Validation of a multi-ancestry polygenic risk score and age-specific risks of prostate cancer: A meta-analysis within diverse populations.

Background: We recently developed a multi-ancestry polygenic risk score (PRS) that effectively stratifies prostate cancer risk across populations. In this study, we validated the performance of the PRS in the multi-ancestry Million Veteran Program and additional independent studies.
Methods: Within each ancestry population, the association of PRS with prostate cancer risk was evaluated separately in each case-control study and then combined in a fixed-effects inverse-variance-weighted meta-analysis. We further assessed the effect modification by age and estimated the age-specific absolute risk of prostate cancer for each ancestry population.
Results: The PRS was evaluated in 31,925 cases and 490,507 controls, including men from European (22,049 cases, 414,249 controls), African (8794 cases, 55,657 controls), and Hispanic (1082 cases, 20,601 controls) populations. Comparing men in the top decile (90-100% of the PRS) to the average 40-60% PRS category, the prostate cancer odds ratio (OR) was 3.8-fold in European ancestry men (95% CI = 3.62-3.96), 2.8-fold in African ancestry men (95% CI = 2.59-3.03), and 3.2-fold in Hispanic men (95% CI = 2.64-3.92). The PRS did not discriminate risk of aggressive versus nonaggressive prostate cancer. However, the OR diminished with advancing age (European ancestry men in the top decile: ≤55 years, OR = 7.11; 55-60 years, OR = 4.26; >70 years, OR = 2.79). Men in the top PRS decile reached 5% absolute prostate cancer risk ~10 years younger than men in the 40-60% PRS category. Conclusions: Our findings validate the multi-ancestry PRS as an effective prostate cancer risk stratification tool across populations. A clinical study of PRS is warranted to determine whether the PRS could be used for risk-stratified screening and early detection. Funding: This work was supported by the National Cancer Institute at the National Institutes of Health (grant numbers U19 CA214253 to C.A.H., U01 CA257328 to C.A.H., U19 CA148537 to C.A.H., R01 CA165862 to C.A.H., K99 CA246063 to B.F.D, and T32CA229110 to F.C), the Prostate Cancer Foundation (grants 21YOUN11 to B.F.D. and 20CHAS03 to C.A.H.), the Achievement Rewards for College Scientists Foundation Los Angeles Founder Chapter to B.F.D, and the Million Veteran Program-MVP017. This research has been conducted using the UK Biobank Resource under application number 42195. This research is based on data from the Million Veteran Program, Office of Research and Development, and the Veterans Health Administration. This publication does not represent the views of the Department of Veteran Affairs or the United States Government.

Introduction

Prostate cancer is the second leading cause of cancer death and represents one of the largest health disparities in the United States, with African ancestry men having the highest incidence rates (Howlader, 2021). Genetic factors play an important role in prostate cancer susceptibility (Mucci et al., 2016; Conti et al., 2021) and racial/ethnic disparities in disease incidence (Conti et al., 2021). Polygenic risk scores (PRS), comprised of common genetic variants, have been shown to enable effective risk stratification for many common cancers (Kachuri et al., 2020; Mars et al., 2020; Balavarca et al., 2020; Pal Choudhury et al., 2020). We recently conducted a multi-ancestry genome-wide association study (GWAS), including 107,247 prostate cancer cases and 127,006 controls (75.8% of European ancestry, 11.7% of East Asian ancestry, 9.1% of African ancestry, and 3.4% Hispanic), where 269 common genetic variants were genome-wide significantly associated with prostate cancer risk (Conti et al., 2021). Although individual genetic variants modulate disease risk only marginally, the aggregated effect of these 269 risk variants, measured by a PRS, was found to stratify prostate cancer risk in independent samples of European and African ancestry (Conti et al., 2021; Plym et al., 2022). As a measure of genetic susceptibility to prostate cancer, the PRS could potentially be an effective tool to identify men across diverse populations at higher risk of developing prostate cancer and allow them to make more informed decisions regarding at what age(s) and how frequently to undergo prostate-specific antigen (PSA) screening.

In this investigation, we evaluated the previously developed multi-ancestry PRS in large independent samples of men from the Veteran Affairs Million Veteran Program (MVP; 21,078 cases and 284,177 controls, including 13,643 cases and 210,214 controls of European ancestry, 6353 cases and 53,362 controls of African ancestry, and 1082 cases and 20,601 controls from Hispanic populations) (Gaziano et al., 2016), the Men of African Descent and Carcinoma of the Prostate (MADCaP) Network (405 cases and 396 controls of African ancestry) (Harlemon et al., 2020), and the Maryland Prostate Cancer Case–Control Study (NCI-MD; 383 cases and 395 controls of African ancestry) (Smith et al., 2017; ‘Methods’). We also included, through meta-analysis, independent replication studies of the multi-ancestry PRS conducted to date in European (UK Biobank and Mass General Brigham [MGB] Biobank) and African ancestry populations (California and Uganda Prostate Cancer Study [CA UG] and MGB Biobank; ‘Methods’; Conti et al., 2021; Plym et al., 2022), bringing the total sample to 31,925 cases and 490,507 controls.

In each of the replication studies included in our analysis, the PRS was constructed by summing variant-specific weighted allelic dosages of the 269 prostate cancer risk variants using the multi-ancestry conditional weights generated from our previous GWAS for prostate cancer (‘Methods’). Within each ancestry population, the association of PRS on prostate cancer risk was evaluated separately in each study and combined in a fixed-effects inverse-variance-weighted meta-analysis. Age-stratified analyses were performed in two large replication studies, UK Biobank and MVP, to assess the age-specific effects of PRS on prostate cancer risk. The absolute risk of prostate cancer was calculated for a given age for each PRS category in men from European, African and Hispanic populations (Antoniou et al., 2010; Kuchenbaecker et al., 2017; Amin Al Olama et al., 2015; Antoniou et al., 2001) using age- and population-specific prostate cancer incidence from the Surveillance, Epidemiology, and End Results (SEER) Program (1999–2013) and age- and population-specific mortality rates from the National Center for Health Statistics, CDC (1999–2013). The PRS was also tested for association with disease aggressiveness in MVP (‘Methods,’ Appendix 1—figure 1).

