Restricted Versus Genome‐Wide Genetic Risk Scores for Coronary Artery Disease

Abstract

Background

Genetic risk scores may be useful for analyzing risks for coronary artery disease (CAD). However, comparisons between restricted and genome‐wide scores have been underexplored, particularly for individuals at increased risk by one score but not the other. Here, we compared restricted polygenic risk scores with 181 high‐confidence genetic variants (PRS₁₈₁) and genome‐wide risk scores that encompass 6.6 million single‐nucleotide polymorphisms (GRS_6.6M).

Methods

Data were from the RS (Rotterdam Study; n=11 001), MESA (Multi‐Ethnic Study of Atherosclerosis; n=2685), and the Sanford Health study (n=25 166). We analyzed score associations with CAD (prevalent and incident), age at onset, and lipid medication use. Combined use of both scores was also examined.

Results

There were robust associations with CAD per SD of the scores for men (PRS₁₈₁: hazard ratio [HR], 1.19 [95% CI, 1.13–1.26]; GRS_6.6M: HR, 1.32 [95% CI, 1.26–1.39]) and women (PRS₁₈₁: HR, 1.24 [95% CI, 1.16–1.32]; GRS_6.6M: HR, 1.32 [95% CI, 1.25–1.40]). PRS₁₈₁ was more strongly associated with early‐onset CAD in men (β=−0.93 [95% CI, −1.36 to −0.50]) and women (β=−0.76 [95% CI, −1.31 to −0.21]). Both scores correlated with lipid medication use, but the scores were also associated with CAD among nonusers. Individuals at high risk by both scores had the highest risk and the earliest age at onset.

Conclusions

PRS₁₈₁ and GRS_6.6M appear to identify different subsets of individuals. Use of both scores together may provide better association information on CAD risk and age at onset than each score alone.

Nonstandard Abbreviations and Acronyms

GRS_6.6M

genome‐wide risk score that encompasses 6.6 million single‐nucleotide polymorphisms

MESA

Multi‐Ethnic Study of Atherosclerosis

NTR

The Netherlands National Trial Register

PRS₁₈₁

restricted polygenic risk scores with 181 high‐confidence genetic variants

RS

Rotterdam Study

Clinical Perspective.

What Is New?

• A restricted polygenic risk score with 181 high‐confidence genetic variants and a genome‐wide score that encompasses 6.6 million single‐nucleotide polymorphisms were both associated with coronary artery disease but identified only moderately overlapping subsets of at‐risk individuals.

What Are the Clinical Implications?

• Combining restricted and genome‐wide genetic scores may improve coronary artery disease risk prediction and identify individuals with earlier age at onset more effectively than either score alone.

• These findings highlight the value of using complementary genetic risk models and support the need for validation in diverse, real‐world populations before clinical adoption.

Coronary artery disease (CAD) is the predominant form of cardiovascular disease and a major cause of morbidity and death. ¹ Early identification of individuals at high risk for CAD should be useful for prevention. ² Genetic factors influence CAD development and progression, with heritability ranging from 40% to 60%. ³ , ⁴ Genome‐wide association studies can provide information on individual genetic predisposition through genetic risk scores. The scores sum the effects of single‐nucleotide polymorphisms (SNPs) determined from independent genome‐wide association study cohorts. Advances in the development of genetic risk scores are leading to their use in diagnosis, intervention decisions, and targeted screening. ⁵ , ⁶ , ⁷ , ⁸ , ⁹

The number of SNPs included in genetic risk scores may vary. ¹⁰ , ¹¹ , ¹² Restricted polygenic risk scores typically include a carefully selected subset of independent genetic variants that meet stringent significance thresholds. ⁸ , ¹³ , ¹⁴ , ¹⁵ Genome‐wide risk scores use millions of SNPs across the genome and may include variants that are in linkage disequilibrium. Genome‐wide risk scores often produce higher risk estimates. ¹⁰ , ¹³ However, it remains unclear how genome‐wide risk scores compare with polygenic risk scores in identifying individuals at risk for CAD, particularly in relation to age at onset or decisions to use lipid medications. There are also questions on risk score performance across demographic factors, such as population structure, age, and sex. ¹⁶ , ¹⁷ , ¹⁸

Our study evaluated 2 distinct genetic risk scores for CAD: the restricted polygenic risk score, which includes 181 high‐confidence genetic variants (PRS₁₈₁) selected from 12 studies, ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ and the genome‐wide risk score, which encompasses 6.6 million SNPs (GRS_6.6M). ¹⁴ Our aim is to better understand their strengths and limitations, focusing on their associations with CAD, age at onset, and lipid medication use.

We used 2 community‐based cohorts (1 in the Netherlands and 1 in the United States) and a US health care system biobank to examine how PRS₁₈₁ and GRS_6.6M perform in different settings. Our findings support an interpretation that these scores may identify different subsets of at‐risk individuals.

Methods

Data Availability

The data that support the findings of this study are available upon reasonable request. Requests should be directed to the management team of the RS (Rotterdam Study), MESA (Multi‐Ethnic Study of Atherosclerosis), or Sanford Health, which have protocols for approving data requests. Because of restrictions based on privacy regulations and informed consent of the participants, data cannot be made freely available in a public repository.

