The association between income and mental health: can sibling comparison designs help resolve this research question?

Sibling comparison designs are increasingly used to address unmeasured familial confounding in observational studies. We propose that three key interpretational fallacies - sibling characteristics, exposure correlation and non-shared confounding, and unmet life course model assumptions - can mislead causal conclusions. We demonstrate these fallacies by investigating childhood family income and mental health. A nationwide Danish cohort of individuals born between 1986 and 1996 (n = 643,814; 404,179 siblings) was followed-up from age 15 until onset of severe mental disorders. Population-wide and within-sibling adjusted hazard ratios (aHR) between childhood family income and offspring mental disorders were estimated, supplemented by descriptive statistics and pseudo-sibling analyses. A $15,000 increase in family income at age 14 was associated with a reduced rate of severe mental disorders (aHR = 0.78; 95% CI: 0.76-0.81), with comparable estimates across measurement ages 1-14 (range: 0.67-0.82). Null results were observed in both a pseudo-sibling cohort of unrelated individuals with the same income differences as the true sibling cohort (aHR = 0.93; 95% CI: 0.85-1.01) and the true sibling cohort (aHR = 1.02; 95% CI: 0.94-1.11). Siblings were typically born three years apart, with an average monthly income difference of $496 at age 14 (IQR;$150-$641). This study advocates for cautious causal interpretation of null results in sibling comparison studies because (1) it may not capture meaningful differences in family income across siblings due to minor income fluctuations; (2) the pseudo-sibling cohort showed evidence of unmeasured non-shared confounding; (3) sibling comparison designs test a critical periods life course model, but the results favour an accumulating/vulnerable model. We present guidelines and R syntax to assess these interpretational fallacies.