Appendix 1—figure 1.

Individual studies of European, African, or Hispanic population included in the polygenic risk score (PRS) association analysis.

Results from previous replication studies (*) in UK Biobank, Mass General Brigham (MGB) Biobank, and California and Uganda Prostate Cancer Study (CA UG) were meta-analyzed with results from Million Veteran Program (MVP), Maryland Prostate Cancer Case–Control Study (NCI-MD), and Men of African Descent and Carcinoma of the Prostate (MADCaP) Network within each ancestry population.

Results

The multi-ancestry PRS was strongly associated with prostate cancer risk in the three populations (Figure 1, Figure 1—source data 1). In European ancestry men, ORs were 3.78 (95% CI = 3.41–3.81) and 7.32 (95% CI = 6.76–7.92) for men in the top PRS decile (90–100%) and top percentile (99–100%), respectively, compared to men with average genetic risk (40–60% PRS category). In African ancestry men, ORs were 2.80 (95% CI = 2.49–2.95) and 4.98 (95% CI = 4.27–5.79) for men in the top PRS decile and percentile, respectively. In Hispanic men, ORs were 3.22 (95% CI = 2.64–3.92) and 6.91 (95% = 4.97–9.60) for men in the top PRS decile and percentile, respectively. PRS associations within each ancestry population were generally consistent across individual replication studies (Figure 1—figure supplement 1). The area under the curve (AUC) increased 0.136 on average across populations upon adding the PRS to a base model of age and principal components of ancestry (Appendix 1—table 1). Compared to the mean PRS in European ancestry controls, African ancestry controls had a mean PRS associated with a relative risk of 2.19 (95% CI = 2.17–2.21), while Hispanic controls had a relative risk of 1.16 (95% CI = 1.15–1.18), consistent with previous findings (Conti et al., 2021).

Figure 1.

Association between the multi-ancestry polygenic risk score (PRS) of 269 variants and prostate cancer risk in men from European, African, and Hispanic populations.

The European ancestry replication studies included Million Veteran Program (MVP), UK Biobank (Conti, Darst et al., Nature Genetics, 2021), and Mass General Brigham (MGB) Biobank (Plym et al., JNCI, 2021). The African ancestry replication studies included MVP, California and Uganda Prostate Cancer Study (CA UG) (Conti, Darst et al., Nature Genetics, 2021), Men of African Descent and Carcinoma of the Prostate (MADCaP) Network, Maryland Prostate Cancer Case–Control Study (NCI-MD), and MGB Biobank (Plym et al., JNCI, 2021). Replication in Hispanic men was conducted in MVP. Results from individual replication studies are shown in Figure 1—figure supplement 1. The x-axis indicates the PRS category. Additional analysis was performed to evaluate the PRS association in men with extremely high genetic risk (99–100%). The y-axis indicates OR with error bars representing 95% CIs for each PRS category compared to the 40–60% PRS. The dotted horizontal line corresponds to an OR of 1. ORs and 95% CIs for each decile are provided in Figure 1—source data 1.

Results in men of European ancestry were meta-analyzed across Million Veteran Program (MVP), UK Biobank, and Mass General Brigham (MGB) Biobank. Results in men of African ancestry were meta-analyzed across MVP, California and Uganda Prostate Cancer Study (CA UG), Maryland Prostate Cancer Case–Control Study (NCI-MD), Men of African Descent and Carcinoma of the Prostate (MADCaP) Network, and MGB Biobank. Results in Hispanic men were from MVP. The PRS association for men in the 99–100% category was not assessed in the MGB Biobank and therefore was not included in the meta-analysis. In each replication study, PRS categories were determined based on the distribution in controls. ORs and 95% CIs were estimated from logistic regression models adjusting for age and principal components of ancestry.

Replication studies in men of European and African ancestry included Million Veteran Program (MVP) (13,643 cases and 210,214 controls of European ancestry and 6353 cases and 53,362 controls of African ancestry), UK Biobank (6852 cases and 193,117 controls of European ancestry), Mass General Brigham (MGB) Biobank (67 cases and 457 controls of African ancestry and 1554 cases and 10,918 controls of European ancestry), California and Uganda Prostate Cancer Study (CA UG) (1586 cases and 1047 controls of African ancestry), Men of African Descent and Carcinoma of the Prostate (MADCaP) Network (405 cases and 396 controls of African ancestry), and Maryland Prostate Cancer Case–Control Study (NCI-MD) (383 cases and 395 controls of African ancestry). The x-axis indicates the PRS category. Additional analysis was performed to evaluate the PRS association in men with extremely high genetic risk (99–100%) in all individual studies except the MGB Biobank. The y-axis indicates OR with error bars representing the 95% CIs for each PRS category compared to the 40–60% PRS category. The dotted horizontal line corresponds to an OR of 1.

Figure 1—figure supplement 1.

Association between the multi-ancestry polygenic risk score (PRS) of 269 variants and prostate cancer risk from individual replication studies of European (A) and African ancestry (B).

Replication studies in men of European and African ancestry included Million Veteran Program (MVP) (13,643 cases and 210,214 controls of European ancestry and 6353 cases and 53,362 controls of African ancestry), UK Biobank (6852 cases and 193,117 controls of European ancestry), Mass General Brigham (MGB) Biobank (67 cases and 457 controls of African ancestry and 1554 cases and 10,918 controls of European ancestry), California and Uganda Prostate Cancer Study (CA UG) (1586 cases and 1047 controls of African ancestry), Men of African Descent and Carcinoma of the Prostate (MADCaP) Network (405 cases and 396 controls of African ancestry), and Maryland Prostate Cancer Case–Control Study (NCI-MD) (383 cases and 395 controls of African ancestry). The x-axis indicates the PRS category. Additional analysis was performed to evaluate the PRS association in men with extremely high genetic risk (99–100%) in all individual studies except the MGB Biobank. The y-axis indicates OR with error bars representing the 95% CIs for each PRS category compared to the 40–60% PRS category. The dotted horizontal line corresponds to an OR of 1.

Appendix 1—table 1.

