Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults

Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults View ORCID Profile Luciana B. Vargas , View ORCID Profile Mariah C. Meyer , View ORCID Profile Iain R. Konigsberg , View ORCID Profile Aastha Kakar , View ORCID Profile Patrick M. Carry , Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium , View ORCID Profile Yun Li , View ORCID Profile Alana C. Jones , View ORCID Profile Hemant K. Tiwari , Vinodh Srinivasasainagendra , View ORCID Profile Nicole D. Armstrong , View ORCID Profile Eimear E. Kenny , View ORCID Profile Bogdan Pasaniuc , View ORCID Profile Marguerite R. Irvin , View ORCID Profile Michael H. Cho , View ORCID Profile Maggie A. Stanislawski , View ORCID Profile Sridharan Raghavan , View ORCID Profile Jonathan A. Shortt , View ORCID Profile Leslie A. Lange , View ORCID Profile Ethan M. Lange doi: https://doi.org/10.1101/2025.06.18.25329573 Luciana B. Vargas 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Luciana B. Vargas Mariah C. Meyer 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Mariah C. Meyer Iain R. Konigsberg 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Iain R. Konigsberg Aastha Kakar 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Aastha Kakar Patrick M. Carry 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA 2 Department of Orthopedics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA 3 Department of Epidemiology, School of Public Health, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Patrick M. Carry Yun Li 4 Department of Biostatistics and Department of Genetics, University of North Carolina at Chapel Hill , Chapel Hill, NC, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Yun Li Alana C. Jones 5 Medical Scientist Training Program, University of Alabama at Birmingham , Birmingham, AL, USA 6 Department of Epidemiology, School of Public Health, University of Alabama at Birmingham , Birmingham, AL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Alana C. Jones Hemant K. Tiwari 7 Department of Biostatistics, University of Alabama at Birmingham , Birmingham, AL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Hemant K. Tiwari Vinodh Srinivasasainagendra 7 Department of Biostatistics, University of Alabama at Birmingham , Birmingham, AL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Nicole D. Armstrong 5 Medical Scientist Training Program, University of Alabama at Birmingham , Birmingham, AL, USA 6 Department of Epidemiology, School of Public Health, University of Alabama at Birmingham , Birmingham, AL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nicole D. Armstrong Eimear E. Kenny 8 Institute for Genomic Health and Department of Genetics and Genomic Sciences, Icahn School of Medicine at Mount Sinai , New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Eimear E. Kenny Bogdan Pasaniuc 9 Department of Genetics, Perelman School of Medicine, University of Pennsylvania , Philadelphia, PA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Bogdan Pasaniuc Marguerite R. Irvin 5 Medical Scientist Training Program, University of Alabama at Birmingham , Birmingham, AL, USA 6 Department of Epidemiology, School of Public Health, University of Alabama at Birmingham , Birmingham, AL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Marguerite R. Irvin Michael H. Cho 10 Channing Division of Network Medicine and Division of Pulmonary and Critical Care Medicine, Brigham and Women’s Hospital , Boston, MA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Michael H. Cho Maggie A. Stanislawski 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Maggie A. Stanislawski Sridharan Raghavan 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA 11 Department of Veterans Affairs, Eastern Colorado Healthcare System , Aurora, CO, USA 12 Department of Medicine, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sridharan Raghavan Jonathan A. Shortt 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jonathan A. Shortt Leslie A. Lange 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA 3 Department of Epidemiology, School of Public Health, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Leslie A. Lange Ethan M. Lange 1 Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus , Aurora, CO, USA 13 Department of Biostatistics and Informatics, School of Public Health, University of Colorado Anschutz Medical Campus , Aurora, CO, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ethan M. Lange For correspondence: ethan.lange{at}cuanachutz.edu Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Polygenic risk score (PRS) distributions vary across populations, complicating PRS risk assessment. We evaluated the impact of post-hoc PRS calibration according to individualized genetic ancestry estimates on PRS