Study Cohorts

The RS is a prospective, single‐center, community‐based cohort of men and women in the Ommoord municipality in the city of Rotterdam, the Netherlands. ³¹ The study is on the incidence and prevalence of all age‐related diseases and their risk factors. There are several subcohorts, each starting in subsequent years and with slightly different inclusion criteria. All subcohorts have longitudinal follow‐up measurements. We used 3 subcohorts for the study. RS‐I started in 1989 (7983 participants). RS‐II commenced in 2000 (3011 participants) and RS‐III in 2007 (3932 participants). The minimum baseline age was 55 years for RS‐I and RS‐II and 45 years for RS‐III. ³¹

Baseline measurements were done through a home interview. Participants visited the research center for extensive physical examinations, imaging, and laboratory assessments, including numerous omics determinations (such as genotypes from whole‐blood DNA). The RS consists mainly of White participants of European background. A total of 11 001 White participants (of 14 926 White participants) were selected for the study on the basis of having SNP array genotype data. The analyzed data set consisted of individuals who clustered with the European ancestry group on the basis of principal component analysis. ³²

MESA examines the characteristics of subclinical cardiovascular disease (detected noninvasively before clinical signs and symptoms) and the risk factors that predict progression to overt disease. The cohort is a community‐based sample of 6814 men and women aged 45 to 84 years without known clinical cardiovascular disease. Self‐reported races and ethnicities are White (38%), Black (28%), Hispanic (22%), and Asian (12%; Chinese descent). Participants were recruited from 6 field centers across the United States: Wake Forest University, Columbia University, Johns Hopkins University, University of Minnesota, Northwestern University, and University of California–Los Angeles. ³³ The analysis here was limited to White individuals (2685 participants).

Each MESA participant had an examination and assessment of standard coronary risk factors as well as sociodemographic, lifestyle, and psychosocial factors. Selected subclinical disease measures and assessment of risk factors are repeated at follow‐up visits. ³³ Examination 1 occurred in 2000 to 2001. Blood samples have been assayed for putative biochemical risk factors and stored. DNA has been extracted, and lymphocytes have been cryopreserved.

The Sanford Health population is a cohort of 26 291 European‐origin patients from the Sanford Biobank and the Imagenetics program. The biobank cohort includes participants aged ≥18 years from the Sanford Health system who elected to provide a specimen to the Sanford Biobank for research purposes, starting in 2011. The Imagenetics cohort includes participants who opted into a preemptive genetic screening program, starting in 2017. The participants were considered to be entered into the Sanford Imagenetics program or the Sanford Biobank at the time of sample collection for genotyping. Sanford Health includes members from the states of North Dakota, South Dakota, Minnesota, and Nebraska.

Coronary Artery Disease Assessment

Rotterdam Study

Information on cardiovascular outcomes was collected through continuous linkage with general practitioner records, hospital discharge letters, and specialist reports. All events were adjudicated by trained study physicians using standardized protocols. For this study, prevalent CAD was defined using the CHD variable, which includes a history of myocardial infarction (MI), percutaneous coronary intervention, or coronary artery bypass grafting at the time of study entry. Incident CAD was defined using the MI variable, which captures only definite nonfatal or fatal MI or death due to CAD during follow‐up. These events were adjudicated on the basis of discharge documentation, ECG changes, cardiac enzyme levels, and, when applicable, autopsy findings, consistent with the universal definition of MI. ³⁴ , ³⁵

Percutaneous coronary intervention and coronary artery bypass grafting procedures were not included in the incident CAD definition, because procedure data from earlier decades may be incomplete or inconsistently recorded. We restricted the incident CAD definition to definite MI events for better diagnostic specificity and better comparability across cohorts and time. Although the CAD definitions for prevalent versus incident cases differ somewhat, they allow more reliable time‐to‐event analyses.

Only first CAD events were included in the analyses. Age at onset for incident cases was based on the date of the first MI. Data on lipid‐lowering medication use were self‐reported at baseline.

Multi‐Ethnic Study of Atherosclerosis

Participants were followed for characterization of cardiovascular disease events, including acute MI and other forms of coronary heart disease, stroke, congestive heart failure, cardiovascular disease interventions, and death. Hospital records with information on potential cardiovascular events were abstracted by qualified personnel, who identified and extracted relevant materials, such as ECGs and discharge summaries. Two physicians with expertise in the field independently reviewed the abstracted records to classify cardiovascular events on the basis of predefined criteria. When there was disagreement between the reviewing physicians, a consensus was reached through further discussion and involvement of the full events committee if necessary. Causes of death obtained from death certificates, medical records, and interviews with next‐of‐kin were used as adjudication criteria for classification of cardiovascular deaths. ³⁶

Incident CAD was defined as occurrence of MI, percutaneous coronary intervention, coronary artery bypass grafting, or coronary death. No participant had a cardiovascular diagnosis at baseline. ³³ Data on baseline lipid medication use were from reviews of medications, which are done during each MESA examination.

Sanford Health Study

Identification of controls and CAD cases was by International Classification of Diseases, Tenth Revision, Clinical Modification (ICD‐10‐CM) codes from the linked electronic medical record (Table S1). ICD‐10‐CM codes that supported classification of participants as CAD cases were codes indicating (1) occurrence of MI (I21.02, I21.09, I21.11, I21.19, I21.21, I21.29, I21.3, I21.4, I21.9, I21.A1, I23.3, I23.6, I23.7, I25.2); (2) performance of coronary artery bypass grafting (I25.708, I25.709, I25.710, I25.719); and (3) medical necessity of percutaneous coronary intervention (I25.110, I25.111, I25.118, I25.119). Sanford Health controls had no ICD‐10‐CM codes for these CAD events. Data on lipid medication use were not available. The ICD‐10‐CM codes that support medical necessity of percutaneous coronary intervention are listed at https://www.cms.gov/medicare‐coverage‐database/view/article.aspx?articleid=57479&ver=13&.

Genotyping

Participants were genotyped by use of the Illumina 550k, 610k, or Global Screening Array version 1 arrays, or the Affymetrix Genome‐Wide Human SNP Array 6.0. Imputation was with the Haplotype Reference Consortium Release 1.1 or 1000 Genomes reference panel. Sample and variant quality control (QC) in the RS and MESA were performed as described. ³⁷ , ³⁸

QC of samples at Sanford Health was performed using the Sanford CHIP Clinical Bioinformatics Pipeline, which compares the sample ID, estimated sex, and sample plate name with information in the Laboratory Information Management System. Occurrences of sample mix‐ups are excluded in this way. In the event of a possible sample mix‐up or poor‐quality metrics, additional header information, such as Sample Well, Cluster File, SNP Manifest, and Scanner Data, is generated and scrutinized. Additional QC metrics, including Contamination, Raw Control X, Raw Control Y, Log R Deviation, GC 10, GC 50, Intensity Percentiles X, and Intensity Percentiles Y, are computed and examined, if the sample does not pass QC standards. QC analysis of Sanford clinical and biobank samples was performed on the basis of sample ID, sex, and call rate. The Sample Management and Genotyping Laboratory cross‐checked the IDs and sexes against the internal list and set a call rate threshold of 0.99 for clinical samples and 0.75 for biobank samples. Samples that fulfilled these criteria were further analyzed for downstream analysis and identification.

