Association between polygenic risk scores and cardiovascular events in prostate cancer patients receiving androgen deprivation therapy in Han Chinese

Abstract

Introduction

Cardiovascular disease (CVD) remains one of the leading non-cancer causes of mortality in prostate cancer patients, with previous studies indicating that androgen deprivation therapy (ADT) may increase cardiovascular risk. In this study, we aimed to predict cardiovascular events by incorporating polygenic risk scores (PRS).

Methods

Data were collected from 24,778 men, including 903 prostate cancer patients at Taichung Veterans General Hospital. Genotyping was performed using the Affymetrix Genome-Wide TWB 2.0 SNP Array. Four PRSs (PGS000337, PGS002262, PGS002725, and PGS003727) associated with coronary artery disease were utilized for analysis. Patients were stratified into quartiles based on their risk levels. Cox proportional hazards models were applied to assess the association between PRSs and the incidence of cardiovascular disease, adjusting for age, androgen deprivation therapy use, and comorbidities.

Results

PGS003727 demonstrated the highest predictive performance, with an area under curve (AUC) of 0.5550. Among ADT-treated patients, those in the highest PRS quartile (Q4) of PGS003727 had a significantly higher risk of CVD compared to those in the lowest quartile (Q1) (HR = 1.892, p = 0.0056). However, this Q4-to-Q1 risk difference was not significant in non-ADT patients (HR = 1.244, p = 0.3951). Kaplan-Meier analysis revealed significant risk variation across Q1 to Q4 in the ADT-treated group (p = 0.028) but not in the non-ADT group. Among Q4 prostate cancer patients, ADT treatment significantly increased CVD risk compared to non-ADT treatment (HR = 1.625, p = 0.0428). Kaplan-Meier analysis further confirmed that ADT treatment was associated with a higher cumulative incidence of CVD in Q4 prostate cancer patients (p = 0.033).

Conclusion

This hospital-based cohort study demonstrated that the PRS effectively predicted CVD risk in prostate cancer patients receiving ADT. Integrating this model into clinical practice could enable more precise cardiovascular risk assessment to aid treatment decisions and improving disease prevention strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40959-026-00445-7.

Introduction

Androgen deprivation therapy (ADT) is the cornerstone of treatment for advanced or high-risk prostate cancer, leveraging the dependence of prostate cancer cells on androgen signaling for growth and survival. Since the seminal work by Huggins and Hodges in the 1940s demonstrated that castration and estrogen administration could markedly reduce serum phosphatases in metastatic prostate cancer, ADT has been refined and incorporated into standard care for multiple disease stages [1]. Contemporary guidelines, including those from the National Comprehensive Cancer Network (NCCN), advocate ADT either alone or in combination with other modalities (e.g., radiation therapy, chemotherapy) for men with advanced, metastatic, or high-risk localized disease [2].

While prostate cancer–specific mortality remains a critical endpoint, cardiovascular disease (CVD) has emerged as a leading cause of non–tumor-related death in prostate cancer patients [3]. Although ADT is a mainstay of therapy, its metabolic and cardiovascular side effects remain an area of debate and ongoing research [4]. Multiple studies have been proposed to point out how ADT could adversely affect cardiovascular (CV) health, including unfavorable changes in body composition, lipid profiles, and insulin sensitivity [5]. Several large observational studies and meta-analyses suggest an association between ADT and elevated risk of CV events and all-cause mortality [6–8]. However, the magnitude of risk increased and specific contributing factors vary across studies, potentially due to differences in study design, patient populations, ADT types and duration, or baseline CV health status. This underscores the complexity of associating ADT use with CV outcomes and highlights the need for more refined stratification tools.

Polygenic risk scores (PRS) represent a novel approach that integrates multiple genetic variants to evaluate an individual’s genetic susceptibility to specific diseases or conditions. PRS has shown considerable utility in predicting various cardiovascular diseases (CVD), including atrial fibrillation [9, 10], heart failure [11], and coronary artery disease (CAD). Studies based on data from the UK Biobank have demonstrated the effectiveness of PRS in predicting CAD incidence and all-cause mortality, although these findings are primarily derived from populations of European ancestry [12, 13]. Furthermore, PRS has also been shown to effectively predict CAD risk in Chinese [14] and Japanese population [15]. Given the genetic heterogeneity among different populations, the impact of PRS on CVD risk may vary across ethnic groups.

