8 software, we run a multilevel model and compared the predicted caesarean cases across hospitals with or without multilevel standardisation (Figure 1).The predicted values were very similar in both models, indicating that residual variance in the model, which did not take in account the hierarchical structure of data, was negligible.In conclusion we consider that Multilevel techniques should not be applied automatically, but should be based on an in depth knowledge of the structure of data. References 1. Merlo J. Changing analytical approaches in European epidemiology -a short comment on a recent article. Eur J Epidemiol 2005; 20(8): 737.