Blood Pressure Polygenic Score Predicts Long-Term Blood Pressure Control and Treatment-Resistant Hypertension.
Jan 14, 2026DOI: 10.1161/hypertensionaha.125.26399Journal ArticleAHA
Written by: So Mi Jemma ChoYunfeng RuanHyeok-Hee LeeSatoshi KoyamaStephen P JuraschekNorrina B AllenEugene YangJohn W McEvoyEric A SecemskyMichael C HonigbergAkl C FahedAniruddh P PatelWhitney E HornsbyPradeep Natarajan
High blood pressure (BP) is a major driver of cardiovascular disease (CVD)-related mortality and disability worldwide [1]. As such, sustained achievement of optimal BP is fundamental for CVD prevention. Observational studies demonstrate a log-linear association of both systolic and diastolic BP (SBP & DBP) with CVD even among young adults [2] or individuals without modifiable risk factors such as hypertension, dyslipidemia, diabetes, and current tobacco use [3]. With clinical trials [4–7] establishing lower rates of major adverse cardiovascular events from intensive SBP lowering below <120 mmHg compared to the standard (<140 mmHg) treatment, contemporary clinical guidelines [8,9] have adopted a starkly lower BP target (<130/80 mmHg) for the general population as well as individuals with concurrent cardiorenal conditions.Nevertheless, suboptimal BP management remains highly prevalent and imposes substantial risk for CVD events [10]. The extent of BP control significantly varies across geographic region, demographics, and comorbidities [11]. Based on the 2017–2020 National Health and Nutrition Examination Survey, U.S. adults prescribed with antihypertensives had varying rates of adequate BP control across sociodemographics [12]. In addition, the prevalence of treatment-resistant hypertension was significantly higher among individuals with concurrent cardiometabolic comorbidities including diabetes mellitus (48.1% versus 23.5% in normoglycemia), chronic kidney disease (50.8% versus 26.5% in normal kidney function), and CVD (40.8% versus 18.4% in without) [13]. However, environmental, contextual, clinical, and diurnal influences contribute to substantial intraindividual measurement variability [14], thereby challenge timely hypertension detection and treatment. Furthermore, rare or ancestry-specific genetic variants modify BP-associated pathophysiological pathways, including mineralocorticoids and glucocorticoids production, β−1 adrenergic receptor encoding, natriuretic peptides clearance, and conversion of angiotensin II to angiotensin III [15. However, whether commonly-occurring genetic variations are associated with prolonged suboptimal BP management is incompletely understood.Given the increasing consensus for rigorous BP management yet pervasively suboptimal control, identifying novel risk factors influencing BP control may further optimize timing and course of BP-lowering treatment. Prior literature has illustrated that genetically-predicted BP independently predicts midlife measured BP and provides incremental discrimination ability for lifetime CVD prediction beyond lifestyle and clinical factors [16]. Nevertheless, whether BP genetic risk is associated with long-term BP stability remains unknown. Understanding the associative and prognostic utility of genetic tools in predicting long-term BP control and treatment-resistant hypertension may improve identification of high-risk individuals warranting earlier and intensive preventive strategies. Here, we examined whether a BP polygenic score (PGS) independently predicts longitudinal BP control and lifetime incidence of treatment-resistant hypertension in adults with hypertension.
Methods
Per the Health insurance Portability and Accountability Act, requests to access the Mass General Brigham Biobank data from qualified researchers trained in human subject confidentiality protocols may be sent to Mass General Brigham Biobank at https://biobank@partners.org. UK Biobank data are available to researchers by application at https://ukbiobank.dnanexus.com.
The Mass General Brigham Biobank is a volunteer healthcare-based cohort comprising eight tertiary care and affiliated community hospitals in New England [17]. Through genotyping and linkage to electronic health records, its objective is to study clinical, genetic, lifestyle, and socioenvironmental risk factors on health outcomes and disease pathophysiology. The protocol was approved by the Massachusetts General Hospital institutional review board (2018P001236). Informed consent was provided by all participants.