Model discrimination and improvement estimated with area under the curve (AUC) upon adding the multi-ancestry polygenic risk score (PRS) to a base model in the Million Veteran Program (MVP) study populations.

		Age and PCs		Age, PCs, and PRS		AUC Change
Population	Sample	AUC	95% CI	AUC	95% CI
European ancestry	All cases and controls	0.582	(0.578–0.587)	0.694	(0.690–0.699)	+0.112
	Aggressive cases and controls	0.533	(0.521–0.545)	0.666	(0.655–0.677)	+0.133
	Nonaggressive cases and controls	0.603	(0.598–0.608)	0.703	(0.698–0.708)	+0.100
African ancestry	All cases and controls	0.512	(0.505–0.520)	0.656	(0.649–0.663)	+0.144
	Aggressive cases and controls	0.547	(0.531–0.564)	0.681	(0.665–0.697)	+0.134
	Nonaggressive cases and controls	0.522	(0.514–0.529)	0.657	(0.649–0.665)	+0.135
Hispanic	All cases and controls	0.530	(0.513–0.547)	0.683	(0.667–0.699)	+0.153
	Aggressive cases and controls	0.568	(0.531–0.607)	0.674	(0.636–0.712)	+0.106
	Nonaggressive cases and controls	0.514	(0.495–0.534)	0.685	(0.667–0.702)	+0.171

Abbreviation: PCs, principal components of ancestry.

Association between the multi-ancestry polygenic risk score (PRS) and prostate cancer risk replicated in men from European, African, and Hispanic populations.

Previously, we found that PRS associations were significantly stronger in younger men (aged ≤ 55 years) than in older men (aged > 55 years) (Conti et al., 2021). In the two large replication studies, UK Biobank and MVP, we further explored effect modification by age (Figure 2, Figure 2—figure supplement 1, Figure 2—source data 1). In European ancestry men, for the top PRS decile, the OR was 7.11 (95% CI = 5.82–8.70) in men aged ≤55, 4.26 (95% CI = 3.77–4.81) in men aged 55–60, and 2.79 (95% CI = 2.50–3.11) in men aged >70. The gradient in PRS risk by age was greater for men in the top PRS percentile, with ORs of 17.2 (95% CI = 13.0–22.8), 9.18 (95% CI = 7.52–11.2), and 5.43 (95% CI = 4.50–6.55) estimated for men ≤55, 55–60, and >70 years of age, respectively. Attenuation of PRS associations with age was also observed in African ancestry men as the OR for men in the top PRS decile decreased from 3.75 (95% CI = 3.04–4.64) in men aged ≤55 to 2.16 (95% CI = 1.76–4.68) in men aged >70. For African ancestry men in the top PRS percentile, the OR decreased from 8.80 (95% CI = 6.16–12.6) in men aged ≤55 to 2.87 (95% CI = 1.76–4.68) in men aged >70. A similar trend was observed in Hispanic men (OR = 6.37, 95% CI = 3.26–12.44 for men ≤55 and OR = 2.15, 95% CI = 1.39–3.32 for men >70 in the top PRS decile). Compared to men in the 40–60% PRS category, men from European, African, and Hispanic populations in the top PRS decile reached 5% absolute risk of prostate cancer 12 years earlier (age 57 vs. 69), 8 years earlier (age 55 vs. 63), and 11 years earlier (age 60 vs. 71), respectively (Table 1, Figure 3). For men in the top PRS percentile, 5% absolute risk was reached by ages 51, 52, and 53 for European, African, and Hispanic populations, respectively.

Figure 2.

Association between the multi-ancestry polygenic risk score (PRS) of 269 variants and prostate cancer risk stratified by age.

PRS associations in men of European ancestry (A) were meta-analyzed from UK Biobank (6852 cases and 193,117 controls) and Million Veteran Program (MVP) (13,643 cases and 210,214 controls; Figure 2—figure supplement 1), whereas PRS associations in men of African ancestry (B) were estimated from MVP (6353 cases and 53,362 controls). The x-axis indicates the PRS category. Additional analyses were performed to evaluate the PRS association in men with extremely high genetic risk (top percentile, 99–100%). The y-axis indicates the OR with error bars representing the 95% CIs for each PRS category compared to the 40–60% PRS category. The dotted horizontal line corresponds to an OR of 1. The number of cases and controls, ORs, and 95% CIs for each PRS category in each age stratum are provided in Figure 2—source data 1.

Results in men of European ancestry were meta-analyzed across UK Biobank and Million Veteran Program (MVP) while results in men of African and Hispanic ancestry were estimated in MVP only. In each replication study, PRS categories were determined based on the distribution in controls. ORs and 95% CIs were estimated from logistic regression models adjusting for age and principal components of ancestry.

The x-axis indicates the PRS category. Additional analysis was performed to evaluate the PRS association in men with extremely high genetic risk (99–100%). The y-axis indicates OR with error bars representing the 95% CIs for each PRS category compared to the 40–60% PRS category. The dotted horizontal line corresponds to an OR of 1.

Figure 2—figure supplement 1.

Association between the multi-ancestry polygenic risk score (PRS) of 269 variants and prostate cancer risk stratified by age in men of European ancestry from UK Biobank (A) and Million Veteran Program (MVP) (B).

The x-axis indicates the PRS category. Additional analysis was performed to evaluate the PRS association in men with extremely high genetic risk (99–100%). The y-axis indicates OR with error bars representing the 95% CIs for each PRS category compared to the 40–60% PRS category. The dotted horizontal line corresponds to an OR of 1.

Table 1.

Age at which 5% absolute risk of prostate cancer is reached in men from European, African, and Hispanic populations.

Absolute risks of prostate cancer were estimated using age- and population-specific Surveillance, Epidemiology, and End Results (SEER) incidence rates, CDC National Center for Health Statistics mortality rates, and polygenic risk score (PRS) associations from Figure 2—source data 1 based on Million Veteran Program (MVP) and the UK Biobank.

PRS category	European	African	Hispanic
[0–10]	>85	74	>85
(10–20%]	81	70	83
(20–30%]	75	67	77
(30–40%]	72	66	71
(40–60%]	69	63	71
(60–70%]	66	61	68
(70–80%]	65	59	66
(80–90%]	62	58	65
(90–100%]	57	55	60
(99–100%]	52	51	53

Figure 3.