Genetic Variant Selection

The PRS₁₈₁ values are based on 181 independent SNPs selected from 259 genome‐wide significant SNPs that were identified in 12 studies ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ (Table S2). We removed 78 SNPs from the original 259 by linkage disequilibrium pruning (r ²<0.2). Palindromic SNPs with allele frequencies between 0.35 and 0.65 were replaced by proxies (with r ²>0.8 and D′>0.8). All 181 variants were available in imputed data for MESA and Sanford Health (with imputation quality r ²>0.3). Therefore, MESA and Sanford Health PRS₁₈₁ values were calculated from 181 SNPs. But in the RS, 4 of the 181 SNPs were excluded, due to low imputation quality (r ²>0.8), and 7 SNPs were replaced by proxies (with r ² > 0.9 and D′ >0.9). Therefore, RS PRS₁₈₁ values were calculated from 177 SNPs.

The GRS_6.6M values are based on effect sizes from a model developed by Khera et al ¹⁴ (validated in the UK Biobank). The score uses 6 630 150 common genetic variants (with minor allele frequencies >1%). ¹⁴ , ²¹ , ³⁹ Briefly, they used the LDPred algorithm to compute candidate genetic risk scores, incorporating pruning and threshold derivation strategies. This Bayesian approach calculates a posterior mean effect size for each variant, taking into account correlation with similarly associated variants in the reference population (HapMap in the current case). Guided by a tuning parameter (ρ=0.001), the Gaussian distribution further accounts for the fraction of causal markers.

Genetic Risk Score Calculation

We calculated weighted continuous PRS₁₈₁ and GRS_6.6M values for CAD as follows:

$$\text{Genetic}{\text{Risk Score}}_i={\sum }_i^k\left({\widehat{\beta }}_j\times {\text{dosage}}_{\mathit{ij}}\right)$$

where Genetic Risk Score _i is the risk score for subject i; dosage _ij is the posterior probability of being a heterozygous (probability around 1.0) or homozygous (probability around 2.0) effect allele carrier after imputations for subject i and variant j; κ is the number of independent variants in the risk score for subject i; and ${\widehat{\beta }}_j$ is the weight for variant j obtained from genome‐wide association study summary statistics. The risk scores were standardized to a mean of 0± (Z transformation) for each of the 5 cohorts (RS‐I, RS‐II, RS‐III, MESA, and Sanford Health).

Statistical Analysis

Analyses were performed separately for each cohort (Table 1). A fixed‐effect meta‐analysis followed all cohort analyses. Individuals with missing genotype array data were excluded, as appropriate.

Table 1.

Overview of Coronary Artery Disease Analyses Performed for the PRS₁₈₁ and the GRS_6.6M

Item analyzed	Method	Results
Analyses of prevalent CAD in the RS and the Sanford Health study
CAD risk	Use continuous risk scores	Odds ratio per SD of the risk score (Table 2A)
Analyses of incident CAD in the RS, MESA, and the Sanford Health study
CAD risk	Use continuous risk scores	HR per SD of the risk score (Table 2B–2D)
Risk score percentiles*	Use categorical risk scores	HRs for risk score percentile cutoffs (Table 3)
Age at onset†	Use continuous risk scores and stratify by age at onset	HR per SD of the risk score (Table 2B–2D)
Age at onset†	Use continuous risk scores	β values and change in age at onset per SD of the risk score (Table 2E)
Lipid medication use‡	CAD risks in medication users vs nonusers, and	HR per SD of the risk score; use continuous risk scores (Table 4)
	associations with medication use in CAD cases vs controls
Combined use of PRS181 and GRS6.6M δ	Assign individuals to risk groups according to their PRS181 and GRS6.6M values (in SD units)	HR and mean age at onset for each risk group (Figure 1)

CAD indicates coronary artery disease; GRS_6.6M, genome‐wide polygenic risk score that encompasses 6.6 million single‐nucleotide polymorphisms; HR, hazard ratio; MESA, Multi‐Ethnic Study of Atherosclerosis; PRS₁₈₁, restricted polygenic risk score with 181 high‐confidence genetic variants (or 177 single‐nucleotide polymorphisms for the RS); and RS, Rotterdam Study.

^*Incident cases and controls were grouped according to percentiles of the risk score distribution, that is, >50th, 60th, 70th, 80th, 90th, 95th, and 98th percentiles. Each group was compared with the bottom 50% of the risk distribution.

^† Age at onset is defined as the age at the time of the first CAD event.

^‡ We stratified incident CAD cases and controls by baseline lipid medication use.

^δ The entire study population was stratified into 8 groups and a reference group, according to the values of PRS₁₈₁ and GRS_6.6M (in SD units). Individuals in the reference group have PRS₁₈₁ and GRS_6.6M values within ±1 SD of the mean for both scores.

Prevalent and Incident CAD

Age‐ (and sex‐) adjusted binomial generalized linear models and Cox proportional hazards models were used to evaluate associations between the Z‐transformed risk scores and prevalent or incident CAD, respectively.

Genetic Risk Score Percentiles

Incident cases and controls were grouped according to their risk score percentiles (ie, >50th, 60th, 70th, 80th, 90th, 95th, or 98th percentile). Each group was compared with the bottom 50% of the risk score distribution, which represents a low‐risk cohort. The comparison emphasizes identifying high‐risk individuals, which is useful for prevention and intervention.

Age at CAD Onset

Early‐onset CAD was defined as onset before age 60 years for both men and women. We used a cutoff of <60 years for both sexes (instead of <55 years for men and <60 years for women) to ensure adequate statistical power for subgroup and cohort analyses. We recognize that risk score effects in relation to age in men may be somewhat attenuated due to use of the older cutoff (<60 years). Gaussian generalized linear models were used to examine the effect of genetic risk scores on age at CAD onset.

