Analysis of left ventricular mass for African American individuals using multi-ancestry polygenic risk scores: the Jackson Heart Study Authors: Douglas McLaurin, Philip Turk, Olivia Affuso, Laura M Raffield, Elizabeth Heitman Published: 2025 ABSTRACT AbstractBackgroundLeft ventricular mass (LVM) is suggested to be a sensitive predictor of adverse cardiovascular outcomes, such as heart failure. In recent years, genome-wide association studies have discovered loci that associate with outcomes related to LVM, providing an opportunity for the development of genetic risk scores. However, the relevance of these genetic variants to non-European ancestry groups requires additional testing. Here, we examined if variants that have been associated with heart failure and LVM in multi-ancestry populations are associated with LVM among African American individuals in the Jackson Heart Study (JHS).MethodsHeart failure and LVM associated variants were identified from two published multi-ancestry studies. Two polygenic risk scores (PRSs) were computed for 2175 African American participants (mean age 53, 63% female). We assessed the linear association of both PRSs with LVM indexed to height (LVMh) and indexed to body surface area (LVMbsa) and fit a multivariate general linear model to LVM containing both PRS and covariates (age, sex, body mass index (BMI), diabetes and hypertension). Type III MANOVA Pillai tests were run to assess the effects of the PRS on the log LVM values.ResultsLinear correlation analysis showed positive associations between age and LVMh (r=0.278) and LVMbsa (r=0.296) as well as BMI with LVMh (r=0.320). A strong linear correlation was observed between LVMh and LVMbsa (r=0.894). Elevated LVM among individuals with diabetes and hypertension was observed. When accounting for age, sex, BMI, diabetes and hypertension, we found insufficient evidence to suggest that the heart failure PRS affected either measure of LVM; the same can be said for the LVM PRS.ConclusionWe find insufficient evidence to suggest that heart failure and LVM genetic variants derived from predominantly European multi-ancestry populations are linearly associated with log LVM among African American participants in the JHS.Keywords: Genetic Association Studies, Heart Failure, Risk Factors, Epidemiology WHAT IS ALREADY KNOWN ON THIS TOPIC Conducting multi-ancestry genome-wide association studies has been advocated to improve the discovery of genetic variants that are relevant to more than just European ancestral groups. However, Europeans often predominate in these multi-ancestry studies, raising the question of the generalisability of these recently discovered genetic variants. Conducting multi-ancestry genome-wide association studies has been advocated to improve the discovery of genetic variants that are relevant to more than just European ancestral groups. However, Europeans often predominate in these multi-ancestry studies, raising the question of the generalisability of these recently discovered genetic variants. WHAT THIS STUDY ADDS When assessing left ventricular mass in an African American cohort, we find insufficient evidence of an association between genetic variants that were previously identified in predominately European, multi-ancestry studies of left ventricular mass and heart failure. When assessing left ventricular mass in an African American cohort, we find insufficient evidence of an association between genetic variants that were previously identified in predominately European, multi-ancestry studies of left ventricular mass and heart failure. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Our findings add to the growing body of knowledge indicating a greater need for the recruitment of and targeted genetic studies in African ancestral cohorts to improve the discovery of genetic variants that may contribute to cardiovascular conditions that impact this disproportionately affected community. Our findings add to the growing body of knowledge indicating a greater need for the recruitment of and targeted genetic studies in African ancestral cohorts to improve the discovery of genetic variants that may contribute to cardiovascular conditions that impact this disproportionately affected community. Introduction The left ventricle is a heart structure that pumps oxygenated blood to the rest of the body and is subject to remodelling induced by various stimuli.1,3 One form of remodelling involves a change in the left ventricular mass (LVM), which can be estimated by echocardiography (ECHO).4,6 Increases in LVM have been associated with ageing and body size.7,11 Diseases such as diabetes and hypertension have also been linked with increased LVM.12,15 When indexed by body size, LVM is a useful diagnostic marker for left