When Sophisticated Models Meet Questionable Premises

Key Takeaways

• MR requires strong, biologically plausible genetic instruments
• Dietary patterns lack robust genetic predictors
• Study relaxed GWAS significance threshold, weakening instruments
• Low‑calorie diet link to psoriatic arthritis is marginal
• Findings may mislead public without clear limitations

Summary

A recent Mendelian randomization (MR) study attempted to determine whether low‑calorie, vegetarian, or gluten‑free diets causally influence inflammatory skin diseases such as psoriasis, psoriatic arthritis, atopic dermatitis, and acne. Using genetic variants from the UK Biobank as proxies for self‑reported dietary patterns, the authors reported a modest 5 % increased risk of psoriatic arthritis for low‑calorie diets and no significant effects for the other diets. The analysis, however, relied on weak genetic instruments that required a relaxed significance threshold and failed to capture lifelong dietary exposure, undermining the core MR assumptions. Consequently, the lone positive finding is likely driven by noise rather than a true causal relationship.

Pulse Analysis

Mendelian randomization has become a cornerstone of modern epidemiology because it leverages the random assortment of genes at conception to mimic a randomized trial. When a genetic variant robustly predicts a physiological exposure—such as LDL‑cholesterol or blood pressure—the method can sidestep confounding and reverse causation that plague observational studies. This genetic‑instrument approach has delivered credible insights into cardiovascular risk, metabolic disease, and drug target validation, earning it a reputation as a “natural experiment” for causal inference.

The recent MR investigation of popular diets and inflammatory skin conditions illustrates why the technique does not automatically translate to behavioral exposures. Dietary patterns are shaped by culture, economics, and personal choice, producing a diffuse genetic signal that often reflects personality or socioeconomic status rather than actual food intake. To compensate, the authors lowered the genome‑wide significance threshold from p < 5 × 10⁻⁸ to p < 5 × 10⁻⁵, admitting that strong predictors were absent. Such weak instruments violate the relevance assumption, and the transient nature of most diets conflicts with the lifetime exposure premise inherent to MR, rendering the modest association with psoriatic arthritis highly suspect.

The broader lesson for researchers and journalists is to scrutinize the biological plausibility of genetic instruments before accepting MR results on lifestyle factors. When robust variants are unavailable, alternative designs—such as longitudinal cohort studies with repeated dietary assessments, randomized feeding trials, or triangulation with metabolomic proxies—may provide more reliable evidence. Until genetics can reliably encode complex behaviors, the epidemiology community must flag MR findings on diet as provisional, ensuring that policy recommendations and public headlines reflect the underlying uncertainty rather than overstated causality.