Despite the growing body of research on PRS and its predictive value for CVD, no studies have specifically examined the association between PRS and CVD risk in prostate cancer patients undergoing ADT, a population known to be at high risk for CVD. Given the well-established link between ADT and increased CVD risk, this study aims to assess the predictive value of PRS for CVD risk in prostate cancer patients receiving ADT. This analysis will be conducted using a cohort from Taichung Veterans General Hospital (TCVGH) as part of the Taiwan Precision Medicine Initiative (TPMI) project.

Methods

Study population

This hospital-based retrospective cohort study included participants aged 20 years or older who were enrolled in the TPMI, a nationwide genetic research program led by Academia Sinica in Taiwan. The TPMI project collected medical records and blood samples from 57,257 participants, recruited from volunteers at TCVGH, a tertiary medical center, between July 2019 and November 2022.

The study population included 903 prostate cancer patients identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 185.x. Participants were categorized into an ADT group and a non-ADT group based on whether they received ADT. The ADT group consisted of patients who had undergone castration therapy, including orchiectomy or treatment with luteinizing hormone-releasing hormone (LH-RH) agonists or antagonists.

The comprehensive TPMI dataset included demographic information, medical procedures, diagnoses, surgeries, and medication prescriptions. Genotyping was performed for all participants using the Affymetrix Genome-Wide TWB 2.0 SNP Array.

Prior to blood sample collection, written informed consent for genetic analysis was obtained from all enrolled participants. All participants provided written informed consent in accordance with the Declaration of Helsinki. The ethical review was approved by the TCVGH Institutional Review Board (IRB no. CE23119A). All procedures were conducted in accordance with relevant guidelines and regulations. Clinical parameters were obtained through de-identified data from the TPMI dataset and electronic medical records at TCVGH.

Genotyping and quality control

DNA was extracted from participants’ blood samples and genotyped using the Axiom Genome-Wide TWB 2.0 Array Plate (Affymetrix, Santa Clara, CA, USA), which contains 714,431 SNPs specifically designed for the Taiwanese Han Chinese population [16].

Genotyping analysis and quality control were performed using Affymetrix Power Tools software. SNPs were excluded if they deviated from Hardy-Weinberg equilibrium (P < 1.0 × 10⁻⁵), had a minor allele frequency (MAF) < 0.05, or a genotype missing rate > 5%. After quality control, 591,048 SNPs were retained for analysis. The feasibility of obtaining GWAS SNP data from Taiwanese Han Chinese populations using customized arrays has been previously validated [17]. Samples with a missing rate > 0.02, an inbreeding coefficient > 0.15, or sex mismatches were excluded. Genotype imputation was conducted for autosomal chromosomes using the Michigan Imputation Server and the ‘minimac4’ algorithm [18], with the 1000 Genomes Phase 3 (Version 5) reference panel [19]. Only biallelic variants meeting an INFO score ≥ 0.3 were included. Genotype data for each participant were integrated with selected SNPs using imputed data as a reference.

Polygenic risk score analysis

To calculate polygenic risk scores (PRS), the scoring function in PLINK version 2.0 was used to aggregate the effects of multiple genetic variants, weighted by their effect sizes from GWAS [20]. PLINK 2.0 automatically resolves issues related to inverted effect alleles or non-effect alleles in datasets. To prevent multicollinearity in PRS development, linkage disequilibrium (LD) pruning and clumping were performed using PLINK 2.0 to select independent and informative SNPs [21, 22]. LD refers to the non-random association of alleles at different loci within a population, often due to physical proximity on a chromosome or other evolutionary factors. For example, when two genetic variants (e.g., SNPs) exhibit strong LD, they are co-inherited more frequently than expected by chance. The PRSs used in this study included PGS000337, PGS002262, PGS002725, and PGS003727, all of which are associated with CVD. PGS002262 was developed based on individuals of East Asian ancestry [14], PGS003727 was based on individuals of European ancestry [13], while PGS000337 and PGS002725 were derived from multi-ancestry populations [15, 23], validated through trans-ancestry GWAS meta-analyses. The standard formula for calculating weighted PRSs is as follows:

Equation (1) calculates the weighted PRS for individual j, where N represents the total number of SNPs included in the PRS calculation from GWAS, denotes the effect size of the effect allele of SNP_i, and dosage_ij refers to the number of effect allele copies in the genotype of individual j.

Assessment of clinical parameters and outcomes

Prostate cancer patients were identified using ICD-9-CM code 185, confirmed by pathological reports. The index date was defined as the date of prostate cancer diagnosis, determined by ICD-9-CM code 185 recorded at least twice in outpatient visits or once during hospitalization between January 2009 and January 2022. Relevant biochemical data were extracted from the TCVGH database, and covariates such as age, body mass index (BMI), and comorbidities were assessed.