To account for temporal changes in BP clinical guidelines, we assessed long-term BP control after the adoption of the 2017 American College of Cardiology/American Heart Association BP guidelines [9] endorsing uniform SBP/DBP thresholds of <130/80 mmHg irrespective of demographic characteristics or underlying comorbidities. We identified 22,456 participants 1) aged 18–79 years; 2) with physician-diagnosed hypertension; 3) at least 3 distinct outpatient BP measurements with less than 2 years gap between each encounter; and 4) without heart failure with reduced ejection fraction, or end-stage renal disease at the beginning of assessment period (Supplemental Methods 1, Figure S1).
Validation analyses were performed on a healthier, population-based UK Biobank observational cohort. The UK Biobank is a prospective cohort study of 0.5 million adults aged 40 to 69 years at recruitment between 2006 and 2010 [18]. Among 151,747 participants with linkage to general practitioner data and genotype information, we identified 27,856 participants with hypertension prior to enrollment (Figure S2). The UK Biobank study protocol was approved by the North West Multi-centre Research Ethics Committee (11/NW/0382) and the secondary data usage (UK Biobank application #7089) was approved by the Massachusetts General Hospital Institutional Review Board (2021P002228). All participants provided signed informed consent. The present study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.
Details on central quality control, imputation of genotypic data, and BP PGS construction and predictive performance are elaborated in [17,18] and Supplemental Methods 2. Briefly, PGSs were derived from multiancestral genome-wide association study (GWAS) results [19–23] based on >1.2 million Hapmap3 variants. SNPs were extracted from the GWAS data and processed with the LDpred2 framework [24]. Given high genetic diversity observed, we applied the Genetic Distance-assisted PRS Combination Pipeline for Diverse Genetic Ancestries [25] framework to generate a harmonized BP PGS for all individuals via linear interpolation. Here, we leveraged multiple PGS weights that were fine-tuned within single ancestry samples and the genetic ancestry continuum information to yield high-accuracy estimation. The scores were calculated based on the four sets of SNP effect sizes (AFR, EAS, EUR, and SAS), regressed on the first 10 principal components, and rescaled to mean of 0 and SD of 1. The combination weights were derived based on the genetic distance, computed via principal component analysis between the target individual and the AFR, EAS, EUR and SAS fine-tuning datasets. Fine-tuning to generate BP PGSs in Mass General Brigham Biobank was performed on UK Biobank samples and vice versa.
In the Mass General Brigham Biobank, covariates were extracted from electronic health records 1 year prior and the most proximate to the beginning of 5-year BP management assessment period (index date). Blood biochemistry and anthropometry measurements were collected during routine outpatient encounters. History of atherosclerotic CVD was recorded for participants with diagnostic or procedural codes pertaining to myocardial infarction, stroke, peripheral artery disease, percutaneous coronary intervention, or coronary artery bypass grafting.
In the UK Biobank, covariates were measured at the cohort enrollment (index date). Current smoking was defined as lifetime smoking of at least 100 cigarettes and without indication of cessation. Body mass index was measured using Tanita BC-418MA body composition analyzer (Tanita, Tokyo, Japan). Blood biochemistry was assayed within 24 hours of non-fasting sample collection.
BPs were measured using automated oscillometric or aneroid auscultatory devices used during routine care; BPs collected during invasive procedures, ambulatory BP monitoring, or pregnancy were excluded. To avoid introducing spurious readings, we excluded measurements with SBP ≥240 mmHg or <60 mmHg, or DBP ≥160 mmHg or <30 mmHg [26].
Two primary outcomes were studied. First, the time above target BP was calculated by re-zeroing the interpolated BP values above SBP ≥130 mmHg and DBP ≥80 mmHg, separately, such that only the duration spent above the thresholds contribute to the cumulative BP burden (Figure 1). We performed linear interpolation on all BP measurements to connect the measured values throughout the 5-year BP control assessment period. Graphically, the time above target represents the sum of duration lived between SBP readings ≥130 mmHg and 130 mmHg or DBP readings ≥80 mmHg and 80 mmHg. Second, apparent treatment-resistant hypertension was defined as either 1) having SBP/DBP ≥140/90 mmHg despite the concurrent use of three or more antihypertensive agents of different classes, including a long-acting calcium channel blocker, renin-angiotensin system inhibitor, and a diuretic at maximally tolerated dose; 2) use of four or more antihypertensives medication classes; or 3) International Classification of Diseases-10 of I1A.0 [27]. We estimated lifetime cumulative incidence of treatment-resistant hypertension.
