This was a cross-sectional study of individuals aged ≥20 years living in the Miyagi Prefecture, northeastern Japan, included in the TMMCommCohort study. The details have been previously described [24, 25]. Briefly, the TMMCommCohort study was started in May 2013; by March 2016, more than 50,000 participants were recruited through the following two approaches: the type 1 survey (n = 41,097 participants), which was performed at specific municipal health check-up sites, and the type 2 survey (n = 13,855), which was conducted at assessment centers. Participants provided information on lifestyle and other potentially health-related aspects through blood and urine samples and a mailed self-reported questionnaire. All participants (n = 54,952) provided written informed consent for this study. The Institutional Review Board of the Tohoku Medical Megabank Organization approved this study (approval number: 2022-4-047; approval date: June 30, 2022).

As several physiological measurements, including home BP, were conducted only in the type 2 survey, we selected participants from this survey group (n = 13,855). We excluded participants: (1) who withdrew from the study before December 11, 2023; (2) who failed to return a self-reported questionnaire (n = 212); (3) without genetic information genotyped on an Affymetrix Axiom Japonica Array (v2; Affymetrix, Santa Clara, CA, USA) (n = 3788); 4) with missing data related to various factors, including BP, home BP measurements for a minimum of 3 days in the morning (n = 2724), height, weight, urinary creatinine, estimated urinary 24-h sodium excretion, estimated urinary 24-h potassium excretion, alcohol consumption status, or physical activity (n = 91); and 5) with a standard deviation (SD) ≥ 6 for each genetic principal component (n = 13). Consequently, 7027 participants fulfilled all inclusion criteria, and their data were analyzed in this study. The data of these participants were randomly categorized into target (n = 1405; 20%) and test (n = 5622; 80%) data. The target data were used to determine the P-thresholds of the best-fit PRS for each trait. The test data were used to examine the associations of the combined PRS and lifestyle score with hypertension or home hypertension (Fig. 1).

Flowchart of study participants

A healthy lifestyle score was constructed based on the following four well-established hypertension risk factors: alcohol consumption status, body mass index (BMI), physical activity, and sodium-to-potassium (Na/K) ratio [4,5,6,7]. The Methods, Supplemental Digital Content 1, includes the definitions of alcohol consumption, BMI, physical activity, and Na/K ratio. Subsequently, the overall lifestyle was categorized into ideal (having at least three ideal lifestyle factors), poor (having at least three poor lifestyle factors), or intermediate (having two ideal lifestyle factors).

A trained nurse measured the BP twice in the upper right arm using a digital automatic BP monitor (HEM-9000AI; Omron Healthcare Co., Ltd., Kyoto, Japan) at the community support center after the participants rested for at least 2 min in a sitting position. The mean values of the two recorded measurements were used for analysis. Home BP was measured using a cuff-oscillometric device (HEM-7080IC; Omron Healthcare Co., Ltd.). Participants measured their home BP in a sitting position after resting for at least 5 min in the morning within 1 h of waking; maintaining the arm at heart level during resting; and if applicable, before taking medications for hypertension, eating breakfast, and after urination. The average home BP in the morning for ≥3 days was used for all analyses. Hypertension was defined as systolic BP (SBP)/diastolic BP (DBP) of 140/90 mmHg or higher measured at a community-support center and/or self-reported hypertension treatment. Home hypertension was defined as home SBP/DBP of 135/85 mmHg or higher, or self-reported hypertension treatment [4].

Detailed information about genotyping and quality control in this study is described in the Methods, Supplemental Digital Content 1. We calculated the PRS based on the summary statistics of a previous GWAS for SBP in the BBJ, which is publicly available at the National Bioscience Database Center [26]. Participants included in our study were distinct from those in the BBJ. All SNPs on the X and Y chromosomes were removed from the data to eliminate the possibility of non-autosomal sex effects. PLINK 1.9 (COVID-19 Genomics UK, Cambridge, UK) was used to calculate the PRS using the clumping and thresholding method. Based on a previous study [19], we performed clumping to capture the right level of the causal signal using the following options: --clump-p1 1 --clump-r2 0.1 –clump-kb 250.

After clumping, we calculated the PRS for each individual in the target dataset (n = 1405) using various variant sets according to different P-value thresholds. The PRS was calculated using the default formula for PRS calculation in PLINK (COVID-19 Genomics UK) (https://choishingwan.github.io/PRS-Tutorial/plink/). As a default setting, we calculated the PRS using the following nine different P-value thresholds: 5.0 × 10−8, 0.001, 0.01, 0,05, 0.1, 0.2, 0.3, 0.4, and 0.5. Among various PRSs with different numbers of SNPs, we chose the list of variants that showed the best fit (determined using a variance explained with the PRS). The settings of the best-fit PRS for both SBP and home SBP in the target data were a PRS with P < 0.001 (Table, Supplemental Digital Content 2). Therefore, we used a PRS constructed using P < 0.001 in the test data (n = 5622) for the analysis. An overview of the PRS calculations and association study is illustrated in Figure, Supplemental Digital Content 3.

Data were presented as means (SD) or median (interquartile range) for continuous variables and number (percentage) for categorical variables. Participants were classified based on their PRS tertile to analyze the potential association between the PRS and hypertension. Multivariate logistic regression analyses were performed to examine the association of the PRS and lifestyle with the prevalence of hypertension and home hypertension, respectively. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) were estimated.

Participants were classified into nine groups based on their PRS and lifestyle to examine the combined effect of genetic and lifestyle risks. For hypertension, logistic regression analyses were adjusted for age at inclusion, sex, and the first six principal components (to adjust for population structure). For home hypertension, logistic regression analyses were adjusted for the above model, as well as added measurement seasons of home BP, as home BP was affected by seasonal temperature changes [27, 28].

To examine the influence of genetic risk on the classification performance, we calculated the area under the receiver operating characteristic curve (AUROC) and 95% CI before and after including the PRS in the statistical model, including healthy lifestyle score, using logistic regression analysis. The AUROCs were compared using the DeLong test.

Additionally, to rule out the influence of hypertension treatment, we excluded participants who were receiving treatment for hypertension (n = 1144) and examined the associations of combined genetic and lifestyle risks with hypertension and home hypertension. Furthermore, to confirm the robustness of our results, we conducted an analysis of covariance and estimated the adjusted least-square means of SBP for 9 categories by genetic and lifestyle risk among participants without treatment for hypertension.

A two-sided P < 0.05 was considered statistically significant. Statistical analysis was performed using R version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria).