Current clinical use of polygenic scores will risk exacerbating health disparities
Alicia R. MartiniD, Masahiro KanaiiD, Yoichiro KamataniiD, Yukinori OkadaiD, Benjamin M. NealeiD, Mark J. DalyiD
Abstract
Polygenic risk scores (PRS) are poised to improve biomedical outcomes via precision medicine. However, the major ethical and scientific challenge surrounding clinical implementation is that they are many-fold more accurate in European ancestry individuals than others. This disparity is an inescapable consequence of Eurocentric genome-wide association study biases. This highlights that--unlike clinical biomarkers and prescription drugs, which may individually work better in some populations but do not ubiquitously perform far better in European populations--clinical uses of PRS today would systematically afford greater improvement to European descent populations. Early diversifying efforts show promise in levelling this vast imbalance, even when non-European sample sizes are considerably smaller than the largest studies to date. To realize the full and equitable potential of PRS, we must prioritize greater diversity in genetic studies and public dissemination of summary statistics to ensure that health disparities are not increased for those already most underserved.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 2,813
- FWCI
- 37.20
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 3 (1 Feb 2019) · CC BY 4.0
Eurocentric GWAS make polygenic scores less accurate in non-Europeans (mean 4.9-fold lower for UK Biobank Africans); equal-sized ancestry-matched GWAS predict better in Japanese and British individuals.
Polygenic risk scores (PRS) are approaching routine clinical use, yet about 79% of GWAS participants are of European descent versus 16% of the world population, so scores transfer poorly and could widen existing health disparities. Beyond prior reports of this gap, the work traces it to allele-frequency and linkage disequilibrium (LD) differences plus environment, quantifies accuracy decay across ancestries in UK Biobank and tests equal-sized British and Japanese GWAS.
Key results
ShownPolygenic scores from European-ancestry GWAS lost accuracy in every other UK Biobank (UKBB) ancestry group, with losses increasing with genetic divergence from the European discovery sample and largest in African-ancestry participants.
Across 17 anthropometric and blood panel traits, mean prediction accuracy relative to Europeans fell 1.6-fold (Hispanic/Latino Americans), 1.7-fold (South Asians), 2.5-fold (East Asians) and 4.9-fold (Africans); targets were unrelated and excluded from the discovery cohort. · Limited generalizability of PRS across diverse…; Figure legends
ShownWith GWAS of identical sample size run separately in UKBB (British) and BioBank Japan (BBJ, Japanese), ancestry-matched summary statistics consistently gave higher prediction accuracy in each target population, for quantitative traits and disease endpoints alike.
Covered 17 anthropometric/blood panel traits and five diseases; accuracy is the best R^2 over five p-value thresholds with bootstrap 95% CIs. BBJ-derived scores were weaker for quantitative traits but stronger for the diseases, attributed to cohort differences in phenotype precision and observed heritability (hospital-based BBJ vs healthier population-based UKBB), not ancestry. · Prioritizing diversity shows early promise for…; Figure legends
ShownGWAS participation is dominated by European-descent samples with no recent move toward diversity, and catalogued associations are skewed toward variants more common in Europeans than in East Asian or African populations.
GWAS Catalog: ~79% of participants of European descent versus 16% of the global population; the non-European share has been flat or falling since late 2014 and non-European mean GWAS sizes remain several-fold smaller. 1000 Genomes frequencies of catalog variants are higher on average in Europeans than in East Asians or Africans, a pattern not seen for genomic variants overall. · Predictable basis of disparities in PRS accuracy; Common discoveries and low-hanging fruit; Figure legends
ShownAfrican-descent UKBB participants were poorly predicted from British and Japanese GWAS alike, consistent with accuracy decaying with genetic divergence from any discovery cohort, while British and Japanese common-variant effect sizes largely agreed.
African-descent targets fared far worse than British or Japanese targets with UKBB or BBJ scores alike, often only marginally better than random, if at all; British-Japanese trans-ethnic genetic correlations were mostly high but lower for height (0.69) and BMI (0.75). Few ICD-10 thalassemia or sickle cell diagnoses argue against these driving blood-trait gaps. · Prioritizing diversity shows early promise for…; Clinical use of PRS may uniquely exacerbate…; Figure legends
ClaimedUnlike clinical biomarkers, whose ancestry-related shifts can often be handled with population-specific reference intervals, PRS are inherently less informative in groups diverged from GWAS cohorts, so current clinical use would mainly benefit European-descent patients.
