Assessment of Confounding and Interaction using the Mantel-Haenszel Risk Estimation Method

Abstract

The association between an exposure of interest (risk factor) and a disease may be confounded by the actionof other separate factors as well as by interactions between risk factors exerting an impact. Crude measures ofeffect may be misleading in such situations. Levels of the potential confounding factor could be estimated usingstratified analysis. Uniformity of the stratum-specific effect estimates can be assessed by performing chi-squaretests for heterogeneity. If the effect is uniform across strata, we can calculate a pooled adjusted summary estimateof the effect using the Mantel-Haenzel (M-H) method. Confidence intervals for the adjusted estimate and theM-H chi-square test are calculated to assess the significance. If the effect is not uniform (presence of interaction),we report stratum-specific estimates, confidence intervals and chi-square for each estimate. In the present paper,assessment of the level of confounding and interaction between risk factors are illustrated using a case-controlstudy of lung cancer conducted at the Regional Cancer Centre, Trivandrum.