Three secondary outcomes were examined. First, time-averaged BP was assessed to account for differences in measurement interval by averaging the cumulative BP from adjacent consecutive healthcare visits indexed to the 5-year assessment period. Second, cumulative BP level above target was calculated by annualizing the sum of the difference between ≥130 mmHg and 130 mmHg or ≥80 mmHg and 80 mmHg thresholds. Third, we categorized and summed the different antihypertensive medication classes.
To reflect differences in clinical practice, the validation analyses using the UK Biobank applied SBP/DBP thresholds of 140/90 mmHg in adherence to the 2006 National Institute for Health and Clinical Excellence hypertension guidelines in align with the UK Biobank enrollment era [28]. Enrollment BP was measured on two consecutive occasions with 1 minute interval using Omron 705 IT electronic BP monitor (OMRON Healthcare Europe, Hoofddorp, Netherlands) after 5 minutes of seated rest; office BP measurements differed by local primary care practices.
The βs or odds ratios (ORs) [95% CIs] for time above target BP and the cumulative incidence of treatment-resistant hypertension were calculated across decile and low (<20th percentile; reference), intermediate (20-<80th percentile) and high (≥80th percentile) SBP and DBP PGS groups of normalized PGS using one-way analysis of variance tests. In parallel, the continuous association was also examined per 1 standard deviation increase in BP PGS using multivariable linear or logistic regression. All models were adjusted for age, sex, first 10 principal components, genotyping array, self-identified race and ethnicity, measured BPs, body mass index, total and high-density lipoprotein cholesterols, estimated glomerular filtration rate, cigarette smoking, lipid-lowering medication, diabetes mellitus, history of CVD, and number of antihypertensives medication classes (except in the case of treatment-resistant hypertension as the outcome) as a time-varying covariate. Given PGS has different explainability for a trait in the context of primary or secondary CVD prevention, comorbidities, and demographics [29,30], we performed sensitivity stratified analyses by age, sex, genetic ancestry, number of healthcare utilization with BP readings, number of antihypertensive classes, and history of cardiometabolic diseases. Given varied definitions of drug-resistant hypertension across clinical guidelines, we further assessed the associations in align with the 2024 European Society of Cardiology [31].
As long-term BP management contributes to CVD, we further assessed the associations between BP PGSs and lifetime risk of cardiovascular-kidney-metabolic syndrome defined as diabetes mellitus, chronic kidney disease, atherosclerotic CVD, and heart failure among participants without subtype-specific condition prior to the assessment period using Cox proportional hazards models adjusted for the covariates above.
To assess the clinical utility of BP PGS in enhancing prediction of long-term BP management, we quantified improvements in discrimination, model fit, and net reclassification made by genetic score. From the Base Model (demographics, lifestyle, clinical risk factors, comorbidities, and medications), we compared improvements in Harrell’s C-index by adding BP PGS. Improvements in discrimination was estimated based on a two-sided Z test after bootstrap resampling of 1000 iterations. Likelihood ratio tests determined whether the inclusion of BP PGS significantly improved the model goodness-of-fit. Improvements in Brier score captured the calibration and discrimination ability of BP PGSs in predicting long-term BP outcomes. Continuous net reclassification index quantified whether addition of BP PGS yields clinically relevant improvements in prediction.
All statistical tests were two-sided, and statistical significance was set at P <0.05. All analyses were performed using R version 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria).