Absolute risk of prostate cancer by polygenic risk score (PRS) category in men from European (A), African (B), and Hispanic populations (C).

The absolute risks were estimated using the age- and population-specific PRS associations from Figure 2—source data 1, the Surveillance, Epidemiology, and End Results (SEER) incidence rates, and the CDC mortality rates corresponding to non-Hispanic White, Black, and Hispanic men. The dotted line indicates the 5% absolute risk of prostate cancer.

Association of multi-ancestry polygenic risk score (PRS) and prostate cancer risk stratified by age.

Similar to previous findings (Conti et al., 2021; Plym et al., 2022), the multi-ancestry PRS did not consistently differentiate aggressive and nonaggressive prostate cancer risk (Appendix 1—table 2). For men in the top PRS decile, ORs were 3.17 (95% CI = 2.77–3.63) and 3.71 (95% CI = 3.48–3.94) for aggressive and nonaggressive prostate cancer in comparison to controls, respectively, in European ancestry men (P-heterogeneity = 0.04), and 1.92 (95% CI = 1.17–3.15) and 3.30 (95% CI = 2.64–4.12), respectively, in Hispanic men (P-heterogeneity = 0.05). In African ancestry men, the association was greater for aggressive (OR = 3.31, 95% CI = 2.71–4.03) than nonaggressive disease (OR = 2.66, 95% CI = 2.43–2.92), although confidence intervals overlapped (P-heterogeneity = 0.05).

Appendix 1—table 2.

The association between the multi-ancestry polygenic risk score (PRS) and prostate cancer aggressiveness in Million Veteran Program (MVP) participants from European, African, and Hispanic populations.

PRS categories were determined based on the distribution in controls in each replication study. ORs and 95% CIs were estimated from logistic regression models adjusting for age and principal components of ancestry. Heterogeneity was assessed via a Q statistic between effects estimates with corresponding tests of significance.

	Aggressive cases vs. controls					Nonaggressive cases vs. controls
PRS category	Controls	Cases	OR	95% CI	p-Value	Controls	Cases	OR	95% CI	p-Value	P-heterogeneity
European ancestry
[0–10%]	21,022	82	0.47	(0.37–0.59)	5.16E-10	21,022	258	0.31	(0.27–0.36)	4.43E-66	4.86E-03
(10–20%]	21,021	96	0.55	(0.44–0.68)	1.63E-07	21,021	423	0.51	(0.46–0.57)	6.29E-34	0.61
(20–30%]	21,021	118	0.67	(0.54–0.83)	1.76E-04	21,021	520	0.63	(0.57–0.70)	1.45E-19	0.60
(30–40%]	21,022	156	0.89	(0.73–1.07)	2.14E-01	21,022	656	0.79	(0.72–0.87)	8.87E-07	0.30
(40–60%]	42,042	352	1.00 (ref.)			42,042	1658	1.00 (ref.)
(60–70%]	21,022	245	1.39	(1.18–1.64)	7.32E-05	21,022	1120	1.35	(1.25–1.45)	6.55E-14	0.71
(70–80%]	21,021	272	1.55	(1.32–1.82)	7.13E-08	21,021	1392	1.67	(1.55–1.79)	1.84E-42	0.41
(80–90%]	21,021	335	1.91	(1.64–2.22)	4.51E-17	21,021	1801	2.15	(2.00–2.30)	5.69E-105	0.16
(90–100%]	21,022	554	3.17	(2.77–3.63)	2.72E-63	21,022	3151	3.71	(3.48–3.94)	<4.35E-283	0.04
(99–100%]	2,103	112	6.49	(5.22–8.07)	1.08E-63	2103	589	6.77	(6.10–7.51)	4.35E-283	0.73
African ancestry
[0–10%]	5,337	29	0.35	(0.24–0.53)	2.96E-07	5337	163	0.35	(0.30–0.42)	2.50E-33	0.98
(10–20%]	5,336	45	0.55	(0.40–0.77)	4.23E-04	5336	247	0.54	(0.46–0.62)	2.28E-17	0.88
(20–30%]	5,336	45	0.55	(0.40–0.77)	4.98E-04	5336	306	0.66	(0.58–0.76)	1.74E-09	0.33
(30–40%]	5,336	70	0.86	(0.65–1.14)	3.05E-01	5336	318	0.69	(0.61–0.79)	3.43E-08	0.16
(40–60%]	10,672	163	1.00 (ref.)			10,672	920	1.00 (ref.)
(60–70%]	5,336	121	1.50	(1.18–1.90)	8.14E-04	5336	556	1.21	(1.08–1.35)	7.65E-04	0.11
(70–80%]	5,336	131	1.62	(1.28–2.04)	4.97E-05	5336	659	1.43	(1.29–1.59)	1.88E-11	0.36
(80–90%]	5,336	151	1.89	(1.51–2.36)	2.59E-08	5336	819	1.78	(1.61–1.97)	9.21E-30	0.64
(90–100%]	5,337	262	3.31	(2.71–4.03)	4.66E-32	5337	1,224	2.66	(2.43–2.92)	8.99E-97	0.05
(99–100%]	534	45	5.840	(4.14–8.22)	5.79E-24	534	220	4.77	(4.02–5.66)	1.62E-71	0.30
Hispanic
[0–10%]	2061	2	0.12	(0.03–0.50)	3.69E-03	2061	21	0.31	(0.20–0.50)	8.67E-07	0.21
[10–20%]	2060	6	0.36	(0.15–0.87)	2.23E-02	2060	31	0.46	(0.31–0.69)	1.29E-04	0.61
(20–30%]	2060	6	0.36	(0.15–0.86)	2.12E-02	2060	47	0.70	(0.50–0.98)	4.00E-02	0.16
(30–40%]	2060	17	1.04	(0.58–1.87)	9.04E-01	2060	59	0.88	(0.65–1.21)	4.39E-01	0.64
(40–60%]	4,120	33	1.00 (ref.)			4120	133	1.00 (ref.)
(60–70%]	2060	20	1.21	(0.69–2.11)	5.05E-01	2060	85	1.28	(0.97–1.68)	8.50E-02	0.87
(70–80%]	2060	24	1.47	(0.86–2.49)	1.55E-01	2060	136	2.06	(1.61–2.63)	7.82E-09	0.26
(80–90%]	2060	22	1.33	(0.77–2.29)	3.01E-01	2060	136	2.05	(1.61–2.62)	8.69E-09	0.15
(90–100%]	2060	31	1.92	(1.17–3.15)	9.37E-03	2060	217	3.30	(2.64–4.12)	7.46E-26	0.05
(99–100%]	206	4	2.580	(0.91–7.38)	7.61E-02	206	46	7.15	(4.96–10.3)	3.99E-26	0.07