Lipid Medication Use

We stratified incident CAD cases and controls according to lipid medication use. We assessed risk score associations with CAD (in relation to medication use) and risk score associations with medication use (in relation to CAD). Lipid medication use (or nonuse) represents a clinician’s overall impression of an individual’s CAD risk. The impression is expected to be based on lipid levels, age, family history, and the presence of other conditions (such as diabetes, hypertension, and smoking). However, our study did not analyze lipid levels (low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, or triglycerides). Our stratified approach aimed to identify genetic determinants associated with clinical CAD risk assessments. The analyses may also provide insights into the interplay between genetic predisposition and clinical decision making. ⁴⁰ , ⁴¹

Combined Use of PRS₁₈₁ and GRS_{6. 6M}

PRS₁₈₁ values were plotted against GRS_6.6M values (in SD units; Figure 1[A]). Participants with both risk scores within ±1 SD of the mean were designated as the reference group. Individuals with at least 1 of the 2 risk scores beyond ±1 SD of the mean (ie, at least 1 score was either greater than the mean+1 SD, or less than the mean−1 SD) were assigned to 1 of 8 groups: G1 (both PRS₁₈₁ and GRS_6.6M<−1 SD); G2 (PRS₁₈₁<−1 and GRS_6.6M within ±1 SD); G3 (PRS₁₈₁<−1 SD and GRS_6.6M>1 SD); G4 (PRS₁₈₁ within ±1 SD and GRS_6.6M<−1 SD); G5 (PRS₁₈₁ within ±1 SD and GRS_6.6M>1 SD); G6 (PRS₁₈₁>1 SD and GRS_6.6M<−1 SD); G7 (PRS₁₈₁>1 SD and GRS_6.6M within ±1 SD); or G8 (both >1 SD) (Figure 1[A]). In this way, the entire population was stratified according to PRS₁₈₁ and GRS_6.6M values. Groups G1 to G8 were each compared with the reference group.

Figure 1. Stratification of PRS₁₈₁ and GRS_6.6M and evaluation of effects of PRS₁₈₁ and GRS_6.6M together.

A, Scatter plot of PRS₁₈₁ vs GRS_6.6M in SD units. Individuals fall into groups G1 through G8 and the reference group, according to their PRS₁₈₁ and GRS_6.6M values (in SD units). The color of each plotted point corresponds to the average of the PRS₁₈₁ and GRS_6.6M values (ie, [ZPRS₁₈₁+ZGRS_6.6M]/2). Colors range from green (low risk) to gray (intermediate risk) to magenta (high risk), as shown in the inset. B, HRs for incident CAD and mean ages of CAD onset for risk groups G1 through G8 compared with the reference group. C, Summary of results on risk assessments that combine PRS₁₈₁ and GRS_6.6M. HRs in this figure reflect categorical comparisons between jointly defined risk groups and a common reference group. CAD indicates coronary artery disease; GRS_6.6M, genome‐wide polygenic risk score that encompasses 6.6 million single‐nucleotide polymorphisms; HR, hazard ratio; PRS₁₈₁, restricted polygenic risk score with 181 high‐confidence genetic variants.

Ethics Statement

Rotterdam Study

The study was approved by the Medical Ethics Committee of Erasmus Medical Center (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272‐159521‐PG). The RS has been entered into the NTR (Netherlands National Trial Register; www.trialregister.nl) and into the WHO International Clinical Trials Registry Platform (www.who.int/ictrp/network/primary/en/) under shared catalog number NTR6831. All participants provided written informed consent to be included in the study and to allow study personnel to obtain medical information from their treating physicians.

Multi‐Ethnic Study of Atherosclerosis

The studies involving human participants were reviewed and approved by institutional review boards. The participants provided written informed consent to be included in the study.

Sanford Health

The study was approved by the Sanford Health institutional review board. Information was recorded by investigators in such a manner that the identities of participants cannot be readily ascertained directly or through identifiers linked to individuals. Furthermore, investigators do not contact or reidentify participants. Therefore, a waiver of Health Insurance Portability and Accountability Act authorization and a waiver of consent were granted from the Sanford Health institutional review board. The project was categorized as exempt from review.

Results

Characteristics of Study Cohorts

Percentages of women, mean ages at baseline, mean years of follow‐up, percentages of individuals on lipid medication, and mean ages at CAD onset are listed in Table 2 and Table S3. MESA had the lowest percentage of women (54.3 among controls for incident CAD, compared with 58.8–65.0 for all other cohorts; Table 2). Sanford Health had the youngest cohort (mean age at baseline, 47.3 years among controls for incident CAD, compared with 56.8–69.2 years for the other cohorts; Table 2). Data on lipid medication use were not available for the Sanford Health cohort. MESA excluded individuals with prevalent CAD at baseline.

Table 2.

