Coronary Artery Disease–Associated and Longevity-Associated Polygenic Risk Scores for Prediction of Coronary Artery Disease Events in Persons Living With Human Immunodeficiency Virus: The Swiss HIV Cohort Study

Abstract

Background

Coronary artery disease (CAD) is in part genetically determined. Aging is accentuated in people with human immunodeficiency virus (HIV) (PLWH). It is unknown whether genetic CAD event prediction in PLWH is improved by applying individual polygenic risk scores (PRSs) and by considering genetic variants associated with successful aging and longevity.

Methods

In the Swiss HIV Cohort Study participants of self-reported European descent, we determined univariable and multivariable odds ratios (ORs) for CAD events, based on traditional CAD risk factors, adverse antiretroviral exposures, and different validated genome-wide PRSs. PRSs were built from CAD-associated single-nucleotide polymorphisms (SNPs), longevity-associated SNPs, or both.

Results

We included 269 patients with CAD events between 2000 and 2017 (median age, 54 years; 87% male; 82% with suppressed HIV RNA) and 567 event-free controls. Clinical (ie, traditional and HIV-related) risk factors and PRSs, built from CAD-associated SNPs, longevity-associated SNPs, or both, each contributed independently to CAD events (P < .001). Participants with the most unfavorable clinical risk factor profile (top quintile) had an adjusted CAD-OR of 17.82 (95% confidence interval [CI], 8.19–38.76), compared with participants in the bottom quintile. Participants with the most unfavorable CAD-PRSs (top quintile) had an adjusted CAD-OR of 3.17 (95% CI, 1.74–5.79), compared with the bottom quintile. After adding longevity-associated SNPs to the CAD-PRS, participants with the most unfavorable genetic background (top quintile) had an adjusted CAD-OR of 3.67 (95% CI, 2.00–6.73), compared with the bottom quintile.

Conclusions

In Swiss PLWH, CAD prediction based on traditional and HIV-related risk factors was superior to genetic CAD prediction based on longevity- and CAD-associated PRS. Combining traditional, HIV-related, and genetic risk factors provided the most powerful CAD prediction.

Aging, coronary artery disease, HIV infection, multivariable analysis, polygenic risk score

Current meta-analyses and guidelines suggest an approximately 2-fold elevated rate of coronary artery disease (CAD) events in people living with human immunodeficiency virus (HIV) (PLWH), including in those with suppressed viremia on antiretroviral therapy (ART), compared with the general population [1, 2]. Individual susceptibility to CAD in PLWH is influenced by traditional CAD risk factors as well as HIV-associated factors, including adverse antiretroviral exposures [3, 4]. CAD also has a strong hereditary component [5–8]. Genome-wide association studies (GWASs) [9–13] have identified common genetic variants that contribute to CAD risk in the general population, with CAD lifetime risk trajectories robustly established based on a polygenic risk score (PRS) consisting of 1.7 million single nucleotide polymorphisms (SNPs) in >480 000 individuals [13].

Our group has previously reported a 1.47-fold increased CAD risk in PLWH with unfavorable genetic background, based on 23 common SNPs [14]. This genetic CAD risk increase was similar to the risk increase attributable to traditional risk factors (eg, dyslipidemia) or adverse antiretroviral exposures (eg, to abacavir or lopinavir) [14]. Because there is concern that aging in PLWH may be accelerated and/or accentuated, there is interest in the potential for improved genetic CAD risk prediction by including SNPs that have been reliably associated with successful aging and longevity in the general population [15, 16]. In the current study, we evaluated CAD event prediction in Swiss PLWH based on traditional, HIV-related and genetic risk factors, including different PRSs built from validated CAD-associated and longevity-associated SNPs.

METHODS

Study Population

We included PLWH enrolled in the Swiss HIV Cohort Study (SHCS; www.shcs.ch) [17] who were participants in our groups’ previous CAD event prediction study [18]. The study was approved by the respective local ethics committees. Participants provided written informed consent for genetic testing. Case patients had a first CAD event and controls were CAD event-free during the study period (1 January 2000 to 31 December 2017). Because previous CAD-GWASs in the general population were conducted in populations of mostly European descent [19], the study was restricted to participants of self-reported European descent. SHCS data are gathered by the 5 Swiss university hospitals, 2 cantonal hospitals, 15 affiliated hospitals, and 36 private physicians (listed in http://www.shcs.ch/180-health-care-providers)

CAD Events

CAD events were defined according to the Data Collection on Adverse events of Anti-HIV Drugs (D:A:D) study and the MONICA Project of the World Health Organization [20], as reported elsewhere [18].