Results
The study included 22,456 Mass General Brigham Biobank participants (mean [SD] age, 63.9 [13.5] years; 10,853 [48.3%] females) with hypertension presenting a mean [SD] SBP of 131.5 [17.7] mmHg and DBP of 75.4 [10.6] mmHg at the beginning of 5-year assessment window (Table 1, Tables S1&S2). Participants contributed a median [IQR] of 7.6 [3.6–13.2] BP readings annually with a median [IQR] of 48.0 [27.7–101.4] days between each reading (Figure S3). The UK Biobank cohort had a lower mean [SD] age of 61.5 [6.4] years but with higher mean SBP [SD] of 143.7 [18.6] mmHg and DBP [SD] of 83.3 [10.3] mmHg at enrollment (Tables S3&S4). Accounting for antihypertensive effects, each SD increase in SBP and DBP PGS was associated with cross-sectionally measured index date SBP [standard error] of 1.86 [0.12] mmHg and DBP of 0.97 [0.07] mmHg in the Mass General Brigham Biobank and 2.95 [0.11] mmHg and 1.46 [0.06] mmHg in the UK Biobank, respectively (Table S5).
The mean duration lived with above target SBP was incrementally longer across higher BP genetic risk as represented by the highest PGS decile. This resulted, on average, as living with above target SBP 51.1% of time compared to the 40.9% in the lowest decile (P<0.001) (Figure 2). Similarly, the highest DBP PGS decile had a significantly longer time with above target DBP compared to the lowest decile (32.3% vs. 20.8%; P<0.001). In parallel, the proportion of participants achieving target BP for the entirety of the assessment period was incrementally lower across higher BP PGS decile (P for trend <0.001) (Table S6&S7). In secondary analyses, the lowest BP PGS decile had significantly lower mean time-averaged SBP [SD] of 128.2 [12.3] mmHg and DBP [SD] of 72.9 [12.3] mmHg compared to the highest decile (time-averaged SBP 132.7 [12.8] mmHg and DBP 76.2 [12.8] mmHg; P<0.001) (Figures S4&S5). Over a lifetime, the highest SBP PGS decile had a 2-fold higher likelihood of treatment-resistant hypertension compared to the lowest PGS group (24.7 % vs. 12.8%) and a higher proportion of individuals requiring triple or higher antihypertensive therapy (25.4% vs. 41.6%; P<0.001).
BP PGS was independently associated with long-term BP stability and treatment-resistant hypertension (Figure 3A, Table S8). In reference to low SBP PGS group, the intermediate and high SBP PGS groups conferred 3.21 [95% CI 2.13–4.29]% and 8.01 [95% CI 6.68–9.34]% longer duration spent with above target SBP, respectively. Simultaneously, higher SBP PGS associated with higher time-averaged SBP level (per SD increase, 1.41 [0.08] mmHg) and cumulative SBP above target (3.34 [95% CI 2.53–3.85] mmHg/year). Similarly, the high DBP PGS group was also associated with 6.19 [95% CI 5.05–7.33]% longer duration lived with above target DBP. In align, the high SBP PGS group associated with 11.19 [95% CI 10.03–12.35]% longer duration spent with above target SBP ≥140 mmHg in the UK Biobank cohort (Figure 3B, Table S9).
Compared to Mass General Brigham Biobank participants with low BP PGS, the high SBP PGS group had 2.36 [95% CI 2.07–2.68]-fold greater odds for lifetime development of treatment-resistant hypertension, whereas the high DBP PGS strata exhibited 1.75 [95% CI 1.55–1.99]-fold higher odds. As such, the ORs [95% CIs] for requiring triple or more intensive antihypertensive therapy were 1.31 [1.27–1.36] in high SBP and 1.22 [1.18–1.26] in high DBP PGS categories compared to the respective low PGS groups.
Subgroup analyses (Tables S10–S28) showed that high SBP PGS is consistently associated with poor long-term hypertension control without significant heterogeneity by age (P=0.122) nor antihypertensive treatment intensity and regimen (P=0.145). Furthermore, high genetically-determined BP associated with treatment-resistant hypertension irrespective of across sex, healthcare utilization, underlying comorbidities, and varying clinical guidelines in both the Mass General Brigham Biobank and UK Biobank.