performance using two large multi-ethnic PRS for type 2 diabetes (T2D) (PRS T2D ) and height (PRS height ), in 8,841 African American (AA) individuals from the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study. We calibrated each participant’s score as a function of estimated genetic similarity to the Yoruba (GSYRI) cohort in the 1000 Genomes Project. Uncalibrated PRSs were significantly skewed by GSYRI. After calibration, 33.6% of individuals in the top decile of PRS T2D were reclassified and performance in the top PRS T2D decile improved from an OR of 7.97 [6.31–10.13] to 10.77 [8.41– 13.91] when compared to the lowest decile. Similarly, 55.0% of individuals in the top PRS height decile were reclassified with GSYRI calibration. The calibrated PRS height showed higher correlation with height (from 0.24 to 0.32, p <10 −7 ), and increased mean height in the top PRS height decile ( p =5.7×10 −5 ) when compared to the uncalibrated PRS height . Lastly, we show that evaluating uncalibrated PRS while adjusting for GSYRI in regression models can lead to inflated and unstable effect size estimates for both the PRS and GSYRI. Introduction Polygenic risk scores (PRS), which aggregate risk estimates across genetic variants associated with a trait of interest, are powerful tools for identifying individuals at increased risk for disease. Individuals with a PRS score in the extreme upper tail of the corresponding PRS distribution can have a similar risk for disease as an individual carrying a rare high-penetrant mutation 1 . Differences in allele frequencies and linkage disequilibrium can result in shifts in the distributions of values for the same PRS both within, for genetically admixed individuals, and between populations, making it difficult to assess risk when combining different populations in the same study 2 – 7 . PRS calibration has been proposed as a solution to regress out the impact of ancestry on PRS. Recent methods for calibrating PRS have been proposed to address these distributional issues, often relying on standardization based on population labels, or, preferably, by regressing out genetic differences in allele frequencies (e.g., using principal components [PCs]) from these scores 8 , 9 . Here, we explore the implications of post-hoc calibration of PRS values according to individualized estimated ancestry proportions in self-identified African American (AA) individuals. We selected PRS functions for type 2 diabetes (T2D) and height, as these traits represent well-studied and common examples of complex human traits. Height is commonly used as a benchmark for genetic studies due to being a classic complex trait (i.e., influenced by thousands of genetic variants with small effect sizes) with broad availability and minimal phenotypic error across studies. Moreover, height is one of the most heritable human complex traits, with 80-90% of the variability in height estimated to be heritable 10 – 12 . PRS height explains over 90% of SNP-based heritability for height, the highest among all available PRS to date 12 . T2D, on the other hand, is a highly prevalent condition, affecting 11.1% of the global adult population 13 , and is a major risk factor for other complex diseases. For example, T2D patients have over 50% increased risk of developing cardiovascular disease and dementia compared to those without T2D 13 . We show that post-hoc calibration is important when assessing PRS performance (e.g., R 2 or AUC), when ranking individuals according to their PRS values, and when interpreting effect sizes of both estimated ancestry and PRS in relation to a phenotype. Methods We leveraged genetic data from 8,841 self-identified non-Hispanic Black study participants from the Reasons for Geographic and Racial Differences in Stroke (REGARDS) cohort. REGARDS is a U.S. national prospective cohort designed to study the reasons for higher incidence of stroke and related comorbidities in African American (AA) individuals and those living in the Southeastern U.S 14 . From the PGSCatalog 15 , we identified two recent PRS based on some of the largest multiethnic GWAS studies to date: one for type 2 diabetes (T2D) (PGS002308) 16 and one for height (PGS002802) 12 . Both PRS T2D and PRS height were developed including 1.1M and 5.3M samples of different ancestral backgrounds (although >75% of primarily European ancestry), and accounted for population structure by applying different meta-analysis approaches, PRS-CSx 17 (PRS T2D ) and SBayesC 18 (PRS height ). We evaluated the performance of PRS height in relation to height (inches), and PRS T2D in the context of T2D status and in relation to fasting glucose levels (mg/dL) in non-diabetic individuals. Further details can be found in Supplemental Materials . We characterized genetic diversity in REGARDS