Comorbidity information was retrieved from TCVGH electronic health records using ICD-9 diagnostic codes, including neurologic diseases (stroke, Parkinson’s disease, paraplegia, dementia; ICD-9-CM codes 430–438, 332.0, 332.1, 342, 3441, 290, 331.0, 331.2), pulmonary diseases (490–496, 500–505), connective tissue diseases (7100, 7101, 7104, 7140–7142, 71481, 5171, 725), chronic kidney diseases (403, 404, 582, 583, 585, 586, 588, V42.0, V45.1, V56), and CVD (excluding stroke; ICD-9-CM codes 427.31, 428, 433.10, 433.11, 441, 4439, 7854, V434, 413.9, 410, 412, 414.00-414.05, 414.8, 401.0, 401.2, 401.9, 402.00, 402.01, 402.10, 402.11, 402.90, 402.91, 250, 268.9, 272.0, 272.1, 272.4, 277.7, 790.21, 790.29). Comorbidities were confirmed if the diagnostic code appeared once during hospitalization or at least twice in outpatient records.

Outcome evaluations included new CV events or high-invasive medical interventions following prostate cancer diagnosis. CVD were identified using ICD-9-CM codes (401-429.9, 785.51), while interventions were determined by hospital billing codes, including neurovascular thrombectomy, coronary artery bypass grafting (CABG), percutaneous transluminal coronary angioplasty (PTCA), coronary artery rotational atherectomy, pacemaker/implantable cardioverter defibrillator (ICD) implantation, electrophysiological study (EPS), transcatheter electrical ablation, and cardioversion. The follow-up period extended from January 2009 to January 2022.

Statistical analysis

Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated using Cox proportional hazards regression models, with time since study entry as the timescale. Outcomes were censored at loss to follow-up, death, or study end (November 2022). Continuous variables were presented as mean ± standard deviation (SD) and analyzed using ANOVA, while categorical variables were presented as numbers (percentages) and analyzed using the Chi-square test. The median follow-up time from prostate cancer diagnosis was reported with interquartile range (IQR) and analyzed using the Kruskal-Wallis test.

Receiver operating characteristic (ROC) curve analysis was used to evaluate the predictive performance of various PRSs for CV events. Participants were stratified into quartiles based on PRS values: Q1 (0–25%), Q2 (26–50%), Q3 (51–75%), and Q4 (76–100%).

The association between PRS quartiles and CV events was assessed using Cox proportional hazards regression, adjusting for potential confounders. Analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, NC) and SPSS version 22.0 (IBM Corp., Armonk, NY, USA).

Results

A total of 57,527 male participants were included in the analysis, among whom 903 were diagnosed with prostate cancer. The baseline characteristics of these patients are summarized in Table 1. Among the 903 prostate cancer patients, 446 received ADT, while 457 did not. During a median 5.01-year follow-up period, 284 CV events were recorded, with 152 occurring in the ADT group and 132 in the non-ADT group. Among patients receiving ADT, CV events were observed across all clinical stages, with 8, 61, 29, and 52 events occurring in stages I, II, III, and IV, respectively. Corresponding event numbers in the non-ADT group were 36, 79, 10, and 6 (Supplementary Table 1).

Table 1.

Characteristics of patient diagnosed with prostate cancer.

Variables		Total (n = 903)		Without ADT (n = 446)		With ADT (n = 457)		P-value
		n	(%)	n	(%)	n	(%)
Age, year	< 65	81	8.97	53	11.88	28	6.13	< 0.0001
	65–75	384	42.52	212	47.53	172	37.64
	> 75	438	48.5	181	40.58	257	56.24
Onset age		68.41	8.34	66.81	8.03	69.96	8.34	< 0.0001
Smoking status	Non-smoker	453	50.17	236	52.91	217	47.48	0.1027
	Ever/current smoker	450	49.83	210	47.09	240	52.52
BMI	< 18.5	1	0.31	1	0.56	0	0	d0.6693
	18.5–24	118	36.76	66	36.87	52	36.62
	≥ 24	202	62.93	112	62.57	90	63.38
Blood pressure	SBP	134.78	18.12	134.47	17.45	135.09	18.76	0.6187
	DBP	77.06	12.37	78.11	12.04	76.03	12.62	0.0141
Lab data	TG	134.67	128.19	127.61	119.81	140.99	135.12	0.1947
	Fasting glucose	126.54	45.49	126.84	41.45	126.24	49.3	0.8705
	eGFR	77.89	21.71	76.62	20.14	79.11	23.08	0.0868
Clinical stage	Stage I	151	16.85	131	29.57	20	4.42	< 0.0001
	Stage II	438	48.88	269	60.72	169	37.31
	Stage III	116	12.95	30	6.77	86	18.98
	Stage IV	191	21.32	13	2.93	178	39.29
PGS000337 (mean/ SD)a		14.18	6.89	14.23	6.91	14.12	6.89	0.8263
PGS002262 (mean/ SD)a		0.17	0.46	0.17	0.46	0.17	0.46	0.8889
PGS002725 (mean/ SD)a		-1.54	0.22	-1.54	0.22	-1.54	0.22	0.6228
PGS003727 (mean/SD)a		1.5	1.12	1.48	1.12	1.52	1.12	0.5834
Comor-bidity (n/%)b	Neurologic disease	66	7.31	31	6.95	35	7.66	0.6828
	Pulmonary disease	61	6.76	28	6.28	33	7.22	0.5724
	Connective tissue disease	5	0.55	4	0.9	1	0.22	d0.2121
	Chronic kidney disease	64	7.09	31	6.95	33	7.22	0.8742
	CVD associated disease	10	1.11	4	0.9	6	1.31	d0.7527
	Hypertension	173	19.16	91	20.4	82	17.94	0.3476
	Hyperlipidemia	96	10.63	55	12.33	41	8.97	0.1015
	DM	91	10.08	48	10.76	43	9.41	0.4995
	Obesity	0	0	0	0	0	0	-
Outcome (n/%)b	CV events	284	31.45	132	29.6	152	33.26	0.2358
Median follow up time (from prostate cancer diagnosis to latest visit) (Median/IQR)c(years)		5.05	3.09–7.74	5.07	3.21–7.79	5.01	2.99–7.67	0.6499

^dUsing Fisher’s exact test.

^aContinuous variables were expressed as mean ± standard deviation (SD) and were analyzed using ANOVA follow a normal data distribution.

^bCategorical variables were expressed as numbers (percent) and were analyzed using the Chi-square test.

^cUsing Kruskal-Wallis test.

Patients under 65 years of age were more prevalent in the non-ADT group compared to the ADT group (11.88% vs. 6.13%; p < 0.0001). Similarly, the mean age at diagnosis was significantly lower in the non-ADT group compared to the ADT group (66.81 vs. 69.96 years; p < 0.0001). The prevalence of stage III and IV disease was significantly higher in the ADT group than in the non-ADT group (29.57% vs. 4.42%; p < 0.0001). At baseline, systolic blood pressure, triglyceride levels, fasting glucose, and renal function were comparable between patients receiving ADT and those not receiving ADT. Diastolic blood pressure was slightly lower in the ADT group (p = 0.0141). No significant differences were observed between the two groups with respect to smoking status, body mass index, PRS, comorbidities, or median follow-up duration.

We conducted ROC curve analysis to evaluate the predictive ability of four PRS scores (PGS000337, PGS002262, PGS002725, and PGS003727) for CV events, as illustrated in Fig. 1. The areas under the curves (AUCs) for these scores were 0.5248 (Fig. 1a), 0.5331 (Fig. 1b), 0.5238 (Fig. 1c), and 0.5550 (Fig. 1d), respectively. Based on its superior predictive performance, we selected PGS003727 and divided participants into quartiles. As shown in Supplementary Table 2, Patient numbers were evenly distributed across PRS quartiles in both groups. We then analyzed CVD risk by comparing Q2, Q3, and Q4 against Q1 as the reference group (Table 2). In Q2 and Q3 group, compared to Q1 group, there were no significant differences in CV event risk regardless of ADT status. In the Q4 vs. Q1 comparison, patients who received ADT had a significantly increased risk of CVD (HR = 1.892; p = 0.0056), while no significant difference was observed among those who did not receive ADT (HR = 1.244; p = 0.3951).

Fig. 1.

Receiver Operating Characteristic (ROC) Curves for Predicting ROC curve analysis illustrating the predictive performance of four PRSs [PGS000337(a), PGS002262 (b), PGS002725 (c), and PGS003727 (d)] for cardiovascular (CV) events. The area under the curve (AUC) values indicates the discriminative ability of each PRS in identifying individuals at risk of CV events. Higher AUC values represent better predictive performance

Table 2.

Risk of CV events in ADT group and without ADT group.