Discussion

Findings from this investigation provide further support for the PRS as a prostate cancer risk stratification tool in men from European, African, and Hispanic populations. Notably, this investigation provides the first evidence of replication of the multi-ancestry PRS in Hispanic men. Consistent with previous findings (Conti et al., 2021; Plym et al., 2022), we observed lower PRS performance in African versus European ancestry men, supporting the need to expand GWAS and fine-mapping efforts in African ancestry men. The stronger association of the PRS with prostate cancer risk observed for younger men supports previous studies (Conti et al., 2021), suggesting that the contribution of genetic factors to prostate cancer is greater at younger ages and that age needs to be considered when comparing PRS findings across studies and populations.

The PRS is an effective risk stratification tool for prostate cancer at both ends of the risk spectrum. Current guidelines consider age, self-reported race, and a family history of prostate cancer in PSA screening decisions (Schaeffer et al., 2021). Although the PRS generally did not differentiate aggressive versus nonaggressive prostate cancer, a substantial fraction of men who will develop aggressive tumors (~40%) are among a subset of men in the population with the highest PRS (top 20%; Appendix 1—table 2), while only ~7% of men who will develop aggressive tumors are among the subset of men in the population with the lowest PRS (bottom 20%; Appendix 1—table 2), suggesting that reduced screening among low PRS men may reduce the overdiagnosis of prostate cancer. Indeed, previous studies in men of European ancestry support that PRS-stratified screening could significantly reduce the overdiagnosis of prostate cancer by 33–42%, with the largest reduction observed in men with lower genetic risk (Pashayan et al., 2015a; Callender et al., 2019; Pashayan et al., 2015b). Risk-stratified screening studies are warranted in diverse populations to evaluate the clinical utility of this multi-ancestry PRS for early disease detection and when in a man’s life genetic risk should be considered in the shared decision-making process of prostate cancer screening.

Materials and methods

Participants and genetic data

We replicated the association between the multi-ancestry PRS and prostate cancer risk in three independent case–control samples from the VA MVP, the MADCaP Network, and the NCI-MD, as described below. Previously, this multi-ancestry PRS was replicated by our group and others in the CA UG (1586 cases and 1047 controls of African ancestry), the UK Biobank (6852 cases and 193,117 controls of European ancestry; updates to the UK Biobank led to slightly different sample sizes in this study of 8483 cases and 193,744 controls of European ancestry), and the MGB (formerly known as the Partners Healthcare Biobank, 67 cases and 457 controls of African ancestry and 1554 cases and 10,918 controls of European ancestry). Results from these studies are described in detail elsewhere (Conti et al., 2021; Plym et al., 2022). To provide a comprehensive assessment of the PRS validation, we meta-analyzed all replication studies, which included a total of 22,049 cases and 414,249 controls of European ancestry (UK Biobank, MGB Biobank, and MVP) and 8794 cases and 55,657 controls of African ancestry (MGB Biobank, MADCaP Network, NCI-MD, and MVP). In men of Hispanic ancestry, the multi-ancestry PRS was only assessed in MVP (1,082 cases and 20,601 controls).

All study protocols were approved by each site’s Institutional Review Board, and informed consent was obtained from all study participants in accordance with the principles outlined in the Declaration of Helsinki.

MVP

The design of the MVP has been previously described (Gaziano et al., 2016). Briefly, participants were recruited from approximately 60 Veteran Health Administration (VHA) facilities across the United States since 2011 with the current enrollment at >800,000. Informed consent was obtained for all participants to provide a blood sample for genetic analysis and access their full clinical and health data. The study received ethical and study protocol approval from the VA Central Institutional Review Board in accordance with the principles outlined in the Declaration of Helsinki.

A total of 485,856 samples from participants enrolled between 2011 and 2017 were genotyped on a custom Axiom array designed specifically for MVP (MVP 1.0). The genotyping array design and data quality controls were extensively described elsewhere (Hunter-Zinck et al., 2020). After excluding variants with high genotype missingness (>5%) and those that deviated from the expected allele frequency observed in the reference populations, genotype data were imputed to the 1000 Genomes Project Phase 3 reference panel (1000 Genomes Project Consortium, 2015). In MVP, genetic ancestry was assessed using HARE (Fang et al., 2019), which assigned >98% of participants with genotype data to one of four nonoverlapping population groups: non-Hispanic White (European), non-Hispanic Black (African), Hispanic, and non-Hispanic Asian. Due to the small number of non-Hispanic Asian individuals, they were excluded from the current analysis.

We identified a total of 21,078 cases and 284,177 controls from MVP, of whom 13,643 cases and 210,214 controls were of European ancestry (73.3%), 6353 cases and 53,362 controls were of African ancestry (19.6%), and 1082 cases and 20,601 controls were Hispanic (7.1%). Prostate cancer cases were identified from the Veterans Affairs Central Cancer Registry (VACCR), which collects cancer diagnosis, extent of disease and staging, first course of treatment, and outcomes from 132 VA medical centers. In this analysis, we only included cases from the VACCR who have a confirmed cancer diagnosis based on their diagnostic code, procedure code, and information from other clinical documents. Among the MVP participants without any prostate cancer diagnostic codes, we limited controls to those aged 45–95 years and had at least one prostate-specific antigen (PSA) test after enrollment. For prostate cancer cases, we obtained additional information on cancer staging and Gleason score to define aggressive prostate cancer phenotypes. Specifically, prostate cancer was considered aggressive if one of the following criteria was met: tumor stage T3/T4, regional lymph node involvement (N1), metastatic disease (M1), or Gleason score ≥8.0. Nonaggressive cases were defined as tumor stage T1/T2 and Gleason score <7.