by estimating the fraction of an individual’s genome that comes from K hypothetical ancestral populations using ADMIXTURE (v. 1.3) 19 , 20 . Based on the prior knowledge that AA individuals have genetic admixture of primarily West African and European continental ancestry 21 , we predefined K=2 and used 1,000 Genomes reference panels (Yoruba in Ibadan, Nigeria [YRI] and Northern Europeans from Utah [CEU]) 22 to supervise ADMIXTURE estimation. Based on this method, each REGARDS individual was assigned two K fractions corresponding to the estimated proportion of genetic similarity to the YRI (GSYRI) and CEU reference panels ( Sup. Fig. 1 ). To validate our estimates, we additionally estimated genetic similarity by calculating principal components (PCs) derived from genetic data in PLINK (v2) 23 . We assessed GSYRI both continuously and by defining bins of genetically similar individuals: >90%, 80-90%, 70-80%, 60-70%, 50-60%, and <50% (the lower deciles (0-50%) were collapsed due to small sample size, N = 246). To remove the PRS and GSYRI correlation, we calibrated each individual’s PRS value by regressing out their GSYRI and dividing the residuals by the standard deviation of the PRS values for all participants within the same GSYRI-defined bin (to account for the possibility that variance in the PRS is also a function of ancestry proportions). This PRS calibration by genetic similarity can be represented as: Where i is REGARDS individual i ; d is the GSYRI decile grouping for individual i ; x i is the observed unscaled PRS for person i ; μ i is the predicted PRS for person i (based on regression model of PRS ~ GSYRI ); and σ d is the estimated standard deviation of PRS scores for REGARDS individuals in GSYRI decile d . To validate our approach, we compared PRS height residuals after regressing out either i) GSYRI, ii) PCs 1-10, or iii) using the Polygenic Score Catalog Calculator 24 ancestry normalization tool ( pgsc_calc ) ( Sup. Fig. 2 ). The pgsc_calc tool performs PC projections based on the 1,000 Genomes sample space and continuously adjusts PRS values based on differences in the mean and variance of the PRS across the PC sample space 9 , 25 . We compared overall model prediction and effect estimates for calibrated and uncalibrated PRS through a series of linear (height, glucose) and logistic (T2D) regression models. All models were adjusted for age and sex. T2D and glucose models were further adjusted for body mass index (BMI). For each outcome and each PRS (uncalibrated or calibrated), models were constructed including a) basic covariates, b) basic covariates and GSYRI, c) basic covariates and PRS, d) basic covariates, GSYRI and PRS, and e) basic covariates, GSYRI, PRS, and an interaction between GSYRI and PRS. Results and Discussion Individuals in REGARDS that self-identified as Black have a median GSYRI of 84.35% [IQR: 75.68 - 90.63%] ( Sup. Fig. 1 and Sup. Table 1 ), which agrees with other AA groups previously described 26 , 27 . GSYRI and the first PC are highly correlated (r = 0.9989, p 0.25, p 0.98, p < 10 −16 ) ( Sup. Fig. 2 ). Thus, we opted to use GSYRI in our analyses, as this measure is more interpretable than PCs and has been independently shown to perform similarly to correction based on PCs 28 . Download figure Open in new tab Figure 1. Impact of ancestry-informed calibration on PRS distribution and prediction. Distribution curves for A) PRS T2D and B) PRS height across African Americans (AA) individuals from the REGARDS cohort, stratified according to GSYRI, before and after PRS calibration. *At-risk was defined as individuals in the upper 10% of PRS distribution. C) Point estimates and confidence intervals of the odds ratio (OR) of type 2 diabetes (T2D) when comparing each decile of the PRS distribution to the lowest decile, for the uncalibrated PRS T2D (yellow) and the calibrated PRS T2D (blue). D) Violin plots showing the distribution of height (after adjusting for age and sex) across deciles of the PRS height distribution, when using the uncalibrated PRS height (yellow) and the calibrated PRS height (blue). It has been suggested that the reason for PRS distributional differences across populations is primarily due to differences in allele frequencies and linkage disequilibrium patters across populations 5 , 6 , 29 . An alternative explanation, however, is that real differences in genetic architecture of traits in different populations lead to the observed PRS shifts. We investigated these hypotheses by studying the association of the outcomes included in our study (height, T2D, and glucose) and GSYRI. Height and glucose levels were not associated with GSYRI (after adjusting for sex, age, [and BMI in glucose models], p > 0.05) while T2D was positively associated with GSYRI (after adjusting for sex, age and