Variables	With ADT (n = 457)				Without ADT (n = 446)
	HR	95% CI		P-value	HR	95% CI		P-valuea
PGS003727
Q2|Q1	1.28	0.791	2.072	0.315	1.022	0.615	1.699	0.9325
Q3|Q1	0.985	0.595	1.63	0.952	1.357	0.825	2.233	0.2295
Q4|Q1	1.892	1.204	2.972	0.0056	1.244	0.752	2.06	0.3951

^aHRs were estimated using univariable and multivariable Cox proportional hazards model, and adjusted for age, neurologic disease, pulmonary disease, chronic kidney disease.

Kaplan-Meier curves were generated to compare the cumulative incidence of CVD events across PGS003727 quartiles (Q1–Q4) within the ADT and non-ADT groups. A significant difference in CV event risk was observed in the ADT group (p = 0.035; Fig. 2a), while no significant difference was found in the non-ADT group (p = 0.75; Fig. 2b). Additionally, Kaplan-Meier analysis comparing Q1 and Q4 within the ADT cohort (Fig. 2c) demonstrated a significantly higher risk of CV events in Q4 patients compared to Q1 patients (p = 0.028).

Fig. 2.

Kaplan-Meier Survival Curves for Cardiovascular Event Risk Across PGS003727 Quartiles. Kaplan-Meier curves illustrating the cumulative incidence of cardiovascular disease (CVD) events across quartiles (Q1 to Q4) of the PRS PGS003727 within androgen deprivation therapy (ADT) and non-ADT groups. a A significant difference in CVD event risk among quartiles was observed in the ADT group (p = 0.035) (b) No significant difference among quartiles was detected in the non-ADT group (p = 0.75). c Kaplan-Meier survival analysis comparing CV event incidence between Q1 and Q4 within the ADT cohort, demonstrating a significant difference (p = 0.028)

To further address potential confounding by prostate cancer clinical stage, we performed stage-stratified Kaplan–Meier analyses by grouping patients into early-stage (Stages I–II) and advanced-stage (Stages III–IV) disease. As shown in Supplementary Fig. 1, the cumulative incidence of cardiovascular events across PRS quartiles demonstrated similar patterns within each stage stratum, with no significant differences in the divergence across PRS quartiles observed in either the ADT or non-ADT groups. However, in stage-stratified analyses focused on patients receiving ADT, a significantly higher cumulative incidence of cardiovascular events was observed in the highest PRS quartile (Q4) compared with the lowest quartile (Q1) among patients with early-stage disease (Stages I–II; p = 0.029) (Supplementary Fig. 2a). In contrast, no significant difference between Q4 and Q1 was observed among patients with advanced-stage disease (Stages III–IV; p = 0.46) (Supplementary Fig. 2b). These findings suggest that the association between PRS and cardiovascular risk among ADT-treated patients may be more pronounced in early-stage disease.

In Table 3, prostate cancer patients were stratified into quartiles (Q1 to Q4) based on their PRS, and hazard ratios for CVD were compared between those who received ADT and those who did not within each quartile. The analysis was adjusted for age, neurologic disease, pulmonary disease, and chronic kidney disease. No significant differences in CV event risk were observed between the ADT and non-ADT groups in Q1 through Q3. In Q4, patients receiving ADT had a significantly higher risk of CVD compared to those in the non-ADT group (HR = 1.625, 95% CI = 1.016–2.600, p = 0.0428).

Table 3.

Risk of CV events in each quartile of PGS003727.

Variables	HR	95% CI		P-valuea
PGS003727
Q1
Without ADT	-	-	-	-
ADT	1.069	0.644	1.774	0.7952
Q2
Without ADT	-	-	-	-
ADT	1.406	0.856	2.311	0.1787
Q3
Without ADT	-	-	-	-
ADT	0.786	0.47	1.314	0.3578
Q4
Without ADT	-	-	-	-
ADT	1.625	1.016	2.6	0.0428

^aHRs were estimated using univariable and multivariable Cox proportional hazards model, and adjusted for age, neurologic disease, pulmonary disease, chronic kidney disease.

To determine whether ADT is an independent risk factor for cardiovascular disease, we performed univariable and multivariable Cox proportional hazards analyses. As shown in Table 4, in univariable analysis, none of the variables including ADT (HR 1.184, 95% CI 0.938–1.495, p = 0.1555) showed significant associations with cardiovascular disease risk, with the exception of comorbidities such as neurologic disease, pulmonary disease, and chronic kidney disease. In multivariable analysis adjusting for age, smoking status, clinical stage, PRS quartiles, and comorbidities, ADT was not independently associated with cardiovascular disease risk. Notably, the highest PRS quartile (Q4) showed a significant association with increased cardiovascular risk compared to Q1 (HR 1.514, 95% CI 1.084–2.116, p = 0.015).

Table 4.

