Polygenic prediction of treatment response for precision medicine is vulnerable to collider bias
- 15 Prime Sciences
Abstract
Recent reanalyses of major lipid-lowering trials have reported substantially greater treatment benefit in individuals with higher polygenic risk scores (PRSs) for coronary artery disease (CAD). We tested whether trial eligibility criteria could generate this pattern through collider bias. In simulations where the true PRS-by-treatment interaction was set to zero, we reproduced apparent PRS-dependent treatment effects similar to those observed in reanalyses of three trials. This included FOURIER-like hazard ratios for lipid-lowering treatment versus placebo of 0.77 in the high CAD PRS stratum and 0.95 in the low CAD PRS stratum. In the Helix Research Network (n = 175,106) and UK Biobank (n = 92,859) cohorts, where we did not apply the trial eligibility criteria, the CAD PRS showed little evidence of modifying the effect of statin use on acute coronary events. We then used LDL cholesterol response as a better-powered test of collider bias: we found no evidence that LDL PRS modified statin response in population cohorts, yet stratifying on baseline LDL cholesterol induced apparent PRS-dependent response matching the simulations. Importantly, treatment randomization does not protect against eligibility-induced collider bias. PRS-dependent response may therefore reflect bias from eligibility rather than biology. These results should inform the planning and execution of clinical trials for precision medicine.
A note on this versionAccess to the UK Biobank Research Analysis Platform has been suspended since April 2026. The UK Biobank estimates reported in this article were generated before the suspension and are based on completed analyses. However, we have not been able to update Extended Data Figures 3 - 5 and Figure 8 to publication-ready versions. This limitation affects the design and presentation of the figures but does not change the underlying results. The figures will be updated once access is restored.
O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=99 SRC="FIGDIR/small/26365857v1_fige3.gif" ALT="Figure 3"> View larger version (13K): [email protected]@b5f36eorg.highwire.dtl.DTLVardef@180ce9corg.highwire.dtl.DTLVardef@17c9f02_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOExtended Data Figure 3:C_FLOATNO Secondary analysis of UKB primary care (n = 92,859) and HRN electronic health records (n = 175,106), testing whether a CAD PRS modifies the effect of statins on risk of acute coronary events within 10 years of statin initiation. IPTW is compared to linear regression with and without adjusting for baseline LDL cholesterol. The heading "modification of statin response by CAD PRS" reports an interaction between statin initiation and CAD PRS on risk of acute coronary events. Positive values of the interaction indicate that the CAD PRS is blunting the effect of statins, while negative values indicate it is conferring greater response.
C_FIG O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=89 SRC="FIGDIR/small/26365857v1_fige5.gif" ALT="Figure 5"> View larger version (9K): [email protected]@1ca4b51org.highwire.dtl.DTLVardef@16f255org.highwire.dtl.DTLVardef@7fb9e1_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOExtended Data Figure 5:C_FLOATNO Cohort-specific estimates underlying the meta-analysis of LDL PRS-dependent statin response in Fig. 6. Interaction estimates between statin initiation and LDL PRS are shown separately for UK Biobank primary care (UKB; lek) and the Helix Research Network (HRN; right), in the full cohort and within each tertile of baseline LDL cholesterol. Within each analysis, inverse probability of treatment weighting was used to balance measured determinants of statin initiation between initiators and non-initiators. Positive interaction coefficients indicate that higher LDL PRS is associated with a blunted LDL cholesterol response to statin initiation; negative coefficients indicate greater response. Error bars show 95% confidence intervals.
C_FIG O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=83 SRC="FIGDIR/small/26365857v1_fig8.gif" ALT="Figure 8"> View larger version (18K): [email protected]@1c3e93dorg.highwire.dtl.DTLVardef@105b30forg.highwire.dtl.DTLVardef@12eb2d3_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOFigure 8:C_FLOATNO Propensity score diagnostics for the primary inverse probability of treatment weighting analysis in UKB primary care (lek) and HRN (right). Lek panels show calibration plots of observed versus predicted statin initiation rates; points lying close to the diagonal indicate that the LASSO propensity model is well-calibrated across the range of predicted values. Right panels show overlapping density histograms of propensity scores stratified by statin initiators and non-initiators; substantial overlap between distributions indicates that both groups are represented across the range of propensity scores, supporting the positivity assumption.
C_FIG
Related
- Maternal–Fetal Oncology Intelligence: Self-Evolving Foundation Models and Digital Twin Ecosystems for Precision Cancer Prediction, Treatment Optimization, and Maternal–Fetal Outcome Forecasting
- Pathway-Specific Polygenic Scores for Predicting Clinical Lithium Treatment Response in Patients With Bipolar Disorder
- Speech markers of depression: cross-sectional discrimination, longitudinal association, and treatment response prediction
- Determinants of variable response to GLP-1 receptor agonists: A mechanism-informed framework for precision treatment of obesity.
- Patient-Derived Organoids in Functional Precision Oncology: Tumor Modeling, Drug-Response Prediction, and Clinical Translation.