MADCaP

The MADCaP Network dataset included 405 prostate cancer cases and 396 controls from sub-Saharan Africa, as previously described (Harlemon et al., 2020; Andrews et al., 2018), with a substantial proportion of cases diagnosed at late stages. The study protocol was approved by each study site’s Institutional Review Board/Ethnic Review Board. Written informed consent was obtained from all participants, and studies were conducted in concordance with the Declaration of Helsinki and the U.S. Common Rule. The MADCaP samples were genotyped on a customized array designed to capture common genetic variation in diverse African populations, and genotyping and quality control have been described in detail elsewhere (Harlemon et al., 2020). GWAS data were imputed using the 1000 Genomes Project Phase 3 reference panel (1000 Genomes Project Consortium, 2015).

NCI-MD

The NCI-MD Study included 383 prostate cancer cases identified from two Maryland hospitals and 395 population-based controls from Maryland and its neighboring states (Smith et al., 2017). The study was approved by the NCI (protocol # 05C-N021) and the University of Maryland (protocol #0298229) Institutional Review Boards. Informed consent was obtained from all participants. About 87% of the cases in this study were considered nonaggressive, with pathologically confirmed T1 or T2 tumor and a Gleason score ≤7. All samples from this study were genotyped on the Illumina InfiniumOmni5Exome array and were imputed to the 1000 Genomes Project Phase 3 reference panel (1000 Genomes Project Consortium, 2015).

PRS construction and association analyses

PRSs were constructed by summing variant-specific weighted allelic dosages from 269 previously identified prostate cancer risk variants (Conti et al., 2021). Variants were weighted using the multi-ancestry conditional weights generated from our previous trans-ancestry GWAS for prostate cancer (Conti et al., 2021). Variants and weights used to generate the PRS can be found in the PGS Catalog: https://www.pgscatalog.org/publication/PGP000122/.

The association of PRS on prostate cancer risk (i.e., case–control status) was estimated separately in each replication study using an indicator variable for the percentile categories of the PRS distribution: [0–10%], [10%–20%], [20%–30%], [30%–40%], [40%–60%], [60%–70%], [70%–80%], [80%–90%], and [90%–100%], where parentheses indicate greater than and square brackets indicate less than or equal to. Additional analysis was performed to obtain the association for the top 1% PRS by splitting the top PRS decile into [90%–99%] and [99%–100%] categories. PRS thresholds were determined in the observed distribution among controls in each study. In all replication studies, logistic regression was performed with the case–control status as the outcome (a binary dependent variable) and the PRS categories as independent predictors, adjusting for age and the up to 10 principal components of ancestry, with the [40%–60%] category as the reference. Age was defined as age at diagnosis for prostate cancer cases and age at last PSA testing (MVP) or age at study recruitment (MADCaP and NCI-MD) for controls.

Discriminative ability was evaluated in MVP by estimating the AUC for logistic regression models of prostate cancer that included covariates only (age and four principal components of ancestry) and for models that additionally included the PRS. All analyses were performed separately within each population.

We performed a fixed-effects inverse-variance-weighted meta-analysis to combine the ORs and standard errors for each PRS decile from individual replication studies by ancestry using R package meta (Schwarzer et al., 2015). This meta-analysis was conducted across the three studies of European ancestry, UK Biobank, MGB Biobank, and MVP, as well as across the five studies of African ancestry, MGB Biobank, CA UG, MADCaP Network, NCI-MD, and MVP.

In the two large replication studies, UK Biobank and MVP, logistic regression analyses were repeated stratifying both cases and controls at ages ≤55, (55–60], (60–65], (65–70], and >70, with adjustments for age (as a continuous variable) and the top principal components of ancestry. The PRS associations estimated in men of European ancestry from UK Biobank and MVP were meta-analyzed using a fixed-effects inverse-variance-weighted method. Heterogeneity between studies and across strata was assessed via a Q statistic between effects estimates with corresponding tests of significance (Schwarzer et al., 2015).

In the three ancestry populations from MVP, we also performed stratified analyses by disease aggressiveness, where cases were stratified as aggressive or nonaggressive and all controls were used in the corresponding stratified analysis. In both the aggressive cases vs. controls and nonaggressive cases vs. controls analyses, logistic regression was performed with the case–control status as the outcome (a binary dependent variable) and the PRS categories as independent predictors, adjusting for age and the up to 10 principal components of ancestry, with the (40–60%] category as the reference. Heterogeneity across strata was assessed via a Q statistic between effects estimates with corresponding tests of significance (Schwarzer et al., 2015).

Estimation of absolute risk

The absolute risk of prostate cancer was calculated for a given age for each PRS category in European, African, and Hispanic ancestry men (Antoniou et al., 2010; Kuchenbaecker et al., 2017; Amin Al Olama et al., 2015; Antoniou et al., 2001). The approach constrains the PRS-specific absolute risks for a given age to be equivalent to the age-specific incidences for the entire population, such that age-specific incidence rates are calculated to increase or decrease based on the estimated risk of the PRS category and the proportion of the population within the PRS category. The calculation accounts for competing causes of death.

Specifically, for a given population and PRS category k (e.g., 80–90%, 90–100%), the absolute risk by age t is computed as . This calculation consists of three components:

is the probability of not dying from another cause of death by age t using age-specific mortality rates, : . In this analysis, the age-specific mortality rates from the National Center for Health Statistics, CDC (1999–2013) were used.

is the probability of surviving prostate cancer by age t in the PRS category k and uses the prostate cancer incidence by age t for category k: .

The prostate cancer incidence by age t for PRS category k is and is calculated by multiplying the population prostate cancer incidence for the reference category, and the corresponding risk ratio, , for PRS category k and age category a (e.g., ages ≤55, 55–60, 60–65, 65–70, and >70) containing age t. These are estimated from the odds ratio obtained from the population-specific individual-level PRS analysis for each age-stratum (African and Hispanic ancestry odds ratios from MVP and European ancestry odds ratios meta-analyzed from MVP and UK Biobank): .