ventricular hypertrophy, a heart condition characterised by an abnormal increase in LVM.16 17 Left ventricular hypertrophy manifests from excess workload on the heart and is a physiological response to stimuli such as intense exercise, which can result in what is often called ‘athlete’s heart’.18 19 However, it can also develop under pathological conditions such as the chronic pressure induced by hypertension.20 21 This pathological form of left ventricular hypertrophy has been associated with an increased risk of heart failure.22,24 As such, LVM is thought to be a sensitive predictor of adverse cardiovascular outcomes.14 25 African American populations have been found to experience a disproportionate burden of cardiovascular disease.26,28 For pathological left ventricular hypertrophy, African American individuals are reported to experience a higher prevalence when compared with the majority population.2629,31 The cause behind elevated LVM among African American adults is enigmatic, but may be partly explained by environmental factors, such as chronic stress26 32 33 or health factors, such as the prevalence of hypertension.34,36 Of growing interest is the influence of genetic variants that have been associated with complex conditions, such as elevated LVM.37,41 Despite this increased burden of disease, African American adults, as well as other populations with significant African ancestry, have been underrepresented in many genetic studies of LVM and related outcomes, such as heart failure.42,45 Polygenic risk scores (PRSs) have emerged as a useful tool for predicting individual disease risk using aggregates of genetic variants that have been identified using genome-wide association studies (GWAS).46,48 However, a major limitation to the application of PRS is the minimal representation of diverse ancestry in source GWAS used to derive these variants.42,45 This under-representation has resulted in inconsistent results for PRS in non-European ancestry populations when compared with their European counterparts, sparking concerns regarding healthcare disparities if PRSs were to be used clinically.49,51 Greater efforts to conduct multi-ancestry GWAS have been attempted, but European ancestry still largely predominates in these studies.42 52 A PRS that could identify individuals most at risk of experiencing increases in LVM would be extremely useful to improving health outcomes, particularly among groups that are disproportionately burdened, such as African Americans. However, the uncertainty of the broad applicability of genetic variants that have recently been identified is a concern. In this study, we examined the linear relationship between genetic risk scores derived from variants that have been associated with heart failure or LVM in European-centric, multi-ancestry studies and measurements of LVM in the African American cohort of the Jackson Heart Study (JHS). Methods Study cohort The JHS is a community-based, prospective epidemiologic study of cardiovascular disease among African American individuals. At the baseline examination (2000–2004) the JHS recruited 5306 men and women between the ages of 21 and 95 living in the Jackson, Mississippi metropolitan area, consisting of Hinds, Madison and Rankin counties. JHS has conducted two follow-up examinations (2005–2008 and 2009–2013). A third follow-up examination is ongoing. All data presented in this study were drawn from extensive phenotypic data collected during the baseline examination (2000–2004).53 Data presented in this study were accessed on 27 July 2023. The work presented in this study was approved for investigation and publication by the JHS presentation and publications committee under project number P1814. This study was reviewed and approved by the University of Mississippi Medical Center Human Research Office under IRB number UMMC-IRB-2023–145. All JHS participants gave written informed consent.54 See the JHS design and methodology report for more information on the informed consent process.54 Anthropometric and chronic disease measures Pertinent characteristics at the baseline examination include: (1) presence of diabetes, defined as fasting glucose ≥126 mg/dL, or use of diabetic medication (actual or self-reported) within 2 weeks prior to the clinic visit; (2) HbA1c ≥6.5%; (3) presence of hypertension, defined as the use of blood pressure lowering medication or blood pressure >140/90 mm Hg; and (4) resting blood pressure measurements taken while sitting using a Hawksley random zero sphygmomanometer with one of four cuff sizes selected by measured arm circumference. Details of variable measurement and collection and of the overall design of the JHS protocol have been described elsewhere.53 Echocardiographic measures Echocardiograms