Reasoned argument: reference intervals can be set from as few as 120 individuals and recentred per population, whereas PRS stay less informative in diverged groups even with multi-ethnic references; the authors urge that any clinical-use discussion quantify economic and health-disparity impacts. · Clinical use of PRS may uniquely exacerbate…
Methods, data and reproducibility
Perspective plus analyses: GWAS Catalog ancestry trends; 1000 Genomes allele-frequency and LD contrasts with an analytical note on LD-driven effect-size differences; European-derived PRS scored in five UKBB continental groups; and equal-sized UKBB and BBJ GWAS cross-applied to British, Japanese and African-descent targets (best of five p-value thresholds).
- GWAS Catalog records of participant ancestry, cumulative over time, compared with global population composition · human
- 1000 Genomes continental African, European and East Asian populations: allele frequencies of GWAS Catalog variants and pairwise LD (example region chr1 51572kb-52857 kb) · human
- UK Biobank targets in five continental ancestry groups scored with European-derived GWAS for 17 anthropometric and blood panel traits; unrelated, non-overlapping with discovery · human
- BioBank Japan (BBJ) Japanese GWAS for the same 17 traits and five diseases at matched sample size (hospital-based, disease-ascertained cohort); also Japanese targets · human
- UK Biobank British GWAS sized identically to BBJ for 17 quantitative traits and five disease endpoints (population-based, healthier-than-average cohort); also British targets · human
- African-descent British individuals in UK Biobank used as targets for both UKBB- and BBJ-derived polygenic scores · human
- Methods & tools
- GWAS Catalog, 1000 Genomes reference data, Genome-wide association studies with sample sizes matched between UKBB and BBJ, Polygenic risk scores, best of five p-value thresholds, Prediction R^2 with bootstrap 95% confidence intervals, Trans-ethnic genetic correlation of common-variant effect sizes, Heritability estimation (observed heritability per cohort), Linkage disequilibrium (r^2) comparison across populations, ICD-10 diagnosis codes
- Reproducibility gaps
- No code, data-availability statement or URLs appear in the text; sample sizes, per-trait R^2 and heritability estimates are deferred to supplementary tables and figures not provided here; access routes for UKBB and BBJ data are not described.
Limitations (6)
- BBJ is hospital-based and disease-ascertained while UKBB is a healthier-than-average population cohort, so cohort differences in phenotype precision and observed heritability, not ancestry, likely explain BBJ's lower quantitative-trait and higher disease accuracy. · stated by the authors
- GWAS Catalog participant tallies mix study availability, accessibility and reuse; cohorts analysed in many GWAS are counted repeatedly, which may disproportionately inflate European-descent representation. · stated by the authors
- Residual population stratification may bias cross-population accuracy by an unknown amount, and most PRS methods ignore recent admixture while none model admixed individuals' local ancestry mosaics, so further methods work is needed. · stated by the authors
- Part of the trait-to-trait fluctuation in relative accuracy across populations probably reflects differing heritability estimates and trans-ethnic genetic correlations, complicating comparisons of individual traits. · stated by the authors
- Accuracy is summarised as the maximum R^2 across five p-value thresholds with no independent tuning sample described, so absolute accuracies may be optimistic; LD-aware scoring such as LDPred was not reported for the new analyses. · noted in this summary
- Ancestry matching was tested only with British and Japanese discovery GWAS; no African or admixed discovery cohort was evaluated, and the five-ancestry fold-decreases cover quantitative traits only, not the diseases central to the clinical argument. · noted in this summary
Who should read:Read this if you develop, evaluate or plan clinical deployment of polygenic risk scores and need quantified cross-ancestry accuracy losses plus concrete arguments for diverse GWAS and open summary statistics.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- FairPRS: a fairness framework for Polygenic Risk Scores
Diego Machado Reyes, Aritra Bose, … Laxmi Parida·medRxiv·Aug 28, 2022
- Inclusion of Variants Discovered from Diverse Populations Improves Polygenic Risk Score Transferability
Taylor B. Cavazos, John S. Witte·Human Genetics and Genomics AdvancesPeer-reviewed·Dec 1, 2020
- Improving Polygenic Prediction in Ancestrally Diverse Populations
Yunfeng Ruan, Kuang Lin, … Tian Ge·Nature GeneticsPeer-reviewed·May 1, 2022
- Selection, optimization, and validation of ten chronic disease polygenic risk scores for clinical implementation in diverse populations
Niall J. Lennon, Leah Claire Kottyan, … Eimear E. Kenny·medRxiv·May 25, 2023
- Polygenic risk score portability for common diseases across genetically diverse populations
Sonia Moreno–Grau, Manvi Vernekar, … Carlos D. Bustamante·Human GenomicsPeer-reviewed·Sep 1, 2024
- Polygenic transcriptome risk scores improve portability of polygenic risk scores across ancestries
Yanyu Liang, Milton Pividori, … Hae Kyung Im·bioRxiv·Nov 12, 2020