Characteristics of the Study Cohorts

	RS			MESA	Sanford
	RS‐I	RS‐II	RS‐III
Prevalent CAD
Cases	490	139	118	…	102
Women, n (%)	157 (32.0)	29 (20.9)	27 (22.9)		31 (30.4)
Age at baseline, mean±SD	55–92.6±70.9	56–89.3±69.1	46.2–86±61.6	…	26–84±63.4
Lipid medication, n (%)	316 (64.5)	79 (56.8)	103 (87.3)	…	…
Controls	5338	2012	2904	…	25 064
Women, n (%)	3312 (62.0)	1140 (56.7)	1678 (57.8)	…	16 071 (64.1)
Age at baseline, mean±SD	55–99.2±68.8	55.1–95.3±64.5	45.5–97.2±56.9	…	3–90±47.7
Years of follow‐up, mean±SD	0.1–24.7±14.4	0.1–15±11.9	0.8–9±7.1	…	0–17.9±10.5
Lipid medication, n (%)	2174±40.7	226±11.2	610±21.0	…	…
Incident CAD
Cases	1162	221	108	321	847
Women, n (%)	603 (51.9)	82 (37.1)	34 (31.5)	116 (36.1)	323 (38.1)
Age at baseline, mean±SD	55–96±69.7	55.3–90.9±66.9	46–80.9±59.1	45–84±66.2	23–90±61.1
Years of follow‐up, mean±SD	0.1–24.2±10.0	0.1–15±7.2	0.1–8.6±3.8	0.15–17.9±8.0	0–17.9±7.4
Lipid medication, n (%)	581 (50)	28 (12.7)	27 (25)	75 (25.3)	…
Age at onset, mean±SD	55.5–102.3±79.7	57.3–97.4±74.1	50.9–87.6±62.9	50.5–98.2±74.3	31–90±69.4
Controls	4550	1791	2796	2364	24 217
Women, n (%)	2947 (64.8)	1058 (59.1)	1644 (58.8)	1285 (54.3)	15 747 (65.0)
Age at baseline, mean±SD	55–99.2±69.2	55.1–95.3±64.2	45.5–97.2±56.8	44–87±62.3	3–90±47.3
Years of follow‐up, mean±SD	0.1–25.5±15.2	0.2–15±12.5	0.8–9±7.2	0–18.4±14.7	0–17.1±10.6
Lipid medication, n (%)	1859 (40.9)	198 (11.1)	583 (20.9)	434 (18.2)	…

CAD indicates coronary artery disease; MESA, Multi‐Ethnic Study of Atherosclerosis; and RS, Rotterdam Study.

All study cohorts had similar normal distributions of unstandardized weighted PRS₁₈₁ and GRS_6.6M values (not shown).

Prevalent and Incident CAD

Both PRS₁₈₁ and GRS_6.6M were associated with prevalent CAD. The odds ratio was higher for GRS_6.6M (2.22 per SD of the risk score) than for PRS₁₈₁ (1.39; Table 3 and Table S4A). Associations for women and men were comparable. Both PRS₁₈₁ and GRS_6.6M were also associated with incident CAD. The hazard ratios (HRs) for incident CAD were slightly lower than the odds ratios for prevalent CAD (HR, 1.21 for PRS₁₈₁ and 1.32 for GRS_6.6M; Table 3 and Table S4B).

Table 3.

Meta‐Analyses of Associations of PRS₁₈₁ and GRS_6.6M With CAD in Relation to Sex and Age at Onset

	PRS181			GRS6.6M
Prevalent CAD
	Odds ratio	95% CI	P value	Odds ratio	95% CI	P value
All	1.39	1.29 to 1.50	<1E‐16	2.22	2.04 to 2.40	<1E‐16
Men	1.39	1.27 to 1.51	5.0E‐14	2.19	1.99 to 2.40	<1E‐16
Women	1.41	1.23 to 1.62	7.8E‐7	2.39	2.09 to 2.74	<1E‐16

Incident CAD
	HR	95% CI	P value	HR	95% CI	P value
All	1.21	1.17 to 1.26	<1E‐16	1.32	1.28 to 1.37	<1E‐16
Men	1.19	1.13 to 1.26	1.9E‐11	1.32	1.26 to 1.39	<1E‐16
Women	1.24	1.16 to 1.32	2.3E‐10	1.32	1.25 to 1.40	<1E‐16

Incident CAD in different age groups, y
	HR	95% CI	P value	HR	95% CI	P value
<60	1.45	1.29 to 1.62	3.0E‐10	1.55	1.39 to 1.74	1.9E‐14
60–69	1.19	1.12 to 1.27	1.0E‐07	1.42	1.32 to 1.53	<1E‐16
70–79	1.13	1.05 to 1.21	6.2E‐04	1.33	1.24 to 1.42	3.2E‐15
≥80	1.14	1.05 to 1.23	1.1E‐03	1.21	1.14 to 1.30	1.1E‐08

Incident CAD before age 60 y and at age ≥60 y
	HR	95% CI	P value	HR	95% CI	P value
<60	1.45	1.29 to 1.62	3.0E‐10	1.55	1.39 to 1.74	1.9E‐14
≥60	1.18	1.13 to 1.23	6.2E‐15	1.30	1.24 to 1.35	<1E‐16

Age at CAD onset
	β value	95% CI	P value	β value	95% CI	P value
All	−0.86	−1.21 to −0.52	8.7E‐07	−0.32	−0.54 to −0.11	3.5E‐03
Men	−0.93	−1.36 to −0.50	2.5E‐05	−0.28	−0.55 to −0.002	4.9E‐02
Women	−0.76	−1.31 to −0.21	6.7E‐03	−0.51	−0.84 to −0.17	2.9E‐03

Odds ratios and HRs are calculated per 1 SD increase in the genetic risk scores. Prevalent CAD cases were defined at baseline. For incident CAD, participants were censored at first diagnosis, death, other loss to follow‐up, the end of the study period, or after 10 years of follow–up. Table S4 shows results for the individual cohorts (RS‐I, ‐II, and ‐III; MESA; and Sanford Health). CAD indicates coronary artery disease; GRS_6.6M, genome‐wide polygenic risk score that encompasses 6.6 million single‐nucleotide polymorphisms; HR, hazard ratio; MESA, Multi‐Ethnic Study of Atherosclerosis; PRS₁₈₁, restricted polygenic risk score with 181 high‐confidence genetic variants (or 177 single‐nucleotide polymorphisms for the RS); and RS, Rotterdam Study.

Genetic Risk Score Percentiles

We pooled data for women and men (for more statistical power) to assess CAD risks across PRS₁₈₁ and GRS_6.6M percentiles (ie, 50th, 60th, 70th, 80th, 90th, 95th, and 98th percentiles; Table 4 and Table S5A through S5F). The bottom 50% of the risk score distributions served as the reference populations. Genetic risk score effect sizes increased at higher risk score percentiles. The top 2% of the genetic risk score distributions had the highest HRs for both scores (2.21 for PRS₁₈₁; 2.87 for GRS_6.6M; Table 4, Table S5A and S5D).

Table 4.