were performed using a Sonos 4500 (Hewlett Packard) ultrasound machine. Final measurements were performed offline by one cardiologist reader who had level 3 training in ECHO (TS). Cardiac dimensions were measured according to the American Society of Echocardiography (ASE) guidelines. LVM was calculated from the ASE corrected formula. To correct for body size, two LVM indexes were calculated as either (1) LVM (g) divided by height2.7 (g/m2.7), denoted as LVMh, or (2) LVM divided by body surface area (g/m2), denoted as LVMbsa. Genotyping Whole genome sequencing of JHS participants was performed as a part of the National Heart, Lung, and Blood Institute’s Trans-Omics for Precision Medicine programme.55 Genetic risk scores PRSs are quantitative measurements of an individual’s estimated genetic risk of developing a trait, which can be calculated based on quantifying the effect of many single nucleotide polymorphisms (SNPs) that have been linked to a trait of interest.56 SNPs for this study were selected from two previously published studies. In one study, 47 genome-wide significant heart-failure loci were identified in multi-ancestry GWAS where individuals of African ancestry comprised less than 15% of the study group.39 In the second study, 12 genome-wide significant LVM associated loci were identified in multi-ancestry GWAS where self-reported Black individuals made up less than <2% of the study cohort.40 Neither of these prior GWAS studies included JHS. Due to non-identification of several variant SNPs within the sequencing data in the JHS cohort (potentially due to either low variant quality which did not pass sequencing quality control or due to no presence of alternate allele in this sample size), PRSs for either heart failure (PRShf) or LVM (PRSlvm) were calculated using 39 heart failure-associated variants39 (online supplemental Table S1) and 10 LVM-associated variants40 (online supplemental Table S2), respectively, and the associated effect sizes determined in their respective prior GWAS. PLINK 1.9 software was then used to compute PRSs for JHS participants using the score function.57 Statistical analysis From the initial study cohort (n=5306), 1906 participants were excluded from analysis due to missing LVM index data (n=3400), and a further 1218 participants were removed due to missing genotyping data. Seven participants had missing diabetes values and were also removed. A comparison was made with respect to both LVM values between participants that had and did not have PRS values using graphics, summary statistics and t-tests. No difference was noticed between these two groups. Therefore, 2175 complete cases were used for modelling. For an initial exploratory data analysis, summary statistics and graphics were generated for the continuous variables age, body mass index (BMI), PRShf, PRSlvm, LVMbsa and LVMh and the categorical variables sex, diabetes and hypertension. Due to the right skew observed for both LVM measurements (figure 1), a log transformation was conducted prior to fitting the model. A multivariate general linear model was fit to both LVM measurements (LVMbsa and LVMh) containing predictors PRShf and PRSlvm; the covariates age and BMI; and the factors sex, diabetes and hypertension. The presence of any multicollinearity among these explanatory variables was assessed using variance inflation factors. No multicollinearity was noted. Matrix scatterplot illustrating pairwise linear correlation analyses using the Pearson correlation coefficient among LVM measurements, PRS and continuous LVM predictors, age and BMI. BMI, body mass index, LVM, left ventricular mass; PRS, polygenic risk score. Since LVMbsa and LVMh capture the same underlying LVM construct and both indexes are strongly correlated (figure 1), we selected a multivariate general linear model to test the association between LVM and the genetic risk scores. Multivariate analysis of variance (MANOVA) directly tests the joint null hypothesis (for the union-intersection test) that the PRS has no association with either outcome while accounting for their covariance, which (i) controls type I error without ad hoc multiple-testing corrections for two separate GLMs across highly dependent outcomes and (ii) preserves or increases power when effects are aligned across correlated responses. Type III MANOVA Pillai tests were run to assess the effects of the explanatory variables. Approximate F-statistics were derived from the Pillai test statistic. P values were adjusted for multiplicity of hypothesis tests according to Holm’s step-down method. To conduct the sensitivity analysis, a linear mixed model was employed between either log LVM measurements and PRS, age, sex, BMI, diabetes, hypertension, the top five ancestry principal