Risk of CV events associated with PGS003727 quartiles and androgen deprivation therapy.

Variables	Univariable				Multivariable
	HR	95% CI		P-valuea	HR	95% CI		P-valuea
Age	1.004	0.991	1.017	0.5791	1.006	0.991	1.02	0.4328
Smoking	1	0.792	1.262	1	1.087	0.854	1.383	0.4973
Clinical stage
Stage I	-	-	-	-	-	-	-	-
Stage II	1.168	0.832	1.639	0.3703	1.034	0.724	1.478	0.8532
Stage III	1.522	0.988	2.344	0.0567	1.382	0.858	2.226	0.1833
Stage IV	1.212	0.818	1.794	0.3376	0.951	0.599	1.509	0.8308
ADT	1.184	0.938	1.495	0.1555	1.174	0.878	1.57	0.2791
PGS003727
Q1	-	-	-	-	-	-	-	-
Q2	1.098	0.775	1.555	0.5995	1.163	0.82	1.651	0.3965
Q3	1.068	0.752	1.517	0.7113	1.085	0.761	1.548	0.6521
Q4	1.383	0.992	1.927	0.0555	1.514	1.084	2.116	0.015
Neurologic disease	0.359	0.196	0.657	0.0009	0.378	0.204	0.699	0.0019
Pulmonary disease	0.557	0.331	0.94	0.0282	0.639	0.368	1.109	0.111
Chronic kidney disease	0.49	0.286	0.84	0.0094	0.483	0.275	0.848	0.0113

^aHRs were estimated using univariable and multivariable Cox proportional hazards model, and adjusted for age, smoking, neurologic disease, pulmonary disease, chronic kidney disease.

Kaplan-Meier survival analysis, adjusted for confounding variables, further illustrated these findings (Fig. 3). No significant differences were detected between the ADT and non-ADT groups in Q1 through Q3 (p = 0.74, 0.25, and 0.35, respectively; Fig. 3a, b and c). In Q4, a significant increase in cumulative CVD incidence was observed in the ADT group compared to the non-ADT group (p = 0.033; Fig. 3d).

Fig. 3.

Kaplan-Meier Survival Curves Comparing ADT and Non-ADT Groups Across PGS003727 Quartiles.Kaplan-Meier curves displaying the cumulative incidence of cardiovascular disease (CVD) events among prostate cancer patients stratified into quartiles (Q1 to Q4) based on the PRS (PGS003727). Comparisons were made between ADT-treated and non-ADT patients within each quartile. No significant differences were observed in Q1, Q2, and Q3 (p = 0.74, 0.25, and 0.35, respectively) (a to c). In Q4, ADT-treated patients exhibited a significantly increased incidence of CV events compared to non-ADT patients (p = 0.033) (d). These findings were further adjusted for confounding variables to ensure robustness of the results

Discussion

Our study is the first to demonstrate a significant association between PRS and the incidence of CVD in prostate cancer patients receiving ADT within the Han Chinese population. Through an analysis of data from the TCVGH-TPMI cohort, we found that a PRS predicting CVD could also serve as a predictor of CVD risk in prostate cancer patients undergoing ADT, a population already known to have an elevated risk of CVD. Notably, prostate cancer patients in the highest PRS quartile who received ADT were at an increased risk of developing CVD.

The mechanisms linking ADT to increased CVD risk are unclear, but ADT may promote metabolic syndrome, elevating risks of CVD and type II diabetes mellitus [24]. Potential mechanisms include insulin resistance, glucose intolerance, hyperinsulinemia, hypertriglyceridemia, elevated low-density lipoprotein (LDL) levels, reduced high-density lipoprotein (HDL) levels, and hypertension [25]. Study showed that ADT has been linked to unfavorable metabolic changes, including alterations in glycemic control, lipid profiles, and body weight in the Asian population [26]. These metabolic changes could potentially lead to increased risk of cardiovascular disease [27, 28]. Fluctuations in hormone levels affect endothelial cell stability and activation, creating conditions that promote the formation and destabilization of atherosclerotic plaques [29, 30]. ADT has been shown to increase the accumulation of both visceral and subcutaneous abdominal fat, while concurrently leading to a decrease in lean body mass [31, 32]. ADT has been shown to contribute to insulin resistance. In men without diabetes mellitus, ADT is also associated with worsening of fasting insulin, fasting glucose, leptin, and the homeostasis model assessment of insulin resistance [33]. Over the past decade, studies have raised concerns regarding the potential association between ADT and an increased risk of adverse cardiovascular events, including myocardial infarction, stroke, and cardiovascular mortality. This heightened risk may be attributed to factors such as plaque destabilization, visceral adiposity, insulin resistance, and endothelial dysfunction [34].