Meta‐Analyses of Associations of PRS₁₈₁ and GRS_6.6M With Incident CAD in Relation to Genetic Risk Score Percentiles

	PRS181			GRS6.6M
All ages at onset
	HR	95% CI	P value	HR	95% CI	P value
Top 50%	1.30	1.20–1.41	5.7E‐10	1.55	1.44–1.68	<1E‐16
Top 40%	1.39	1.27–1.51	3.1E‐14	1.60	1.48–1.73	<1E‐16
Top 30%	1.43	1.30–1.57	2.2E‐13	1.70	1.55–1.85	<1E‐16
Top 20%	1.53	1.37–1.70	3.8E‐15	1.84	1.68–2.03	<1E‐16
Top 10%	1.69	1.49–1.91	2.2E‐16	2.11	1.87–2.38	<1E‐16
Top 5%	1.98	1.69–2.33	<1E‐16	2.47	2.12–2.88	<1E‐16
Top 2%	2.21	1.75–2.78	1.4E‐11	2.87	2.29–3.60	<1E‐16

Age at onset <60 y
	HR	95% CI	P value	HR	95% CI	P value
Top 50%	1.60	1.26–2.04	1.3E‐04	2.10	1.65–2.69	2.9E‐09
Top 40%	1.65	1.29–2.11	6.0E‐05	2.29	1.79–2.93	5.6E‐11
Top 30%	1.77	1.36–2.31	1.9E‐05	2.55	1.97–3.32	1.8E‐12
Top 20%	2.14	1.62–2.82	7.6E‐08	2.76	2.07–3.69	5.8E‐12
Top 10%	2.71	1.96–3.75	1.6E‐09	3.70	2.67–5.13	3.9E‐15
Top 5%	3.60	2.47–5.25	2.5E‐11	4.36	2.93–6.51	4.9E‐13
Top 2%	5.41	3.32–8.81	1.2E‐11	4.94	2.87–8.51	8.5E‐09

Age at onset ≥60 y
	HR	95% CI	P value	HR	95% CI	P value
Top 50%	1.28	1.17–1.40	6.1E‐08	1.51	1.39–1.64	<1E‐16
Top 40%	1.36	1.25–1.49	1.1E‐11	1.55	1.43–1.69	<1E‐16
Top 30%	1.39	1.26–1.53	1.1E‐10	1.63	1.49–1.79	<1E‐16
Top 20%	1.47	1.31–1.64	1.6E‐11	1.79	1.61–1.99	<1E‐16
Top 10%	1.59	1.39–1.82	3.2E‐11	1.98	1.74–2.26	<1E‐16
Top 5%	1.82	1.53–2.17	3.0E‐11	2.32	1.96–2.74	<1E‐16
Top 2%	1.92	1.47–2.50	1.7E‐06	2.72	2.12–3.48	2.4E‐15

HRs in this table are derived from categorical analyses comparing individuals in the top X% of the genetic risk score distribution with those in the bottom 50% (reference group). The bottom 50% of the genetic risk score distribution was the reference group. For incident CAD, participants were censored at first diagnosis, death, other loss to follow‐up, the end of the study period, or after 10 y of follow‐up. Table S5 shows results for the individual cohorts (RS‐I, ‐II, and ‐III; MESA; and Sanford Health). CAD indicates coronary artery disease; GRS_6.6M, genome‐wide polygenic risk score that encompasses 6.6 million single‐nucleotide polymorphisms; HR, hazard ratio; MESA, Multi‐Ethnic Study of Atherosclerosis; PRS₁₈₁, restricted polygenic risk score with 181 high‐confidence genetic variants (or 177 SNPs for the RS); and RS, Rotterdam Study.

Age at CAD Onset

Early‐onset cases (age<60 years) had the strongest associations across 4 age groups (Table 3 and Table S4D). Both PRS₁₈₁ and GRS_6.6M also showed stronger effects for early‐onset CAD when incident CAD was stratified by age (<60 years versus ≥60 years; P _interaction=0.00088 for PRS₁₈₁; P _interaction=0.0042 for GRS_6.6M; Table 3 and Table S4E).

Early‐onset cases with scores above the 98th percentile had the largest HRs (5.41 for PRS₁₈₁; 4.94 for GRS_6.6M; Table 4, Table S5B and S5E). The corresponding HRs for late‐onset cases were 1.92 and 2.72 (Table 4; Table S5C and S5F).

PRS₁₈₁ was associated with age at onset for both men (β=−0.93) and women (β=−0.76; Table 3 and Table S4C). Onset was almost 1 year earlier per SD of PRS₁₈₁. A GRS_6.6M effect on age at onset was smaller for women (β=−0.51; roughly 0.3 years earlier per SD of GRS_6.6M). No GRS_6.6M effect on age at onset was observed for men (β=−0.28 [95% CI, −0.55 to 0.01]; Table 3 and Table S4C). No statistically significant differences were found between men and women.

Lipid Medication Use

We explored associations between genetic risk scores and lipid medication use for incident CAD. Lipid medication use (or nonuse) may represent a clinician’s overall impression of an individual’s CAD risk. Participants were separated into 4 groups: (1) nonuser controls; (2) nonuser incident CAD cases; (3) user controls; and (4) user incident CAD cases.

Association of Risk Scores With CAD in Relation to Medication Use

Both PRS₁₈₁ and GRS_6.6M were associated with CAD risk, regardless of medication use. As expected, the strongest risk score associations with CAD were seen when cases with CAD who were lipid medication users were compared with controls who were nonusers. HRs were 1.33 for PRS₁₈₁ and 1.49 for GRS_6.6M (Table 5, Table S4F.1 and S4F.6). We also observed associations between risk scores and CAD when we compared cases with CAD who were lipid medication users with controls who were also users (HRs, 1.20 for PRS₁₈₁ and 1.32 for GRS_6.6M; Table 5, Table S4F.3 and S4F.8). Finally, increased CAD risks were seen for the comparison of cases with CAD who were nonusers with controls who were nonusers (HRs, 1.18 for PRS₁₈₁ and 1.25 for GRS_6.6M; Table 5, Table S4F.2 and S4F.7).

Table 5.