components and family ID as covariates. Principal components were generated using the principal component analysis in related (PC-AiR)58 samples approach. Statistical analyses were performed using R V.4.3.1 (R Project for Statistical Computing). Results Study cohort characteristics and exploratory data analysis At baseline, the JHS recruited 5306 participants. Among those participants, a total of 2175 participants had the LVM measurements and genetic data required to pursue this study. A description of the relevant demographic, anthropometric and echocardiographic characteristics of the study cohort is provided in table 1. This study group consisted of individuals with a mean (± SD) age of 52.6±12.6 years, BMI of 31.5±7.0 kg/m2, LVMh of 36.2±9.8 g/m2.7, LVMbsa of 75.0±18.7 g/m2 and 52% were observed to be hypertensive. Table 1Characteristics of the study cohortCharacteristicResultn (% sample of 2175) Female1360 (63%) Hypertension1125 (52%) Diabetes471 (22%)Mean (SD) Age, years52.6 (12.6) Body mass index, kg/m231.5 (7.0) Left ventricular mass indexed by height36.2 (9.8) Left ventricular mass indexed by BSA75.0 (18.7)BSA, body surface area. Characteristics of the study cohort BSA, body surface area. Our first analysis of this cohort explored the relationships between our study variables with a particular interest in associations between LVM and the explanatory variables. For the continuous variables, we applied naïve linear correlation analysis with Pearson correlation coefficients for age, BMI, PRS and LVM measures (figure 1). Our results show moderate positive correlations between age and both LVMh (r=0.278) and LVMbsa (r=0.296). We also observe a positive correlation for BMI with LVMh (r=0.320). To be expected, a strong positive correlation was also observed between LVMh and LVMbsa (r=0.894). We also note that the distributions for LVMbsa and LVMh were right-skewed, while the distributions for PRShf and PRSlvm were approximately normal. For the categorical variables, we assessed the distributions of LVM measurements by sex, diabetes and hypertension (online supplemental Figure S1). We found an apparent increase in LVMh and LVMbsa among individuals with diabetes and hypertension compared with individuals who did not experience those conditions. For sex, there was also increased LVMbsa for males compared with females, and no notable difference between sexes with respect to LVMh. Multivariate analysis for LVM After exploring the relationships among our variables and concluding that the associations between LVM and the explanatory variables coincide with published literature, we moved forward with our primary analysis. To correct for the skewness of the LVM indexes, these variables were log-transformed. Because both indexes measure the same outcome, LVM, and have a strong correlation (r=0.894), both log LVMh and log LVMbsa were fit to a multivariate general linear model that included the explanatory variables (table 2). For both log LVMh and log LVMbsa, significant effects were observed for age (F=49.0; p<1.5 x 10−15), sex (F=972.4; p<1.5 x 10−15), diabetes (F=4.7; p=0.028), hypertension (F=46.7; p<1.5 x 10−15) and BMI (F=3787.8; p<1.5 x 10−15). However, there was insufficient evidence to suggest PRShf had an effect on either log LVMbsa or log LVMh, accounting for the other explanatory variables in the model (F=1.6; p=0.401). The same can be said of PRSlvm (F=0.2; p-val=0.833). Table 2Multivariate analysis of variance table for log LVMh and log LVMbsaVariables F P valueAge49.0 <1.5e-15 Sex972.4 <1.5e-15 Diabetes4.7 0.028 Hypertension46.7 <1.5e-15 BMI3787.8 <1.5e-15 PRShf1.60.401PRSlvm0.20.833Approximate F derived from Pillai test statistic. P value represents adjusted p value according to Holm’s step-down method. Note reference F distribution has two numerator df and 2166 denominator df. P values in bold are considered statistically significant (p <0.05).BMI, body mass index; PRS, polygenic risk score. Multivariate analysis of variance table for log LVMh and log LVMbsa Approximate F derived from Pillai test statistic. P value represents adjusted p value according to Holm’s step-down method. Note reference F distribution has two numerator df and 2166 denominator df. P values in bold are considered statistically significant (p <0.05). BMI, body mass index; PRS, polygenic risk score. Following the MANOVA, we conducted a post-hoc sensitivity analysis using a linear mixed model that examined the association between the log LVM measurements and PRS individually (table 3). For this analysis, we also adjusted for genetic ancestry using five principal components and family random effects. Similar to our previous results, we also observed no statistically significant association between either LVM measurement