As such, BP PGSs were significantly associated with higher lifetime risk of cardiovascular-kidney-metabolic outcomes independent of measured BPs and phenotypic risk factors (Figures S6–S9). In the Mass General Brigham Biobank, relative to the low PGS group, the high SBP genetic risk group was associated with HR [95% CI] of 1.36 [1.23–1.50] for chronic kidney disease, 1.19 [1.12–1.27] for atherosclerotic CVD, and 1.18 [1.09–1.28] for heart failure. Likewise in UK Biobank, the high SBP PGS group conferred 1.11-, 1.14-, and 1.18-fold higher risk for chronic kidney disease, atherosclerotic CVD, and heart failure, respectively.
Table 2 and Figure S10 illustrates the discrimination and net reclassification abilities of BP genetic scores beyond phenotypic risk factors. In addition to demographic, clinical, and lifestyle factors, SBP PGS enhanced the prediction of uncontrolled SBP from C-index [95% CI] 0.82 [0.81–0.83] to 0.84 [0.84–0.85] (ΔC-index [0.02–0.03]) and net reclassification index [95% CI] by 0.02 [0.01–0.03]. Furthermore, BP PGSs significantly improved discrimination for treatment-resistant hypertension from C-index [95% CI] of 0.73 [0.72–0.74] to 0.76 [0.76–0.78] (ΔC-index [0.03–0.04]). BP PGSs consistently enhanced discrimination, calibration, and reclassification abilities for all long-term BP burden and treatment indices in the UK Biobank (Table S29, Figure S11).
Discussion
In large prospective healthcare- and population-based cohorts of adults with hypertension, BP PGS was independently associated with and improved prediction of long-term BP control, cumulative BP burden, treatment intensity, and treatment-resistant hypertension regardless of age or history of cardiorenal diseases. Since contemporary clinical guidelines underscore sustained achievement of target BP for primary and secondary prevention of CVD, our findings permit several conclusions regarding the clinical utility of polygenic assessment toward optimal hypertension management.
First, genetic predisposition to elevated BP is associated with extent of guideline-recommended target BP level achievement and cumulative BP burden over 5 years. These findings extend prior observations on how BP-associated genetic alleles independently associate with incident hypertension [32] and exhibit greater predictability for CVD [16,33]. A cross-sectional population surveillance data identified sociodemographic (current or former marriage, consistent access to the same healthcare provider) and lifestyle (dietary sodium reduction, physical activity, weight management) factors associated with better BP control in U.S. adults with hypertension [34]. We further observed a role for BP PGS to predict longitudinal BP control beyond established phenotypic risk factors using real-world healthcare utilization data and minimizing practical challenges of high BP measurement variability [35]. Given that 1) the risk of cardiovascular events increases log-linearly even below SBP of 120 mmHg [36] and 2) elevated BP is implicated with high population attributable risk across primordial [3] to secondary [37] CVD prevention, BP PGSs may augment clinical actionability for hypertension management—especially for high-risk patients warranting more vigilant monitoring and frequent risk re-assessments towards optimal BP preservation.
Second, BP PGSs may augment prioritization of patients benefiting from earlier or more intensive BP-lowering therapy. Current hypertension guideline acknowledges rare, monogenic forms of hypertension (i.e., glucocorticoid-remediable aldosteronism, Liddle’s syndrome) but does not formally recognize BP PGS as a risk-enhancing factor due to the nominal contribution of individual BP loci to measured BP [9]. Nevertheless, we demonstrate that higher BP PGS is associated with 8% longer duration spent above target BP and 2.4-fold higher risk of treatment-resistant hypertension across healthcare and general populations conferring wide gradient of cardiorenal disease risk and comorbidities burden. Furthermore, using novel multiancestral PGSs, we establish the independent associations and robust predictive utility of both SBP and DBP PGSs in treatment-resistant hypertension prediction (ΔC-index [95% CI], 0.04 [0.04–0.05]) otherwise unobserved in prior study [38]. Akin to our finding, a pharmacogenomic substudy of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial demonstrated that participants in the highest BP PGS quintile had 50% to 70% higher odds of having treatment-resistant hypertension compared to those at intermediate risk [39]. Such finding suggests that BP-regulating genes and pathobiology (i.e., vasodilation activation of the renin-angiotensin aldosterone system) may contribute to BP-lowering response [40] and implies prognostic utility of BP PGS for risk stratification beyond traditional risk factors.