BMI, β = 1.10, p < 10 −07 ). The latter observation has been previously reported in the literature, and it remains unclear whether the association is due to genetic differences between groups, or due to confounding environmental differences, such as social determinants of health (SDoH) 30 – 35 . We further adjusted T2D models for three SDoH variables (income, education, and medical care), and the association between T2D and GSYRI was attenuated ( β = 0.70, p = 0.01). When stratified by sex, the association between T2D and GSYRI attenuated more strongly in males ( β = 0.52, p = 0.13) than females ( β = 0.85, p = 0.0034), consistent with previous reports 36 . These results suggest that, in our study, shifts in the PRS distributions associated with GSYRI are primarily due to population allele frequency differences rather than genetically driven differences in height or T2D between individuals with different GSYRI profiles. Ancestry Calibration Improves Risk Stratification Based on observed values in the upper tail of the PRS distributions, the uncalibrated PRS functions either systematically under- or over-estimated risk for AA individuals with lower estimated GSYRI relative to the remaining AA individuals ( Fig. 1A-B ) . For PRS T2D , all AA individuals with ≦50% GSYRI were systematically excluded from the upper 10% of the uncalibrated PRS distribution “at-risk” classification (none had a PRS in the upper decile), and most (over 99%) of the individuals with ≦70% GSYRI were also excluded. The opposite shift was observed for PRS height , where the majority (79.5%) of individuals with ≦50% GSYRI were placed at the top decile of the uncalibrated PRS height ( Fig. 1B ) , although height itself is not associated with GSYRI ( p > 0.05). After calibration of PRS T2D , 33.6% of individuals in the top decile of PRS T2D distribution were reclassified. We found a higher median percentile change in PRS T2D rank with calibration in the subset of individuals previously excluded from risk stratification (≦70% GSYRI) compared to the remaining individuals (>70% GSYRI) (26.7 increase versus 4.5 decrease, respectively, p < 0.001) ( Sup. Table 2 and Sup. Fig. 4 ). Calibration increased the odds ratio (OR) of T2D from 7.97 [95% CI:6.31-10.13] when using the uncalibrated PRS T2D to 10.77 [95% CI: 8.41-13.91] using the calibrated PRS T2D ( p = 0.032), when comparing dichotomized PRS (upper 10% vs. lower 10%) ( Fig. 1C ) . This reclassification also resulted in higher correlation with glucose levels (after adjusting for sex and age) in non-diabetics, from 0.083 for the uncalibrated PRS T2D ( p = 2.4 × 10 −9 ) to 0.10 for the calibrated PRS T2D ( p = 2.2 × 10 −13 ) ( Sup. Fig. 3A ). After calibration of PRS height , 55.0% of individuals in the top decile of the uncalibrated PRS distribution were reclassified using the newly calibrated PRS rankings. Individuals ranked in the top decile of the calibrated PRS height had significantly higher median height as compared to the uncalibrated PRS height (median residual height percentiles: 64.3 [IQR: 38.2-84.5] and 71.7 [IQR: 46.7-88.6], p = 5.7 × 10 −5 ) ( Fig. 1D ). Moreover, the correlation between PRS height and height (after adjusting for age and sex) increased from 0.24 to 0.32 after GSYRI calibration ( p < 10 −7 ) ( Sup. Fig. 3B ). As expected, we observed that individuals with greater levels of genetic admixture had greater percentile rank changes between the PRS height distributions before and after the calibration. Individuals with ≦70% GSYRI had a median 28.0 percentile decrease in PRS height after calibration, compared to a median of 4.2 percentile increase in the remaining individuals (>70% GSYRI) ( p < 0.001) ( Sup. Table 2 and Sup. Fig. 5 ). Ancestry: Covariate Adjustment versus Calibration on Effect Estimates It is common practice to model genetic similarity (e.g., GSYRI or PCs) as a covariate when evaluating the performance of uncalibrated PRS values to account for possible confounding between genetic similarity and the PRS. We demonstrate that, for some PRS functions, this practice alone does not remedy biased PRS effect estimates in admixed groups, such as AA individuals. We found similar overall model performance (AUC/R 2 ) between prediction models including calibrated or uncalibrated PRS when also including GSYRI ( Table 1 and Sup. Table 3). However, we observe biased parameter estimates for both GSYRI and the uncalibrated PRS when jointly including those in the same model. For example, GSYRI was not associated with height in models adjusted for age and sex ( β = 0.21 ± 0.23, p = 0.37). However, when including the uncalibrated PRS, GSYRI became strongly associated with height ( β = 5.69 ± 0.28, p = 8.4 × 10 −88 )( Table 1 ). Furthermore, the effect estimates for the uncalibrated PRS changed