In our multivariable Cox regression analysis, ADT was not identified as an independent risk factor for cardiovascular disease after adjustment for age, smoking status, clinical stage, PRS, and comorbidities. While some prior studies have reported an association between ADT and increased cardiovascular risk [34, 35], these findings have been inconsistent across different populations and study designs [36, 37]. The lack of independent association in our cohort may be attributed to several factors, including the heterogeneity of disease stages-particularly the inclusion of stage IV patients with shorter life expectancies, which may have attenuated the long-term cardiovascular effects of ADT.

Several murine models have demonstrated that ADT exacerbates atherosclerotic lesion formation. In studies involving castrated male mice on a high-fat diet, these mice developed larger atherosclerotic lesions at the aortic origin compared to their intact counterparts. Furthermore, testosterone supplementation in orchiectomized mice significantly reduced the size of atherosclerotic lesions [38]. Similarly, androgen receptor (AR) knockout mice on an apolipoprotein E-deficient background exhibited larger atherosclerotic lesions in the aortic root than mice with an intact AR, suggesting that the disruption of testosterone signaling may promote atherogenesis [39]. Additionally, testosterone was found to augment cholesterol efflux from human monocyte-derived macrophages in a dose-dependent manner by upregulating scavenger receptor B1, providing a potential mechanism for how testosterone can decrease the cholesterol content of atherosclerotic lesions [40]. Collectively, these preclinical findings support the hypothesis that ADT drives the progression of atherosclerosis.

Epidemiological studies have demonstrated a significant association between family history and CVD, suggesting a potential hereditary component in CVD risk [41]. Genetic variations associated with lipid disorders play a crucial role in CVD risk. For instance, mutations in the low-density lipoprotein (LDL) receptor or its associated genes can lead to familial hypercholesterolemia, a condition characterized by markedly elevated LDL cholesterol levels, xanthomas, and a predisposition to CVD [42–44]. If left untreated, individuals with familial hypercholesterolemia face an exceptionally high risk of CVD; however, early intervention with statin therapy can significantly mitigate this risk [45]. Additionally, triglyceride-rich lipoproteins have been identified as a causal factor for CVD [46]. Elevated triglyceride levels, driven by deleterious mutations in lipoprotein lipase or apolipoprotein A5, are associated with an increased risk of CVD. These mutations impair triglyceride metabolism, leading to prolonged circulation of triglyceride-rich lipoproteins, endothelial dysfunction, and heightened atherogenic potential. Consequently, individuals carrying these genetic variants may face a significantly elevated risk of atherosclerosis and subsequent cardiovascular event [47, 48]. A single nucleotide polymorphism (SNP) in the lipoprotein(a) (LPA) locus on chromosome 6q26-27 has been associated with elevated serum Lp(a) levels and a significantly increased risk of coronary artery disease [49].

The PRS, derived from multiple genetic markers associated with CVD, offers a comprehensive risk assessment by incorporating an individual’s genetic predisposition [50]. The first study to utilize a PRS for CVD was based on 13 SNPs associated with coronary heart disease. By incorporating weighted genetic variants, this PRS was able to identify the top 20% of individuals of European ancestry who have an approximately 70% increased risk of experiencing a first coronary heart disease event [51]. On the other hand, PRS incorporating genetic variants related to cellular adhesion, leukocyte migration, atherosclerosis, insulin resistance, and transendothelial migration has been found to be associated with an increased risk of CVD [52–54]. Additionally, a PRS comprising 12 genetic variants associated with atrial fibrillation has been significantly linked to an increased risk of ischemic stroke [55]. Furthermore, a PRS derived from six million SNPs in the UK Biobank has been associated with an increased risk of coronary artery disease. Notably, the prevalence of individuals at high genetic risk is approximately 20 times higher than the carrier frequency of rare monogenic mutations that confer a comparable level of risk [56]. Similar studies have also been reported in Asian populations. In the Japanese population, Hachiya et al. identified a genome-wide PRS associated with an increased risk of ischemic stroke events [57]. Furthermore, Koyama et al. utilized data from Biobank Japan and the UK Biobank to develop a PRS for predicting CAD in the Japanese population, known as PGS000337 [15]. However, this PRS was not predictive in our study population. In our study, PGS003727 had the best predictive performance (AUC = 0.5550), though this reflects only a modest improvement over random classification. This highlights the complexity of genetic contributions to disease risk, driven by multiple genetic and environmental factors. Though PGS003727 shows potential as a predictive tool, its limited performance advises against standalone clinical use. It could complement established risk factors (clinical, biochemical, lifestyle), with further validation and multi-omics integration possibly improving PRS utility in personalized medicine.