and either PRS through this sensitivity analysis (table 3). Table 3Sensitivity analysis adjusted with genetic ancestry principal components and family structurelog LVMPRSEstimateSEtP valueLVMhPRSlvm0.0580.1280.4520.651LVMhPRShf3.8792.6921.4410.150LMVbsaPRSlvm0.0650.1220.5310.595LMVbsaPRShf2.9222.5831.1310.258LVM indexes were log-transformed and fit to a linear mixed model including PRS, sex, BMI, diabetes, hypertension, five genetic ancestry principal components and family structure.BMI, body mass index; PRS, polygenic risk score. Sensitivity analysis adjusted with genetic ancestry principal components and family structure LVM indexes were log-transformed and fit to a linear mixed model including PRS, sex, BMI, diabetes, hypertension, five genetic ancestry principal components and family structure. BMI, body mass index; PRS, polygenic risk score. Discussion PRSs are a relatively new and innovative technology that has the potential to enhance risk prediction models and move our society closer toward the promise of precision medicine.59 For example, in a breast cancer PRS, it was observed that women in the top 10% of the distribution have a lifetime risk of breast cancer that was comparable to the risk of individuals with pathogenic mutations in CHEK2 and ATM genes.48 An indexed LVM PRS, comprised of 465 variants, has been shown to associate with higher risk of incident atrial fibrillation, HF, ventricular arrhythmias, dilated cardiomyopathy and implantable cardioverter defibrillator.40 Among individuals receiving hypertension treatment, an increased risk of experiencing HF was associated with elevated scores in a HF PRS that was derived from 12 genetic variants.60 However, a major point of concern regarding the use of PRS is their generalisability towards those of non-European ancestry, as the genetic variants that are being discovered and applied to these PRS calculations often come from studies that lack diversity in ancestry.61 In this study, we sought to determine if heart failure or LVM-associated variant loci that have been identified in predominantly European multi-ancestry populations are associated with LVM within the African American cohort in the JHS. We derived PRSs for JHS participants based on heart failure or LVM-associated variants and looked for associations with two indexes of LVM and included LVM covariates and factors age, BMI, sex, diabetes and hypertension in our model. Consistent with published literature, we observed statistically significant results when assessing the relationships between the latter and LVM. However, no statistically significant associations were observed between either PRS or LVM. In the exploratory analysis, when assessing unadjusted correlations among LVM, PRS and the continuous covariates age and BMI, we found moderate positive associations with age and both LVM indexes, and between BMI and LVMh (figure 1). It is well reported that structural remodelling of the left ventricle is inextricably linked with ageing.7 While there is debate surrounding the link between ageing and actual changes in LVM, the consensus appears to be that the traditional means of calculating LVM is positively associated with increasing age.8 BMI was positively correlated with LVMh, while an approximate correlation coefficient of zero (−0.054) was observed with LVMbsa. Regarding this lack of BMI association with LVMbsa, previous studies have pointed out that LVMbsa poorly estimates changes in LVM among overweight/obese individuals, particularly when compared with the performance of LVMh.962,64 This study cohort consists of an average BMI of 31.5, which falls within the range of obesity according to standard BMI classifications and thus might explain our results. To be expected, we also observed a strong positive correlation between LVMh and LVMbsa. Other categorical variables including sex, diabetes and hypertension were also assessed through side-by-side box plots of LVM indexed values within each category level (online supplemental Figure S1). We did not observe a difference between males and females in LVMh. However, for LVMbsa, males were elevated over females. This may be related to findings suggesting that indexing to height may not sufficiently account for sex differences that can lead to miscalculations of LVM.9 Similarly, for both LVM indexes, we observed an elevation among participants with diabetes and hypertension compared with those without either condition, consistent with published work.12 13 We fit a multivariate general linear model to the log transformation of both LVM measurements and then used MANOVA hypothesis tests to determine the effect of each variable on LVM (table 2). Our linear model found that age, sex, diabetes, hypertension and BMI were each important