considerably depending on whether GSYRI was or was not included in the model (from β =0.68 ± 0.03 to β =1.13 ± 0.04). In contrast, when including the calibrated PRS in the model along with GSYRI, the effect estimate for GSYRI did not change, remaining nonsignificant, and the effect estimate for the calibrated PRS also was not impacted by the inclusion of GSYRI. A similar pattern of results was observed for PRS T2D and glucose levels ( Sup. Table 3 ). Of note, GSYRI was associated with T2D before inclusion of the PRS T2D in the models ( β = 1.10 ± 0.21, p = 7.5 × 10 −8 ). However, the direction of the GSYRI effect changed when including the uncalibrated PRS T2D in the model (from β =1.10 ± 0.21 to β = - 2.68 ± 0.26), whereas the effect estimates for GSYRI remained similar when including the calibrated PRS in the model ( Table 1 ). We demonstrate that, for some PRS functions, covariate adjustment for uncalibrated PRS estimates results in biased effect estimates, and that post-hoc calibration of PRS based on GSYRI results in more stable estimates. View this table: View inline View popup Download powerpoint Table 1. Change in effect sizes of PRS height and PRS T2D with the inclusion of estimated genetic ancestry (GSYRI) as a covariate in the models. T2D, type 2 diabetes. β, effect size of one unit change in the outcome. s.e., standard error. VIF PRS , variance inflation factor for the PRS in the model. R 2 , fraction of variance explained by the model. AUC, area under the receiver operating characteristic (ROC) curve. For interaction models (noted by the inclusion of the interaction term GSYRI x PRS), parameter estimates are given for each of the main effects (GSYRI or PRS) and the interaction effect (GSYRI x PRS). The reason for unstable effect estimates in models that include uncalibrated PRS and GSYRI is not clear. We found low to moderate multicollinearity between GSYRI and the uncalibrated PRS values (variance inflation factor, VIF=[1, 1.5]), which indicates that the correlation between PRS and GSYRI (| r | > 0.25) does not fully explain the unreliability of the estimates observed. For height and T2D, we found no significant evidence supporting an interaction between GSYRI and PRS values in models including the calibrated PRS, but did find such evidence for modest interaction effects using the uncalibrated PRS ( Table 1 ). We found no evidence of an interaction between GSYRI and PRS in models with glucose level as the outcome ( Sup. Table 3 ). An alternative hypothesis to explain the unreliability of models with the uncalibrated PRS and GSYRI would be that there is a source of collider bias that is removed by directly regressing out the correlation between PRS and GSYRI (i.e., calibration) before inclusion in the regression models, although further investigation would be needed to confirm this. In conclusion , this work demonstrates that post-hoc calibration of PRS scores using a continuous estimate of genetic similarity (e.g., GSYRI or PCs) is needed for more accurate interpretation and clinical utility of PRS in AAs. Although we show that overall model performance (as measured by R 2 or AUC) when including both GSYRI and continuous PRS in prediction models is similar for both calibrated and uncalibrated PRS, there was an instability of coefficient estimates for both the PRS and GSYRI when using uncalibrated PRS. Moreover, identification of “at-risk” individuals, as would be applied in a clinical setting, differed substantially with calibration. This observation highlights the need to disentangle the impact of genetic similarity on PRS to improve interpretation of PRS as a stand-alone risk factor in genetically admixed populations. It is critical to highlight that the calibration does not fully address the current differences in PRS performance across the genetic ancestry continuum, and there are several other limitations that must be addressed before clinical implementation of PRS is optimized in genetically admixed individuals that were not included in the scope of this work 37 . Existing PRS calibration methods rely on broadly regressing out measures of global ancestry (e.g., principal components or admixture estimates), and do not account for local admixture patterns to inform PRS adjustment. Other limitations to clinical utility of PRS include underrepresentation of non-European ancestry individuals in GWAS, which lead to biased effect estimates when applying PRS developed in predominantly European populations to other populations, in addition to missing important population-specific variants that increase genetic risk in non-European populations 2,3,5,29,37 . PRIMED Consortium The members of the Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium are Clement Adebamowo, Sally Adebamowo, Adebowale Adeyemo, Nicholette Allred, Paul