Prostate cancer incidence for age t for the reference category, was obtained by constraining the weighted average of the population cancer incidences for the PRS categories to the population age-specific prostate cancer incidence, . , where is the frequency of the PRS category k with for all nonreference categories in our primary PRS analysis by deciles (e.g., 0–10%, 10–20%, 20–30%, etc.).

By leveraging the definition that , for all k, the absolute risks were calculated iteratively by first getting , then , then , and finally . Subsequent values were then calculated recursively for all t.

For each population, absolute risks by age t were calculated using age- and population-specific prostate cancer incidence, , from theSEER program (1999–2013) and age- and population-specific mortality rates, , from the National Center for Health Statistics, CDC (1999–2013).

This article is mainly for an audience of genetic epidemiologists interested in the evaluation and portability of polygenic scores. The authors rigorously estimate the association of their multi-ancestry polygenic risk scores (PRS) for prostate cancer across multiple ancestries in a meta-analysis and show effect modification by age. The authors show that their PRS is effective in risk stratification for prostate cancer.

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Validation of a Multi-Ancestry Polygenic Risk Score and Age-Specific Risks of Prostate Cancer: A Meta-analysis Within Diverse Populations" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Ricardo Azziz as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Arjun Bhattacharya (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

The reviewers have agreed that the work presented here is an important contribution to the literature on polygenic risk scores and their applicability to cohorts of different ancestries. They also have agreed that the manuscript is clear, strong, and informative, and have only a small number of suggestions for improving clarity.

1. A reviewer mentions that added context for PRS is needed in the introduction. At the moment, the paper seems written for cancer epidemiology or genetic epidemiology audience, but it would be helpful if it were clear for a broader life sciences audience, who may not be completely aware of the utility of a PRS or may have preconceived notions about PRS research. Could the authors please add explanations and/or figures in the Introduction for:

– Explaining what a PRS is,

– what a PRS estimates,

– how the specific PRS was trained,

– the utility of a PRS in a clinical setting.

2. Can the authors also please briefly state how "prostate cancer risk" is calculated, either in the main text or in "Materials and methods" to guide the reader?

3. Could the authors please address how the results would be different if a random-effects meta-analysis model were to be used? A reviewer mentions that there is substantial heterogeneity in the environmental effects on prostate cancer risk across these different populations.

4. Have the authors looked into the PRS performance across ancestry proportion estimates and PRS associations with prostate cancer susceptibility across bins of ancestry proportions (i.e., AFR ancestry 0-20%, 20-40%, etc)? Could they comment a little on this topic? This could be an interesting follow-up.

5. Would it make sense to model select on other demographic/clinical covariates (e.g., SES measures) when estimating the odds ratios?

Other details

1. Can the authors please clarify what is the difference between Figure 1 and Figure 1 – source data 1? The caption of Figure 1 states that "ORs and 95% CIs for each PRS category are provided in Figure 1 – source data 1." but they seem to be already present in Figure 1. Also, please clarify the difference between Figure 2 and Figure 2 – source data 1.

2. The acronym "Partners" is used in the legend of Figure 1 instead of the acronym "MGB" used in the main text.

3. Page 20.- "Age-specific mortality rates are provided from a reference cohort." A reviewer recommends specifying which reference cohort is used.

Essential revisions:

The reviewers have agreed that the work presented here is an important contribution to the literature on polygenic risk scores and their applicability to cohorts of different ancestries. They also have agreed that the manuscript is clear, strong, and informative, and have only a small number of suggestions for improving clarity.

1. A reviewer mentions that added context for PRS is needed in the introduction. At the moment, the paper seems written for cancer epidemiology or genetic epidemiology audience, but it would be helpful if it were clear for a broader life sciences audience, who may not be completely aware of the utility of a PRS or may have preconceived notions about PRS research. Could the authors please add explanations and/or figures in the Introduction for:

– Explaining what a PRS is,

– what a PRS estimates,

– how the specific PRS was trained,

– the utility of a PRS in a clinical setting.

In response to this comment, in the Introduction section we include more background information regarding the development and validation of the multi-ancestry PRS for prostate cancer and what the PRS estimates (page 5). Further discussion on the clinical utility of this multi-ancestry PRS in identifying individuals at high risk of developing prostate cancer is included in the last paragraph of the Discussion section (page 9).

2. Can the authors also please briefly state how "prostate cancer risk" is calculated, either in the main text or in "Materials and methods" to guide the reader?

In our statistical analyses, “prostate cancer risk” refers to the relative risk estimated by the odds ratio calculated from a prostate cancer case-control analysis in each replication study. In all analyses assessing the association of PRS with prostate cancer risk, logistic regression models were used to estimate the odds ratio with case-control status as the outcome (a binary dependent variable) and the PRS categories as independent predictors, adjusting for age and up to ten principal components. The text in the Materials and methods section is modified to elaborate on these details in the statistical analysis (page 13 and page 14).

3. Could the authors please address how the results would be different if a random-effects meta-analysis model were to be used? A reviewer mentions that there is substantial heterogeneity in the environmental effects on prostate cancer risk across these different populations.

Within the European (UK Biobank, MVP, and MGB Biobank) and African population (MVP, MGB Biobank, MADCaP Network, CA UG, and NCI-MD) population, we also meta-analyzed the PRS associations with prostate cancer risk from individual replication studies using a random-effects method. We didn’t observe appreciable differences between the results from a fixed-effects and a random-effects meta-analysis in these two ancestry populations (see , Author response table 1). Given the similarities between these results, we only reported the PRS associations from the fixed-effects inverse-variance-weighted meta-analysis.

Author response table 1.

Association of the Multi-ancestry PRS with Prostate Cancer Risk in men of European and African Ancestry.

The PRS association on prostate cancer risk estimated from individual replication studies within each ancestry population were meta-analyzed using (a) a fixed-effects method and (b) a random-effects method.