Furthermore, BP PGSs significantly improved discrimination for stability of BP control and treatment-resistant hypertension, which may help to prioritize subgroups who are likely to derive greater benefit from a specific therapy when other phenotypic risk factors are equalized. Nevertheless, evidence of genetic interaction and heterogeneity of BP-lowering treatment effect across BP PGS remain scarce and inconsistent. Within limited scope, current pharmacogenomics focus on the identification of candidate genes responsible for inter-individual variability in antihypertensives response. Few modestly-powered epidemiological studies have demonstrated divergent findings on the effect of BP PGS on measured BP response to select antihypertensives classes. Whereas Black adults in the lowest SBP PGS quintile derived 3.5 mmHg-greater SBP reduction at 6 months from hydrochlorothiazide compared to the median PGS quintile [39], nonsignificant effect modification was observed in Finnish male population [41]. Furthermore, no associations were observed between the SBP PGS and BP response to lisinopril [39] or diuretics [41]. The lack of high-powered, methodologically optimized, and diverse trials examining the treatment effect between BP PGSs and class-specific antihypertensives underscores the need to refine strategies to improve hypertension management in the context of heightened genetic risk. A recent primary and secondary prevention study [42] demonstrated that atherogenicity of even moderately elevated low-density lipoprotein cholesterol is higher in individuals with a high PGSs. In this context, BP PGS may similarly identify an actionable genotype in whom targeted intensive BP lowering therapy could derive the greatest absolute CVD risk reduction. Nevertheless, whether new modalities emerging for treatment-resistant hypertension, such as zilbesiran [43] and renal denervation [44], will benefit those with high BP PGS to a greater extent need future investigations.
Based on multi-ancestry BP PGSs and rigorously characterized longitudinal BP control using international biobanks with linkage to electronic health records, we demonstrated the translational potential of BP genetic score in hypertension management. Nevertheless, our results have potential limitations. First, BPs were collected using varied devices and protocols under diverse contextual and physical environments. Therefore, direct comparison to studies using standardized BP measurements may be limited. Second, white-coat hypertension or masked hypertension may have led to misclassification of on-target BP achievement. Conversely, healthcare providers may have set higher target BPs to minimize adverse effects such as hypoperfusion, syncope, frailty, or electrolyte abnormalities; in such case, pseudo-resistance is possible. Nevertheless, we have rigorously excluded patients with advanced cardiorenal conditions and further validated in the U.K. guideline adhering to more liberal BP thresholds. Third, patients may have undergone BP measurement at out-of-network practices. Similarly, our results may have limited external generalizability across sociodemographically or clinically heterogeneous populations adhering to differing BP-lowering practice. Nevertheless, we expect missingness to be minimal as the Mass General Brigham is the largest multi-institutional health system in New England, and the National Health Service serves as the single-payer system for UK Biobank. Lastly, our results may have limited generalizability across external populations with different population structure, BP distribution, and hypertension management practices.
In summary, BP PGSs exhibited discernible associations and predictive capabilities for long-term BP control and treatment-resistant hypertension in adults with hypertension. Given ubiquity of suboptimal BP control, genetic assessment may augment opportunities for earlier treatment intensification for patients with high underlying risk, quantification of individualized treatment threshold for more effective and safe hypertension management, and identification of candidates meriting novel BP-lowering modalities. With increasing availability of large-scale multiancestral biobanks, the cost effectiveness of routine incorporation of BP genetic instruments into clinical roadmap toward personalized hypertension management warrants future exploration.
AP
Dr. Aniruddh Patel
Cardiology · General Non-Invasive CardiologyMassachusetts General Hospitalaniruddh