with respect to their effect on LVM, while neither PRS was significantly associated with either LVMhf or LVMbsa. Based on the exploratory data analysis, these results are not surprising, and our sensitivity analysis also supported these findings. The associations with the non-genetic explanatory variables are consistent with published work that finds that LVM increases with age8 and BMI65 and that individuals with diabetes12 and hypertension13 are at a greater risk of increases in LVM. We find insufficient evidence of an association between these specific genetic variants and LVM in the JHS. There are a few explanations for why we have observed these results in our study that are incongruent with the results of the referenced studies that these variants are obtained from.39 40 According to general practices, PRS analyses should involve a sample size of at least 100 individuals.57 However, such a sample size may not be sufficient to detect a more modest PRS effect. While the PRS developed in this study is sufficient by these general terms, it has been suggested that PRS performance can be improved with increased sample size.66 When taking this into consideration, plus the fact that the reference studies from which we sampled the variant loci included over 43 000 participants for the LVM study and over 100 000 for the heart failure study, it is possible that this study cohort of 2175 participants was not sufficient to replicate the findings from the reference studies. Also, the sizeable difference in genetic ancestry composition between the referenced studies in which these variants were collected and our cohort in the JHS may play a significant role. The heart failure and LVM studies consisted of less than 15% African ancestry and less than 2% black individuals, respectively. We also note that the referenced LVM GWAS used cardiac magnetic resonance (CMR) imaging to estimate LVM versus the ECHO estimates used in JHS. As it has been suggested that ECHO overestimates LVM when compared with CMR,67,69 the technical differences in these two techniques may have contributed to how the genetic variants collected from the reference GWAS associated with LVM in this study. Heart failure GWAS variants may also exert their effects through intermediate phenotypes other than LVM. Finally, the prevalence of health factors such as hypertension and obesity within this cohort could simply outweigh the genetic effect of these variants. Both conditions are associated with overload of the heart, with hypertension being associated with pressure overload and obesity with volume overload.70 71 Irrespective of a genetic predisposition, the presence of health factors such as these can stimulate cardiac remodelling and induce changes in LVM. Limitations There are a few limitations within our statistical approach that should be acknowledged. First, our approach for this study followed a parsimonious model that sought to focus on key variables of interest and avoid model expansion. In doing so, we did not include certain environmental explanatory variables that are reported to associate with LVM such as smoking72,74 and alcohol use.75 76 We acknowledge that the inclusion of these variables would provide an opportunity to gain more insight from our statistical models, especially in relation to the non-genetic explanatory variables. A second limitation of our statistical approach is the decision to pursue a complete case analysis rather than attempting imputation of the missing LVM measurements. In preparation for this study, a comparison was done with respect to both LVMs between subjects that had and did not have PRS values using graphics, summary statistics and t-tests (data not shown). No difference was noticed between these two groups. Thus, while the analysis was restricted to complete cases, the risk of bias introduced by exclusion appears minimal. Multiple imputation was not pursued for the missing genetic data because genotype information is not missing at random but structurally absent for participants who did not undergo genotyping. In addition, imputation of LVMs would have required more nuanced outcome modelling that could obscure true associations. However, we acknowledge that the missing data may impact the generalisability of the results of this study. Third, the multivariate general linear model, the primary analysis tool for this study, did not include principal components of genetic ancestry or family random effects. The JHS cohort is a relatively homogenous population of African American individuals located in the Jackson, MS metropolitan area. While there is obviously genetic ancestry variability within the participants of the JHS, our assumption was that any effect from genetic ancestry differences among