BibTeX
@misc{tudball2026polygenic,
title = {{Polygenic prediction of treatment response for precision medicine is vulnerable to collider bias}},
author = {M. J. Tudball and E. T. Cirulli and M. V. Holmes and G. Davey Smith and J. B. Richards},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.18.26365857},
url = {https://www.medrxiv.org/content/10.64898/2026.10.18.26365857v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Polygenic prediction of treatment response for precision medicine is vulnerable to collider bias
AU - M. J. Tudball
AU - E. T. Cirulli
AU - M. V. Holmes
AU - G. Davey Smith
AU - J. B. Richards
PY - 2026
DA - 2026/10/18
PB - medRxiv
DO - 10.64898/2026.10.18.26365857
UR - https://www.medrxiv.org/content/10.64898/2026.10.18.26365857v1
AB - Recent reanalyses of major lipid-lowering trials have reported substantially greater treatment benefit in individuals with higher polygenic risk scores (PRSs) for coronary artery disease (CAD). We tested whether trial eligibility criteria could generate this pattern through collider bias. In simulations where the true PRS-by-treatment interaction was set to zero, we reproduced apparent PRS-dependent treatment effects similar to those observed in reanalyses of three trials. This included FOURIER-like hazard ratios for lipid-lowering treatment versus placebo of 0.77 in the high CAD PRS stratum and 0.95 in the low CAD PRS stratum. In the Helix Research Network (n = 175,106) and UK Biobank (n = 92,859) cohorts, where we did not apply the trial eligibility criteria, the CAD PRS showed little evidence of modifying the effect of statin use on acute coronary events. We then used LDL cholesterol response as a better-powered test of collider bias: we found no evidence that LDL PRS modified statin response in population cohorts, yet stratifying on baseline LDL cholesterol induced apparent PRS-dependent response matching the simulations. Importantly, treatment randomization does not protect against eligibility-induced collider bias. PRS-dependent response may therefore reflect bias from eligibility rather than biology. These results should inform the planning and execution of clinical trials for precision medicine. A note on this versionAccess to the UK Biobank Research Analysis Platform has been suspended since April 2026. The UK Biobank estimates reported in this article were generated before the suspension and are based on completed analyses. However, we have not been able to update Extended Data Figures 3 - 5 and Figure 8 to publication-ready versions. This limitation affects the design and presentation of the figures but does not change the underlying results. The figures will be updated once access is restored. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=99 SRC="FIGDIR/small/26365857v1_fige3.gif" ALT="Figure 3"> View larger version (13K): [email protected]@b5f36eorg.highwire.dtl.DTLVardef@180ce9corg.highwire.dtl.DTLVardef@17c9f02_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOExtended Data Figure 3:C_FLOATNO Secondary analysis of UKB primary care (n = 92,859) and HRN electronic health records (n = 175,106), testing whether a CAD PRS modifies the effect of statins on risk of acute coronary events within 10 years of statin initiation. IPTW is compared to linear regression with and without adjusting for baseline LDL cholesterol. The heading "modification of statin response by CAD PRS" reports an interaction between statin initiation and CAD PRS on risk of acute coronary events. Positive values of the interaction indicate that the CAD PRS is blunting the effect of statins, while negative values indicate it is conferring greater response. C_FIG O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=89 SRC="FIGDIR/small/26365857v1_fige5.gif" ALT="Figure 5"> View larger version (9K): [email protected]@1ca4b51org.highwire.dtl.DTLVardef@16f255org.highwire.dtl.DTLVardef@7fb9e1_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOExtended Data Figure 5:C_FLOATNO Cohort-specific estimates underlying the meta-analysis of LDL PRS-dependent statin response in Fig. 6. Interaction estimates between statin initiation and LDL PRS are shown separately for UK Biobank primary care (UKB; lek) and the Helix Research Network (HRN; right), in the full cohort and within each tertile of baseline LDL cholesterol. Within each analysis, inverse probability of treatment weighting was used to balance measured determinants of statin initiation between initiators and non-initiators. Positive interaction coefficients indicate that higher LDL PRS is associated with a blunted LDL cholesterol response to statin initiation; negative coefficients indicate greater response. Error bars show 95% confidence intervals. C_FIG O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=83 SRC="FIGDIR/small/26365857v1_fig8.gif" ALT="Figure 8"> View larger version (18K): [email protected]@1c3e93dorg.highwire.dtl.DTLVardef@105b30forg.highwire.dtl.DTLVardef@12eb2d3_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOFigure 8:C_FLOATNO Propensity score diagnostics for the primary inverse probability of treatment weighting analysis in UKB primary care (lek) and HRN (right). Lek panels show calibration plots of observed versus predicted statin initiation rates; points lying close to the diagonal indicate that the LASSO propensity model is well-calibrated across the range of predicted values. Right panels show overlapping density histograms of propensity scores stratified by statin initiators and non-initiators; substantial overlap between distributions indicates that both groups are represented across the range of propensity scores, supporting the positivity assumption. C_FIG
ER -