Auer, Jennifer Below, Palwende Romuald Boua, Kristin Boulier, Michael Bowers, Joseph Breeyear, Nilanjan Chatterjee, Brian Chen, Tinashe Chikowore, Jaewon Choi, Ananyo Choudhury, Matthew P. Conomos, David Conti, Nancy Cox, Sinead Cullina, Burcu Darst, Aaron Deutsch, Yi Ding, Todd Edwards, Eleazar Eskin, Segun Fatumo, Jose Florez, Nelson Freimer, Stephanie Fullerton, Tian Ge, Daniel Geschwind, Chris Gignoux, Stephanie Gogarten, Mark Goodarzi, Xiuqing Guo, Christopher Haiman, Neil Hanchard, Scott Hazelhurst, Ben Heavner, Susan Heckbert, Jibril Hirbo, Whitney Hornsby, Kangcheng Hou, Qinqin Huang, Alicia Huerta, Guoqian Jiang, Katherine Johnston, Linda Kachuri, Takashi Kadowaki, Abram Bunya Kamiza, Eimear Kenny, Sarah Kerns, Alyna Khan, Elena Kharitonova, Joohyun Kim, Iain Konigsberg, Charles Kooperberg, Matt Kosel, Peter Kraft, Iftikhar Kullo, Soo-Heon Kwak, Ethan Lange, Leslie Lange, Loic Le Marchand, Hyunsuk Lee, Aaron Leong, Yun Li, Meng Lin, Kirk Lohmueller, Ruth Loos, Yingchang Lu, Ravi Mandla, Alisa Manning, Iman Martin, Alicia Martin, Hilary Martin, Rasika Mathias, James Meigs, Josep Mercader, Rachel Mester, Mariah Meyer, Tyne Miller-Fleming, Braxton Mitchell, Nicola Mulder, Jie Na, Pradeep Natarajan, Sarah C. Nelson, Maggie Ng, Kristján Norland, Franklin Ockerman, Loes Olde Loohuis, Suna Onengut-Gumuscu, Ebuka Onyenobi, Roel Ophoff, Paivi Pajukanta, Bogdan Pasaniuc, Bogdan Pasaniuc, Aniruddh Patel, Ulrike Peters, Jimmy Phuong, Michael Preuss, Laura Raffield, Michèle Ramsay, Alexander Reiner, Kenneth Rice, Stephen Rich, Jerome Rotter, Bryce Rowan, Robb Rowley, Yunfeng Ruan, Lori Sakoda, Sriram Sankararaman, Dan Schaid, Dan Schrider, Philip Schroeder, Ruhollah Shemirani, Jonathan Shortt, Megan Shuey, Xueling Sim, Roelof A.J. Smit, Johanna Smith, Lucia Sobrin, Lauren Stalbow, Adrienne Stilp, Daniel Stram, Quan Sun, Ken Suzuki, Lukasz Szczerbinski, Ran Tao, Bamidele Tayo, Timothy Thornton, Buu Truong, Teresa Tusié, Miriam Udler, David van Heel, Luciana B. Vargas, Vidhya Venkateswaran, Ying Wang, Jia Wen, Jennifer Wessel, Laura Wiley, Lynne Wilkens, Riley Wilson, John Witte, Genevieve Wojcik, Quenna Wong, Toshimasa Yamauchi, Lisa Yanek, Yue Yu, Yuji Zhang, Haoyu Zhang, and Michael Zhong. Funding Funding for this work was supported by the National Institutes of Health for the project ‘Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium’, with grant funding for Study Site CAPE (U01HG011715) and the Coordinating Center (U01HG011697). M.H.C. was supported by R01HL168199, and R01HL153248, and R01HL135142. This work utilized the Alpine high performance computing resource at the University of Colorado Boulder 38 . Alpine is jointly funded by the University of Colorado Boulder, the University of Colorado Anschutz, Colorado State University, and the National Science Foundation (award 2201538 and 2322260). Data storage supported by the University of Colorado Boulder ‘PetaLibrary’ 39 . Data availability REGARDS data are not publicly available due to ethical and legal restrictions. To abide by its obligations with NIH/NINDS and the Institutional Review Board of the University of Alabama at Birmingham, REGARDS facilitates data sharing through data use agreements. Any investigator is welcome to access the REGARDS data, including statistical code, through this process. Requests for data access may be sent to the REGARDS Executive Committee at regardsadmin{at}uab.edu . More information can be found at https://www.uab.edu/soph/regardsstudy/ . Acknowledgements This work utilized data generated by the Reasons for Geographic and Racial Differences in Stroke (REGARDS) project. This research project is supported by cooperative agreement U01 NS041588 co-funded by the National Institute of Neurological Disorders and Stroke (NINDS) and the National Institute on Aging (NIA), National Institutes of Health, Department of Health and Human Service. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NINDS or the NIA. Representatives of the NINDS were involved in the review of the manuscript but were not directly involved in the collection, management, analysis or interpretation of the data. The authors thank the other investigators, the staff, and the participants of the REGARDS study for their valuable contributions. A full list of participating REGARDS investigators and institutions can be found at: https://www.uab.edu/soph/regardsstudy/ . The REGARDS study protocol was reviewed and approved by the Institutional Review Board at the University of Alabama at Birmingham. All participants provided written informed consent, and all data used for these analyses were previously de-identified. Footnotes ↵ * A list of PRIMED banner authors appears at the end of the paper. References 1. ↵ Khera , A. V. et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations . Nat. Genet . 