Meta‐Analyses of Associations of PRS₁₈₁ and GRS_6.6M in Relation to Incident CAD and Lipid Medication Use

	PRS181			GRS6.6M
	HR	95% CI	P value	HR	95% CI	P value
Associations of risk scores with incident CAD
User cases vs nonuser controls	1.33	1.23–1.45	1.5E‐11	1.49	1.40–1.60	<1E‐16
User cases vs user controls	1.20	1.10–1.31	2.0E‐05	1.32	1.24–1.41	3.3E‐16
Nonuser cases vs nonuser controls	1.18	1.09–1.27	1.1E‐05	1.25	1.15–1.35	1.3E‐07
Associations of risk scores with lipid medication use
User cases vs nonuser cases	1.12	1.02–1.24	2.1E‐02	1.14	1.05–1.23	9.7E‐04
User controls vs nonuser controls	1.13	1.09–1.17	9.7E‐10	1.10	1.05–1.14	3.1E‐06

HRs are based on the values below
		Incident CAD
		Control	Case
Lipid medication	Nonuser	a	b
	User	c	d

All HRs in this table are derived from continuous models, calculated per 1 SD increase in the genetic risk scores (PRS₁₈₁ and GRS_6.6M). For incident CAD, participants were censored at first diagnosis, death, other loss to follow‐up, the end of the study period, or after 10 y of follow‐up. Table S4 shows results for the individual cohorts (RS‐I, ‐II, and ‐III; MESA; and Sanford Health). CAD indicates coronary artery disease; GRS_6.6M, genome‐wide polygenic risk score that encompasses 6.6 million single‐nucleotide polymorphisms; HR, hazard ratio; MESA, Multi‐Ethnic Study of Atherosclerosis; PRS₁₈₁, restricted polygenic risk score with 181 high‐confidence genetic variants (or 177 single‐nucleotide polymorphisms for the RS); and RS, Rotterdam Study.

Association of Risk Scores With Lipid Medication Use in Relation to CAD

We examined associations of genetic risk scores with lipid medication use among cases and controls with incident CAD. The HRs for lipid medication use for the comparison of users who were cases with CAD versus nonuser cases with CAD were 1.12 for PRS₁₈₁ and 1.14 for GRS_6.6M (Table 5, Table S4F.4 and S4F.9). HRs were similar for the comparison of user controls versus nonuser controls (1.13 and 1.10, for PRS₁₈₁ and GRS_6.6M, respectively; Table 5, Table S4F.5 and S4F.10).

Combined Use of PRS₁₈₁ and GRS_{6. 6M}

Direct comparison of PRS₁₈₁ versus GRS_6.6M showed correlations of 53% in Sanford Health, 51% in the RS, and 47% in MESA (Figure 1[A]). We stratified participants according to their PRS₁₈₁ and GRS_6.6M values (in SD units). Each participant was classified as low risk (<−1 SD from the mean), intermediate risk (within ±1 SD from the mean), or high risk (>1 SD from the mean) for each risk score (PRS₁₈₁ and GRS_6.6M). In this way, there were 9 groups representing PRS₁₈₁ and GRS_6.6M combined (see Methods). Individuals with intermediate risk for both scores served as the reference group, with which other groups were compared (G1 through G8; Figures 1[A] and 1[B]).

Individuals classified as G1 (low risk by both scores) had the lowest CAD risk (HR, 0.69) and a mean age of onset of 73.6±8.9 years. Conversely, G8 individuals (high risk by both scores) had the highest HR for CAD (1.88) and a mean age of onset of 69.3±9.6 years (Figure 1[B]; Tables S3 and S4G).

Of the 2 groups with low‐to‐intermediate risk (G2 and G4), group G4 (low risk by GRS_6.6M and intermediate risk by PRS₁₈₁) was associated with low CAD risk (HR, 0.69) and had a mean age of onset of 73.6±8.9 years. Group G2 (low risk by PRS₁₈₁ and intermediate risk by GRS_6.6M) did not have a CAD risk association (HR, 0.96) but had a higher age at onset (73.3±9.3 years) compared with the reference group.

The 2 groups with intermediate‐to‐high risk (G5 and G7) both demonstrated significantly increased CAD risks (HRs, 1.42 and 1.20, respectively) and ages at onset around 70 years.

The 2 discordant risk groups (high risk in one score and low risk in the other; G3 and G6) did not show associations with CAD risk but had the highest ages at onset among all groups (76.1±2.1 years for G3, with high GRS_6.6M; 75.9±10.2 years for G6, with high PRS₁₈₁).

Discussion

We evaluated strengths and limitations of a PRS₁₈₁ and a GRS_6.6M genetic risk score in relation to associations with CAD. Our study had 38 852 individuals from 3 cohorts (1 from Europe and 2 from the United States). We assessed the scores in relation to CAD risk, age at onset, and lipid medication use. Both PRS₁₈₁ and GRS_6.6M were associated with CAD risk in men and women. However, PRS₁₈₁ showed a stronger and statistically significant association with earlier CAD onset in both sexes, whereas GRS_6.6M showed a modest effect in women and no significant effect in men. In addition, PRS₁₈₁ and GRS_6.6M were only moderately correlated across our 3 cohorts. Use of PRS₁₈₁ and GRS_6.6M together provided better association information on CAD risk and age at CAD onset than use of each score alone.

CAD Risk Prediction by PRS₁₈₁ and GRS_{6. 6M}

Both scores were associated with prevalent and incident CAD in men and women (Table 3). CAD risks increased with higher genetic scores, particularly among early‐onset cases. For both scores, individuals in the top 2% of the risk score distribution had the highest HRs (Table 4).

Age at CAD Onset

PRS₁₈₁ showed a linear relation with age at CAD onset. One SD of the score was associated with CAD onset nearly 1 year earlier. The trend was not as strong for GRS_6.6M, which showed no age association for men and a modest effect for women (≈0.3 years earlier per SD; Table 3). On the other hand, we observed fairly similar risk score effects for both PRS₁₈₁ and GRS_6.6M among early‐onset cases ³ , ¹⁶ (Tables 3 and 4). CIs were wide, however. Further work is needed to interpret the age‐at‐onset findings.

