Simple methods for checking for possible errors in reported odds ratios, relative risks and confidence intervals

Abstract

Meta-analyses of data from epidemiological studies are often based on odds ratios (ORs) or relative risks (RRs) and their 95 per cent confidence intervals (CIs) as reported by the authors. Where possible ORs, RRs and CIs should be checked against the source data. Some simple methods are presented for checking the validity of reported ORs, RRs and CIs where the source data are not available. These methods include inferring the minimum total number of subjects in the study, the minimum total number of cases and the minimum number there must be in any disease/exposure category. Examples taken from the literature on environmental tobacco smoke exposure (ETS) illustrate that errors in published data are not infrequent and may stay undetected in meta-analyses.