Contingency Table Analysis in Obstetrics and Gynaecology

Abstract

Contingency table analysis for investigating the association between two categorical variables is common. Methods often employed are the χ2 test and Fisher’s exact test. However, they  have been sometimes misused and there has been no systematic evaluation in obstetrics and gynaecology. Therefore, we aimed to describe the available methods of analysing contingency tables and to evaluate the performance of recent literature in obstetrics and gynaecology. Searching the first three issues of the Obstetrics and Gynecology journal in 2008 and 2003 identified respectively 34 (57%) and 46 (62%) studies with the use of χ2 test, Fisher’s exact test, or exact χ2 test specified or actually performed. However, only 22 (in 2008) and 28 (in 2003) studies have sufficient data for verifying the actual use of tests; of which, 82% (exact 95% confidence interval = 60% to 95%) and 67% (exact 95% confidence interval = 46% to 83%) respectively had at least one inadequate use. Those common ones included non-reproducible p values, using χ2 test without fulfilling its requirements, and using χ2 test for 2x2 tables when Fisher’s exact test is available. No significant differences between the two publication years were observed. Exact χ2 test, though not widely implemented, is more appropriate for contingency table analysis. Substantial inadequate use of tests in contingency table analysis was observed in a highly ranked journal in obstetrics and gynaecology and no notable improvement if not worsening was observed over the past 5 years. More effective communications between clinicians and statisticians are required.