CAD Genetic Risk Scores and Lipid Medication Use

As mentioned above, lipid medication use (or nonuse) may represent a clinician’s overall impression of an individual’s risk for CAD. However, both PRS₁₈₁ and GRS_6.6M were associated with CAD, independent of lipid medication use. And both scores were associated with lipid medication use regardless of whether an individual had CAD (Table 5). Larger studies may be needed to draw definite conclusions from such results. Data on lipid medication use were not available for the Sanford Health cohort.

Combined Use of PRS₁₈₁ and GRS_{6. 6M}

PRS₁₈₁ and GRS_6.6M showed only moderate correlation across cohorts (53% in Sanford, 51% in the RS, 47% in MESA). Thus, the 2 scores capture different aspects of CAD genetic architecture, even though risk information in PRS₁₈₁ and in GRS_6.6M overlap. The differences in risk score values likely reflect differences among the genome‐wide association study data sets used to identify variants, as well as differences in the variants that make up the scores.

When CAD risks were analyzed by combined use of PRS₁₈₁ and GRS_6.6M, individuals at high CAD susceptibility by both scores had the highest risk estimate and the youngest age at onset (HR, 1.88; age at onset, 69.3; group G8; Figure 1[B]). Results support an interpretation that the scores identify distinct individuals at increased CAD risk. Therefore, combining PRS₁₈₁ and GRS_6.6M approaches appears to improve association information on CAD, particularly for individuals at the highest risk.

In contrast, individuals with high risk by one score but low risk by the other did not show significantly increased CAD risks and had the highest mean ages of onset (HR, 0.92–0.98; age at onset, 75.9–76.1). Thus, risk interpretation may be unclear for individuals with discordant genetic risk scores (Figure 1[C]). The findings support a conclusion that relying on a single genetic risk score may overlook important information when assessing risks for CAD.

Large analyses of the extent to which combined scores may improve risk assessment, beyond predictions made through traditional risk factors, are now needed. For example, use of combined risk scores should be analyzed in relation to the pooled cohort equations, Framingham risk scores, or Systematic Coronary Risk Evaluation 2 assessments.

Strengths and Limitations

Strengths of our study are construction of PRS₁₈₁ from validated variants across 12 independent studies, which enhances generalizability, and comparison of PRS₁₈₁ to GRS_6.6M, a well‐established genome‐wide score. We assessed both scores in relation to prevalent and incident CAD, with stratification by sex, age, and lipid medication use. Our findings are also strengthened by use of both community‐based and health care system cohorts (the RS, MESA, and Sanford Health). However, not all analyses applied to MESA and Sanford Health due to differences in study designs. Another limitation is the predominantly European ancestry of participants. Further studies should evaluate score performance in more diverse populations. Our analysis focused on 1 genome‐wide score (GRS_6.6M, constructed using LDpred). Future work should assess genome‐wide scores constructed by other methods, such as thresholding plus principal components or least absolute shrinkage and selection operator regression. Finally, we focused on only genetic contributions to CAD risk. Factors that could be analyzed in future models include triglycerides, low‐density lipoprotein cholesterol, blood pressure, and subclinical atherosclerosis.

Conclusions

We compared 2 genetic risk scores for CAD: the restricted PRS₁₈₁ score and the GRS_6.6M score. Both were associated with CAD risk, age at onset, and lipid medication use. PRS₁₈₁ consistently produced strong effect sizes. Because there was only moderate correlation between the 2 risk scores, the scores appear to represent different components of the genetic risk for CAD. Combining both scores yielded better evaluations of CAD risk and age at onset. If confirmed, the results suggest that use of multiple genetic risk scores for CAD, rather than a single score, may be advantageous in clinical practice.

Sources of Funding

The current project was instigated as an extension of the Genotyping on ALL Patients project, a multidisciplinary collaborative effort involving various departments at Erasmus Medical Center. The Genotyping on ALL Patients project was funded by the internal Koers23 program from the Erasmus Medical Center (Project No. 109433). The authors thank the other investigators, staff members, and participants who contributed substantially to the Genotyping on ALL Patients project. The RS is funded by Erasmus Medical Center and Erasmus University, Rotterdam; Netherlands Organization for the Health Research and Development (ZonMw); the Research Institute for Diseases in the Elderly; the Ministry of Education, Culture and Science; the Ministry for Health, Welfare and Sports; the European Commission (DG XII); and the Municipality of Rotterdam.

MESA and the MESA SHARe projects are conducted and supported by the National Heart, Lung, and Blood Institute in collaboration with MESA investigators. Support for MESA is provided by contracts 75N92020D00001, HHSN268201500003I, N01‐HC‐95159, 75N92020D00005, N01‐HC‐95160, 75N92020D00002, N01‐HC‐95161, 75N92020D00003, N01‐HC‐95162, 75N92020D00006, N01‐HC‐95163, 75N92020D00004, N01‐HC‐95164, 75N92020D00007, N01‐HC‐95165, N01‐HC‐95166, N01‐HC‐95167, N01‐HC‐95168, N01‐HC‐95169, UL1‐TR‐000040, UL1‐TR‐001079, UL1‐TR‐001420, UL1TR001881, DK063491, and R01HL105756. Funding for SHARe genotyping was provided by National Heart, Lung, and Blood Institute Contract N02‐HL‐64278. This study was also supported in part by National Heart, Lung, and Blood Institute contracts R01HL151855 and R01HL146860.

The Sanford Health project is funded by the Sanford Health System. Supported in part by Sanford Health Care contract number 2018–1690, by the National Center for Advancing Translational Sciences, CTSI grant UL1TR001881, and the National Institute of Diabetes and Digestive and Kidney Disease Diabetes Research Center grant DK063491 to the Southern California Diabetes Endocrinology Research Center. Infrastructure for the CHARGE Consortium is supported in part by the National Heart, Lung, and Blood Institute grant R01HL105756.

Disclosures

None.

Supporting information

Tables S1–S5