50 , 1219 – 1224 ( 2018 ). OpenUrl CrossRef PubMed 2. ↵ Reisberg , S. , Iljasenko , T. , Läll , K. , Fischer , K. & Vilo , J. Comparing distributions of polygenic risk scores of type 2 diabetes and coronary heart disease within different populations . PLoS ONE 12 , ( 2017 ). 3. Ding , Y. et al. Polygenic scoring accuracy varies across the genetic ancestry continuum . Nature 618 , 774 – 781 ( 2023 ). OpenUrl CrossRef PubMed 4. Carlson , C. S. et al. Generalization and Dilution of Association Results from European GWAS in Populations of Non-European Ancestry: The PAGE Study . PLoS Biol . 11 , e1001661 ( 2013 ). OpenUrl CrossRef PubMed 5. ↵ Wang , Y. et al. Theoretical and empirical quantification of the accuracy of polygenic scores in ancestry divergent populations . Nat. Commun . 11 , 3865 ( 2020 ). OpenUrl CrossRef PubMed 6. ↵ Hou , K. et al. Causal effects on complex traits are similar for common variants across segments of different continental ancestries within admixed individuals . Nat. Genet . 55 , 549 – 558 ( 2023 ). OpenUrl CrossRef PubMed 7. ↵ Martin , A. R. et al. Human Demographic History Impacts Genetic Risk Prediction across Diverse Populations . Am. J. Hum. Genet . 100 , 635 – 649 ( 2017 ). OpenUrl CrossRef PubMed 8. ↵ Khan , A. et al. Genome-wide polygenic score to predict chronic kidney disease across ancestries . Nat. Med . 28 , 1412 – 1420 ( 2022 ). OpenUrl CrossRef PubMed 9. ↵ Khera , A. V. et al. Whole-Genome Sequencing to Characterize Monogenic and Polygenic Contributions in Patients Hospitalized With Early-Onset Myocardial Infarction . Circulation 139 , 1593 – 1602 ( 2019 ). OpenUrl CrossRef PubMed 10. ↵ Silventoinen , K. et al. Heritability of adult body height: a comparative study of twin cohorts in eight countries . Twin Res. Off. J. Int. Soc. Twin Stud . 6 , 399 – 408 ( 2003 ). OpenUrl 11. Polderman , T. J. C. et al. Meta-analysis of the heritability of human traits based on fifty years of twin studies . Nat. Genet . 47 , 702 – 709 ( 2015 ). OpenUrl CrossRef PubMed 12. ↵ Yengo , L. et al. A saturated map of common genetic variants associated with human height . Nature 610 , 704 – 712 ( 2022 ). OpenUrl CrossRef PubMed 13. ↵ International Diabetes Federation . IDF Diabetes Atlas, 11th Edition . https://diabetesatlas.org ( 2025 ). 14. ↵ Howard , V. J. et al. The Reasons for Geographic and Racial Differences in Stroke Study: Objectives and Design . Neuroepidemiology 25 , 135 – 143 ( 2005 ). OpenUrl CrossRef PubMed Web of Science 15. ↵ Lambert , S. A. et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation . Nat. Genet . 53 , 420 – 425 ( 2021 ). OpenUrl CrossRef PubMed 16. ↵ Ge , T. et al. Development and validation of a trans-ancestry polygenic risk score for type 2 diabetes in diverse populations . Genome Med . 14 , 70 ( 2022 ). OpenUrl CrossRef PubMed 17. ↵ Ruan , Y. et al. Improving polygenic prediction in ancestrally diverse populations . Nat. Genet . 54 , 573 – 580 ( 2022 ). OpenUrl CrossRef PubMed 18. ↵ Lloyd-Jones , L. R. et al. Improved polygenic prediction by Bayesian multiple regression on summary statistics . Nat. Commun . 10 , 5086 ( 2019 ). OpenUrl CrossRef PubMed 19. ↵ Alexander , D. H. & Lange , K. Enhancements to the ADMIXTURE algorithm for individual ancestry estimation . BMC Bioinformatics 12 , 246 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 20. ↵ Alexander , D. H. , Novembre , J. & Lange , K. Fast model-based estimation of ancestry in unrelated individuals . Genome Res . 19 , 1655 – 1664 ( 2009 ). OpenUrl Abstract / FREE Full Text 21. ↵ All of Us Research Program Genomics Investigators . Genomic data in the All of Us Research Program . Nature 627 , 340 – 346 ( 2024 ). OpenUrl CrossRef PubMed 22. ↵ The 1000 Genomes Project Consortium . A global reference for human genetic variation . Nature 526 , 68 – 74 ( 2015 ). OpenUrl CrossRef PubMed 23. ↵ Chang , C. C. et al. Second-generation PLINK: rising to the challenge of larger and richer datasets . GigaScience 4 , 7 ( 2015 ). OpenUrl CrossRef PubMed 24. ↵ Lambert , S. A. et al. Enhancing the Polygenic Score Catalog with tools for score calculation and ancestry normalization . Nat. Genet . 56 , 1989 – 1994 ( 2024 ). OpenUrl CrossRef PubMed 25. ↵ Khan , A. et al. Genome-wide polygenic score to predict chronic kidney disease across ancestries . Nat. Med . 28 , 1412 – 1420 ( 2022 ). OpenUrl CrossRef PubMed 26. ↵ Bryc , K. , Durand , E. Y. , Macpherson , J. M. , Reich , D. & Mountain , J. L. The Genetic Ancestry of African Americans, Latinos, and European Americans across the United States . Am. J. Hum. Genet . 96 , 37 – 53 ( 2015 ). OpenUrl CrossRef PubMed 27. ↵ Zakharia , F. et al. Characterizing the admixed African ancestry of African Americans . Genome Biol . 