For the present analysis, information from the case
notes was abstracted using a standardized collection form. Once the complete data set had been collected, it was analysed by a non-clinical statistician who was blinded to the aetiological relationships and hypotheses being tested. These methods have been subsequently described and validated by Gilbert et al.41 Seven cases were excluded from analysis because either the history and/or the documentation with respect to alcohol consumption was considered to be unreliable, leav ing 553 (99%) of the complete series in the present analysis.