	European Ancestry							African Ancestry
	22,049 cases, 414,249 controls							8,794 cases, 55,657 controls
	Fixed-Effects Meta-analysis				Random-Effects Meta-analysis			Fixed-Effects Meta-analysis				Random-Effects Meta-analysis
PRS Category	OR	(95% CI)	P value		OR	(95% CI)	P value	OR	(95% CI)	P value		OR	(95% CI)	P value
[0–10%]	0.32	(0.29–0.35)	5.67E-126		0.32	(0.28–0.36)	3.72E-60	0.37	(0.32–0.43)	4.17E-41		0.39	(0.27–0.54)	6.49E-08
(10%–20%]	0.48	(0.44–0.51)	5.74E-77		0.47	(0.40–0.55)	2.49E-20	0.55	(0.48–0.62)	9.75E-22		0.58	(0.41–0.81)	1.58E-03
(20%–30%]	0.64	(0.59–0.68)	9.81E-35		0.64	(0.59–0.68)	6.29E-34	0.66	(0.59–0.74)	1.61E-12		0.65	(0.54–0.79)	6.85E-06
(30%–40%]	0.79	(0.75–0.85)	2.47E-12		0.78	(0.71–0.87)	1.12E-06	0.7	(0.62–0.78)	1.16E-10		0.7	(0.62–0.78)	8.81E-10
(40%–60%]	1.00 (reference)			1.00 (reference)				1.00 (reference)			1.00 (reference)
(60%–70%]	1.33	(1.26–1.40)	8.90E-24		1.32	(1.23–1.42)	5.44E-15	1.23	(1.11–1.35)	3.01E-05		1.26	(1.05–1.51)	1.09E-02
(70%–80%]	1.62	(1.54–1.71)	3.26E-72		1.61	(1.48–1.75)	5.21E-29	1.45	(1.32–1.59)	6.00E-16		1.47	(1.27–1.70)	3.13E-07
(80%–90%]	2.18	(2.08–2.29)	2.28E-216		2.16	(1.88–2.48)	4.28E-27	1.79	(1.64–1.95)	7.51E-40		1.8	(1.62–2.00)	3.36E-27
(90%–100%]	3.78	(3.62–3.96)	<5.00E-324		3.86	(3.52–4.23)	2.30E-184	2.8	(2.59–3.03)	1.38E-144		2.97	(2.57–3.44)	1.46E-49
(99%–100%]	7.32	(6.76–7.92)	<5.00E-324		7.7	(5.74–10.33)	3.28E-42	4.98	(4.27–5.79)	5.02E-95		5.17	(3.87–6.92)	1.57E-28

4. Have the authors looked into the PRS performance across ancestry proportion estimates and PRS associations with prostate cancer susceptibility across bins of ancestry proportions (i.e., AFR ancestry 0-20%, 20-40%, etc)? Could they comment a little on this topic? This could be an interesting follow-up.

We appreciate the reviewer’s insightful suggestion. In three of the five replication studies included in the African-ancestry meta-analysis (CA UG, MADCaP Network, and NCI-MD), we estimated the proportion of African ancestry (%AFR) from an unsupervised (K=2) ADMIXTURE analysis using the 1000 Genomes Project phase 3 European and African samples as the reference populations. We do not have access to the admixture information in the MVP and MGB Biobank data.

The %AFR was higher and less dispersed in MADCaP (mean = 95.7%, standard deviation [SD] = 4.6%) than in CA UG (mean = 79.2%, SD = 13.5%) and NCI-MD (mean = 77.2%, SD = 10.2%; see Author response table 2). Despite the differences in %AFR, we found that the PRS associations with prostate cancer risk were similar across these three studies, particularly in the top PRS decile (Figure 1 —figure supplementary 1). Although we were not able to formally test the effect modification of African ancestry on the PRS association due to the limited sample sizes of these three studies, the similar PRS associations observed in these studies support the robustness of this multi-ancestry PRS in risk stratification across African populations with varying degrees of admixture. Future investigations with sufficient sample sizes are warranted to better understand the interaction between admixture and PRS.

Author response table 2.

	Sample Size		% AFR Ancestry
	No. Case	No. Control	Mean	SD
CA UG	1,586	1,047	79.2	13.5
MADCaP Network	405	396	95.7	4.6
NCI-MD	383	395	77.3	10.2

5. Would it make sense to model select on other demographic/clinical covariates (e.g., SES measures) when estimating the odds ratios?

For prostate cancer, only age, family history, and race are considered well-established risk factors. Given the potential correlation between family history and PRS, family history of prostate cancer is typically not adjusted for in the association analysis of PRS. Other demographic or clinical factors are unlikely to be associated with genotype status and thus not considered as potential confounders in the association between PRS and prostate cancer risk. Therefore, adjusting for these variables will not be expected to influence PRS associations.

Other details

1. Can the authors please clarify what is the difference between Figure 1 and Figure 1 – source data 1? The caption of Figure 1 states that "ORs and 95% CIs for each PRS category are provided in Figure 1 – source data 1." but they seem to be already present in Figure 1. Also, please clarify the difference between Figure 2 and Figure 2 – source data 1.

Figure 1 is a graphic presentation of the results in Figure 1 – source data 1. Figure 1 – source data 1 provides the actual numbers of odds ratios (ORs), 95% confidence intervals (CIs), and P values for each PRS category in each ancestry population, which were cited in the main text. Although they present the same PRS association results, the detailed information included in Figure 1 – source data 1 would allow direct comparison or meta-analysis with other/future replication studies. The same applies to Figure 2 and Figure 2 – source data 2. We believe it is important to include both the figures and the tables (source data) in the manuscript to provide sufficient information for interpretation.

2. The acronym "Partners" is used in the legend of Figure 1 instead of the acronym "MGB" used in the main text.

We thank the reviewer for spotting this error. The legend of Figure 1 – source data 1 has been updated to be consistent with the main text, replacing “Partners” with “MGB Biobank” (page 22).

3. Page 20.- "Age-specific mortality rates are provided from a reference cohort." A reviewer recommends specifying which reference cohort is used.

We modified the text in the Materials and methods (page 15) to specify that the age-specific mortality rates from the National Cancer for Health Statistics, CDC (1993-2013) were used in the calculation of absolute risk of prostate cancer.