For patients identified as being at high genetic risk based on PRS, additional interventions may help reduce the risk of CVD. Analyses from two randomized controlled primary prevention trials, ASCOT and JUPITER, demonstrated that individuals at high genetic risk exhibit a greater burden of subclinical atherosclerosis and derive both greater relative and absolute benefits from statin therapy in preventing a first coronary heart disease event [58]. Furthermore, among patients with a high PRS, adopting a favorable lifestyle was associated with nearly a 50% lower relative risk of coronary artery disease compared to those with an unfavorable lifestyle [59]. Thus, for prostate cancer patients with a high PRS for cardiovascular disease CVD, early intervention, such as statin therapy, may theoretically provide benefits in preventing further cardiovascular complications.

Currently, no study has specifically investigated the association between PRS and the risk of CVD in prostate cancer patients. Therefore, we selected four PRS models specific to coronary heart disease for analysis, including those derived from Japanese and Chinese populations, as well as a European-based mode [13–15, 23]. Among these, PGS003727 from the UK Biobank demonstrated the highest predictive value for CVD in prostate cancer patients undergoing hormone therapy in our population. PGS003727 has been strongly associated with both prevalent and incident coronary artery disease events in UK Biobank participants. Additionally, it has shown excellent performance in predicting prevalent CAD across diverse ancestry groups within the UK Biobank, including individuals of African, East Asian, and South Asian ancestry [13]. In our study, we also confirmed its predictive power for CVD. Although the AUC was not high, the score still demonstrated significant predictive value. This finding underscores the potential of PRS as a complementary tool in assessing risk for multifactorial diseases such as cardiovascular events.

From a clinical perspective, our results advocate for a more personalized approach to treatment selection and monitoring in prostate cancer. While traditional risk stratification often emphasizes clinical and demographic factors [60, 61], the integration of genetic data could identify a subgroup of patients requiring enhanced cardioprotective strategies. For instance, patients identified as genetically high-risk could undergo more frequent CV assessments, lifestyle interventions, or even preemptive initiation of cardioprotective therapies during ADT [35, 62]. Such measures align with the broader thrust of precision medicine, which encourages tailoring interventions to an individual’s unique genetic, environmental, and lifestyle background. Further, our work underscores the importance of extending PRS analyses to diverse populations and validating these findings in larger, multi-ethnic cohorts, ensuring that genetic risk models are both equitable and generalizable. Developing ancestry-specific PRS or incorporating genomic data from understudied populations will be critical in providing accurate, actionable risk information to all patients.

There are several limitations to this study that should be acknowledged. First, while we leveraged a large cohort from the TPMI program at TCVGH, the sample size of patients receiving ADT who also experienced CV events remained modest, potentially affecting the precision of our estimates. Second, the follow-up period of approximately five years may not capture long-term CV outcomes, and extending longitudinal analyses would provide a more comprehensive risk profile. Third, our PRS and findings primarily reflect data derived from populations of European ancestry; validation in more diverse populations is imperative to enhance generalizability. Finally, confounding factors, including lifestyle variables, medication use and subtle comorbidities, may influence the observed associations and require more detailed adjustment in future analyses.

Conclusions

This hospital-based cohort study demonstrated that the PRS effectively predicted CVD risk in prostate cancer patients receiving ADT. Integrating this model into clinical practice could enable more precise cardiovascular risk assessment, thereby aiding treatment decisions and improving disease prevention strategies. A large-scale prospective study with a longer follow-up period is needed to further validate our findings.

Authors’ contributions

Q-YN, L-WC, H-WY, and J-RL contributed to data collection, analysis, and interpretation. S-CH and I-CC conceived and designed the study, supervised the research process, and critically reviewed the manuscript. S-CH and I-CC also contributed to manuscript drafting and revision. All authors read and approved the final version of the manuscript.

Funding

This study was funded by Taichung Veterans General Hospital, Taiwan (grant numbers TCVGH-1137302B, TCVGH-1135003B and TCVGH-1143903B] and National Science and Technology Council, Taiwan (NSTC 113-2314-B-075 A-005).

Data availability

The data used in this article will be shared on reasonable request to the corresponding author. All data used in this study are available in this article. However, the individual-level PRS and prostate cancer information data are not currently available within the paper.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Institutional Review Board of the Taichung Veterans General Hospital (IRB no. CE23119A). All participants completed a written inform consent form according to the Helsinki declaration.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.