Clinical Trials

❘❙❚■ Chapter 26 | ConfoundingTable 3. Hormone replacement therapy (HRT) trial example: association between socioeconomic statusand treatment.Treatment Number in each socioeconomic status group (%) TotalLowHighNo HRT 300 (75) 100 (25) 400HRT 200 (50) 200 (50) 400χ 2 = 53.27; P < 0.0001.The results in Table 2 show that there is no difference in the proportions of womenwith mental function improvement between the two treatment groups if they areconsidered separately by socioeconomic class. The proportion of women withimproved mental function was 40% in the high socioeconomic status group and8% in the low socioeconomic status group, regardless of treatment. Accordingly,the estimated odds ratio is 1.00 in each group for both levels of socioeconomicclass. Therefore, socioeconomic status did indeed confound the associationbetween HRT treatment and improvement in mental function, and so theapparent difference found by the original analysis in Table 1 is spurious.How can we confirm whether a variable is a confounder?For a variable to be a confounder it must satisfy three conditions [1,2]:• It must be associated with the treatment.• It must be a predictor of the outcome being measured.• It must not be a consequence of the treatment itself.We can illustrate how to identify a confounder using the hypothetical HRT studyas an example.Step 1. Assess whether the potential confounder has an associationwith the treatment groupFrom the HRT example, it is clear that a variable can confound the relationshipbetween treatment and outcome only if it is unevenly distributed between thetreatment groups. In our example, 50% of the women taking HRT were of highsocioeconomic status compared with only 25% of women in the non-HRTgroup, indicating that the distribution of socioeconomic status among the twotreatment groups was imbalanced. The chi-square test shown in Table 3 confirmsthat this imbalance was highly statistically significant (χ 2 = 53.27; P < 0.0001).298