Polygenic risk and COVID-19 severity diverged once pre-infection health entered the model

Polygenic scores tied to several disease domains tracked with COVID-19 severity in two Danish cohorts, but the strength of those associations changed once pre-infection health was folded into the analysis. The pattern was not uniform. Some genetic signals weakened after adjustment for body mass index, physical health, mental health, or comorbidity burden, while others held nearly steady despite those additions.

The study examined two complementary populations: 14,364 participants in the Danish Blood Donor Study, where the outcome was self-reported bedridden COVID-19, and 115,515 participants in the Copenhagen Hospital Biobank, where severity was defined through registers among individuals with documented SARS-CoV-2 infection. Across the two cohorts, eight polygenic scores were associated with bedridden COVID-19 in the blood donor sample, and nine were associated with register-defined severity in the hospital biobank. Seven of those associations appeared in both cohorts and moved in the same direction.

Why the pre-infection baseline mattered so much

The strongest message from the analysis is not simply that genetics tracked with severe infection. It is that some of the observed genetic signal overlapped with health status already present before infection, and the degree of overlap depended on which trait was being measured. In other words, a polygenic score may appear to predict worse COVID-19 outcomes partly because it tags biological pathways that also shape chronic health before the virus ever arrives.

This becomes clear in the sequential adjustment models. In the Danish Blood Donor Study, the risk difference per 1-SD higher ADHD polygenic score dropped from 1.75 percentage points after adjustment for pandemic period to 1.31 after additional adjustment for pre-infection BMI and physical and mental health. The corresponding influenza polygenic score association changed only modestly, from 2.00 to 1.83 percentage points. In the Copenhagen Hospital Biobank, the ADHD score decreased from 0.49 to 0.40 percentage points after adjustment for pre-pandemic comorbidity, while the COVID-19 severity polygenic score stayed at 0.19 percentage points before and after that adjustment.

A single adjustment can separate a direct infectious-disease signal from a broader marker of underlying health burden.

Different genetic pathways behaved differently after adjustment

The changing estimates matter because they point to heterogeneity across traits. A polygenic score for ADHD did not behave like the score for COVID-19 severity itself, and the influenza score did not attenuate to the same degree as the ADHD score. Those differences suggest that cross-trait polygenic associations are not interchangeable. Some may capture shared biology relevant to immune response or inflammatory tone, while others may function more like proxies for pre-existing health behaviors, cardiometabolic status, or psychiatric comorbidity that alter the course of infection once it begins.

Vaccination adjustment added relatively little further attenuation overall. That finding leaves the central pattern intact: measured pre-infection health explained part of the observed relationship, but not all of it. The residual association after those adjustments implies that genetic liability still contributed to severity, although the size of that contribution varied across cohorts and across scores.

The study used population-standardised risk differences and 5,000 bootstrap iterations to generate 95% confidence intervals, a statistical approach that gives the estimates a common scale and helps stabilize uncertainty around the reported effects. The analysis also applied false-discovery-rate correction before carrying forward the polygenic scores into the sequential models, which reduces the chance that isolated findings are being driven by multiple testing alone.

Two cohorts, two severity definitions, one recurring pattern

The two Danish settings were deliberately different, and that difference strengthens the interpretation. One cohort relied on self-reported bedridden illness, a direct patient-centered measure of acute functional impact. The other used register-defined severity, anchored in documented infection and clinical records. Despite those different outcome definitions, seven associations were shared across cohorts and remained directionally concordant.

That convergence argues against the associations being a measurement artifact from one particular dataset. It also raises a practical question for human genetics and infectious disease research: how much of a polygenic association with severe infection reflects susceptibility to the pathogen itself, and how much reflects the pre-existing health state that makes a person less resilient once infected? This study does not collapse those possibilities into a single answer. Instead, it shows that the balance shifts depending on the trait under study.

For clinicians and researchers, the implication is straightforward. Genetic associations with infectious disease severity should not be interpreted in isolation from the patient’s baseline health profile. BMI, physical health, mental health, and comorbidity can move the estimate meaningfully, and the direction of movement is not the same for every polygenic score.

What the Danish data can and cannot settle

These findings come from two large Danish cohorts, but they still leave important constraints on interpretation. The study design can show association and attenuation after adjustment, but not the full causal chain linking genotype, baseline health, vaccination, and severity. Pre-infection health was measured in two different ways across the cohorts, and those differences may partly explain why attenuation was heterogeneous rather than uniform.

Even so, the clinical logic is compelling. Severe COVID-19 is rarely the product of one factor alone; it emerges from an interaction between host genetics, chronic disease burden, immune competence, and exposure context. A polygenic score linked to worse outcomes may therefore be marking vulnerability that was already present before the virus appeared. When that score keeps its association after adjustment, it suggests a more direct biological contribution to outcome severity. When it fades, the score may be reflecting upstream health status more than acute infectious risk.

The study leaves one narrow but important conclusion: pre-existing health conditions can reshape the apparent effect of polygenic liability on COVID-19 severity, and the extent of that reshaping differs across genetic traits. For future infectious-disease genetics, that is not a statistical footnote. It is part of the phenotype itself.


Reporting by Dr. A. Nikhil Krishnan, MBBS for Medicine World Foundation, with additional data provided by Nature Medical Research.