Polygenic prediction of treatment response for precision medicine is vulnerable to collider bias

Collider bias induced by trial eligibility criteria, creating an apparent interaction between polygenic risk score and treatment effect

Reanalyses of major lipid-lowering trials have reported substantially greater treatment benefit in individuals with higher coronary artery disease polygenic risk scores (PRSs); this study tests whether trial eligibility criteria could generate this pattern through collider bias. In simulations where the true PRS-by-treatment interaction was set to zero, the authors reproduced apparent effects similar to those of three trial reanalyses, including FOURIER-like hazard ratios of 0.77 in the high PRS stratum and 0.95 in the low PRS stratum. In the Helix Research Network (n = 175,106) and UK Biobank (n = 92,859) cohorts, without trial eligibility criteria, the coronary PRS showed little evidence of modifying the effect of statins on acute coronary events, and the LDL PRS did not modify LDL cholesterol response to statins. By contrast, stratifying on baseline LDL cholesterol induced an apparent PRS-dependent response matching the simulations, and the authors stress that randomization does not protect against this eligibility-induced bias.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

The work is methodological but its consequences are concrete: a "greater benefit in high-PRS individuals", invoked to target a treatment, may be an artifact of participant selection rather than a biological property. Any proposal to guide prescribing with a polygenic score should therefore be tested outside trial populations filtered on a biomarker related to the score. A non-peer-reviewed preprint, some of whose figures could not be finalized because UK Biobank access was suspended — the estimates themselves are stated to be unchanged.

Analysis by Dr Thibaut Benquey