this population would have no sizeable impact on the association with LVM. To address our assumption, we included five genetic ancestry PCs, computed using PC-AIR,58 in the sensitivity analysis and observed no change in our overall results. Heart failure is a complex, heterogenous condition with numerous risk factors.77 In the referenced study by Levin et al,39 the authors applied a broad ‘all-cause’ definition of HF to identify the genetic variants. It is possible that these genetic variants may be too varied in their correlative HF causes, which is highlighted by the authors in the noted pleiotropic effect of certain loci.39 Thus, this variability in associated traits may have weakened possible associations due to changes in LVM, which might explain why we observe no association with PRShf. The influence of genetic ancestry on the applicability of PRS across diverse groups has been an important topic of discussion in recent years.49 It is possible that assessing the European-derived variants in this study to an African American cohort might yield different results compared with analyses in a similar predominantly European multi-ancestry population. Additionally, the number of variants applied to GRS has also been noted as an issue in the predictive performance.47 Our study applied a very modest number of variants. It is possible that a true PRS containing hundreds of variants may have performed better. Ultimately, additional PRS models should be examined in future work with larger cohorts. Finally, the influence of environmental factors such as socioeconomic status (SES) and stress on LVM cannot be overstated.3578,82 Chronic stress and lower SES have been implicated in higher incidences of hypertension,79 80 which is itself a leading cause of increases in LVM.78 These relationships can also be found within the JHS as stress and lower SES have been linked with an increased prevalence of hypertension.35 82 Thus, for a trait such as LVM, which is linked closely with hypertension, it is possible that the contribution of effects from the selected genetic variants was simply outweighed by the environment. Additionally, as mentioned previously, factors such as smoking and alcohol use are of notable interest and should also be considered. Future studies that investigate the genetic risk for changes in LVM should assess the interaction between genetics and environmental factors. Conclusion Collectively, we find insufficient evidence to suggest that these specific heart failure and LVM-associated variants identified in predominantly European multi-ancestry populations are associated with LVM among African American adults in the JHS, though modest impacts cannot be excluded in our relatively small cohort. Future studies may be able to extend this work by optimising the number of genetic variants applied to PRS calculations and incorporating specific environmental factors into analyses. Future GWAS in African ancestry populations for LVM and heart failure may also improve available GWAS training data for PRS scores. Supplementary material Acknowledgements The Jackson Heart Study (JHS) is supported and conducted in collaboration with the Jackson State University (HHSN268201800013I), the Tougaloo College (HHSN268201800014I), the Mississippi State Department of Health (HHSN268201800015I) and the University of Mississippi Medical Center (HHSN268201800010I, HHSN268201800011I and HHSN268201800012I) contracts from the National Heart, Lung and Blood Institute (NHLBI) and the National Institute on Minority Health and Health Disparities (NIMHD). The authors also wish to thank the staff and participants of the JHS. The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the National Heart, Lung and Blood Institute; the National Institutes of Health; or the US Department of Health and Human Services. Footnotes Funding: This study was supported by the National Institutes of Health – National Heart, Lung, and Blood Institute (T32HL105324). Provenance and peer review: Not commissioned; externally peer-reviewed. Patient consent for publication: Not applicable. Ethics approval: This study involved human participants and was approved for investigation and publication by the Jackson Heart Study presentation and publications committee under project number P1814. This study was reviewed and approved by the University of Mississippi Medical Center Human Research Office under IRB number - UMMC-IRB-2023-145. Participants gave informed consent to participate in the study before taking part. Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information. References Associated Data Supplementary Materials Data Availability Statement All data relevant to the study are included in the article or uploaded as supplementary information.