10 , R141 ( 2009 ). OpenUrl CrossRef PubMed 28. ↵ Rosenthal , E. A. et al. Comparing Ancestry Standardization Approaches for a Transancestry Colorectal Cancer Polygenic Risk Score . Genet. Epidemiol . 49 , e22590 ( 2025 ). OpenUrl PubMed 29. ↵ Wang , Y. et al. Polygenic prediction across populations is influenced by ancestry, genetic architecture, and methodology . Cell Genomics 3 , 100408 ( 2023 ). OpenUrl PubMed 30. ↵ Robbins , J. M. Excess type 2 diabetes in African-American women and men aged 40-74 and socioeconomic status: evidence from the Third National Health and Nutrition Examination Survey . J. Epidemiol. Community Health 54 , 839 – 845 ( 2000 ). OpenUrl Abstract / FREE Full Text 31. Lam , V. et al. Ancestry-attenuated effects of socioeconomic deprivation on type 2 diabetes disparities in the All of Us cohort . BMC Glob. Public Health 1 , 22 ( 2023 ). OpenUrl PubMed 32. Cheng , C.-Y. et al. African ancestry and its correlation to type 2 diabetes in African Americans: a genetic admixture analysis in three U.S. population cohorts . PloS One 7 , e32840 ( 2012 ). OpenUrl CrossRef PubMed 33. Nagar , S. D. , Nápoles , A. M. , Jordan , I. K. & Mariño-Ramírez , L. Socioeconomic deprivation and genetic ancestry interact to modify type 2 diabetes ethnic disparities in the United Kingdom . EClinicalMedicine 37 , 100960 ( 2021 ). OpenUrl PubMed 34. Florez , J. C. et al. Strong association of socioeconomic status with genetic ancestry in Latinos: implications for admixture studies of type 2 diabetes . Diabetologia 52 , 1528 – 1536 ( 2009 ). OpenUrl CrossRef PubMed Web of Science 35. ↵ Chande , A. T. et al. Influence of genetic ancestry and socioeconomic status on type 2 diabetes in the diverse Colombian populations of Chocó and Antioquia . Sci. Rep . 7 , 17127 ( 2017 ). OpenUrl PubMed 36. ↵ Tang , M. , Chen , Y. & Krewski , D. Gender-related differences in the association between socioeconomic status and self-reported diabetes . Int. J. Epidemiol . 32 , 381 – 385 ( 2003 ). OpenUrl CrossRef PubMed Web of Science 37. ↵ Kachuri , L. et al. Principles and methods for transferring polygenic risk scores across global populations . Nat. Rev. Genet . 25 , 8 – 25 ( 2024 ). OpenUrl CrossRef PubMed 38. ↵ University of Colorado Boulder Research Computing . Alpine . doi: 10.25811/k3w6-pk81 ( 2023 ). OpenUrl CrossRef 39. ↵ University of Colorado Boulder Research Computing . PetaLibrary . doi: 10.25811/81nc-wv41 ( 2021 ). OpenUrl CrossRef View the discussion thread. Back to top Previous Next Posted June 18, 2025. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults Luciana B. Vargas , Mariah C. Meyer , Iain R. Konigsberg , Aastha Kakar , Patrick M. Carry , Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium , Yun Li , Alana C. Jones , Hemant K. Tiwari , Vinodh Srinivasasainagendra , Nicole D. Armstrong , Eimear E. Kenny , Bogdan Pasaniuc , Marguerite R. Irvin , Michael H. Cho , Maggie A. Stanislawski , Sridharan Raghavan , Jonathan A. Shortt , Leslie A. Lange , Ethan M. Lange medRxiv 2025.06.18.25329573; doi: https://doi.org/10.1101/2025.06.18.25329573 Share This Article: Copy Citation Tools Ancestry Calibration of Polygenic Risk Scores Improves Risk Stratification and Effect Estimation in African American Adults Luciana B. Vargas , Mariah C. Meyer , Iain R. Konigsberg , Aastha Kakar , Patrick M. Carry , Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium , Yun Li , Alana C. Jones , Hemant K. Tiwari , Vinodh Srinivasasainagendra , Nicole D. Armstrong , Eimear E. Kenny , Bogdan Pasaniuc , Marguerite R. Irvin , Michael H. Cho , Maggie A. Stanislawski , Sridharan Raghavan , Jonathan A. Shortt , Leslie A. Lange , Ethan M. Lange medRxiv 2025.06.18.25329573; doi: https://doi.org/10.1101/2025.06.18.25329573 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genetic and Genomic Medicine Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4435) Dentistry and Oral Medicine (444) Dermatology (382) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1509) Epidemiology (15227) Forensic Medicine (30) Gastroenterology (1124) Genetic and Genomic Medicine (6597) Geriatric Medicine (668) Health Economics (997) Health Informatics (4534) Health Policy (1368) Health Systems and Quality Improvement (1613) Hematology (540) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15916) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (146) Nephrology (667) Neurology (6599) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1144) Occupational and Environmental Health (957) Oncology (3332) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (691) Primary Care Research (711) Psychiatry and Clinical Psychology (5447) Public and Global Health (9230) Radiology and Imaging (2198) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a004370dbcc7593a',t:'MTc3OTU0MDkyNw=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();