Case-Control Matching

As reported elsewhere [18], we aimed to select 3 controls who were CAD event free at the CAD event date for the corresponding case patient (matching date) using risk-set sampling [21]. We used incidence density sampling for matching [22]; that is, we matched controls for similar observation durations, with observation during similar calendar periods, to account for differences in ART (with different potential CAD risk associations [4, 23]) in use at different times and other differences during the observation period. As reported elsewhere [18], we used as matching criteria sex, age (within ±4 years), and date of SHCS enrollment (within ±4 years). Case patients were observed until the matching date, and controls until the first regular SHCS follow-up examination after the CAD event date for the corresponding case patient, respectively. We allowed reuse of controls for up to 3 case patients [21].

Nongenetic Clinical CAD Risk Factors

As reported elsewhere [18], covariables included smoking status (current, past, or never), age (per 1 year older), and family history of CAD, diabetes mellitus, hypertension, or dyslipidemia (defined as published elsewhere [24]). HIV-related covariables included HIV viremia at the matching date (HIV RNA <50 or >50 copies/mL), CD4 cell count nadir, and ART exposures, defined a priori, based on their CAD association in the D:A:D study, that is, current (last 6 months) exposure to abacavir [4] and cumulative exposure (>1 year) to lopinavir, indinavir, darunavir [23], and stavudine [25, 26] until the matching date, cytomegalovirus (CMV) seropositivity [27], hepatitis C virus seropositivity [28, 29], and intravenous drug use.

Genotyping

DNA samples were obtained from peripheral blood mononuclear cells and genotyped with the Global Screening Array v2.0 + MD (Illumina), or in the setting of previous SHCS genetic studies. All quality control, filtering, and imputation steps before the merging of batches were performed separately for each batch of samples as described in the Supplementary Methods. For the final merged data set used to calculate the PRSs, only variants with a minor allele frequency >5% and missingness <10% were kept.

Calculating Genome-Wide PRSs

PRSs were calculated using PRSice software (version 2.3.3). The CAD-PRS was calculated by directly applying the variant information from the CAD-PRS previously validated by Inouye et al [13]. Information on included variants in this score and their weights were downloaded from the PGS Catalog [30]. In total, 607 895 PRS variants from Inouye et al [13] were successfully matched and included in the CAD-PRS. The longevity-PRS was calculated using the P values and effect sizes from a large longevity GWAS reported by Deelen et al [16], using the 90th survival percentile as phenotype. After matching between the genotype data and summary statistics, the variants were clumped using windows of 250 kilobases and an r² value of 0.1. The best-fit model with 4 independent genome-wide significant SNPs (P < 5 × 10⁻⁸) from Deelen et al [16] was found by P value thresholding with PRSice software. Of note, a favorable longevity-PRS is associated with longevity in the reference study, and an unfavorable longevity-PRS with CAD events in the present study. We also applied a combined “meta-PRS” (ie, CAD-PRS plus longevity-PRS), which we calculated following the same principles described by Inouye et al [13] (Supplementary Methods).

Power Calculation

To detect CAD event odds ratios (ORs) of >1.6, 255 case patients and 2 controls per case patient would be required [31]. As recommended, the calculations assume a correlation of exposure between pairs in the case-control set of 0.2 [31].

Statistical Analyses

Univariable and multivariable conditional logistic regression analyses were used to estimate associations of the different clinical and genetic risk factors. We decided a priori to stratify the genetic risk factors into quintiles for better visualization of their potentially nonlinear associations with CAD events. Clinical variables were entered into the multivariable model if their association in the univariable model had P < .2. We combined all traditional and HIV-related risk factors into a single measure of “clinical” CAD event risk by creating quintiles of the individually predicted CAD event probabilities from the multivariable model with the clinical risk factors as described above. These clinical risk quintiles were then used to check for and visualize interactions with genetic risk factors. Model fit and interactions were analyzed using Akaike and bayesian information criteria and likelihood ratio tests. CAD event variation explained by the different models with combinations of clinical and genetic risk factors were documented with pseudo-R² values (as in our group’s 2013 CAD genetic study [14]) and receiver operating characteristic (ROC) values. To assess for an association between telomere length (TL), longevity-PRS, and CAD prediction, we added TL to the multivariable models. We used Stata/SE 16.1 software (StataCorp).

RESULTS

After exclusion of 46 case patients and 31 controls because of excessive missingness in the genotyping data, and 98 participants because of incomplete case-control pairs, the study population consisted of 269 case patients and 567 controls, based on 357 individual control participants who were matched with 1 (n = 212), 2 (n = 80), or 3 case patients (n = 65). CAD events included myocardial infarction (n = 143), coronary angioplasty/stenting (n = 102), coronary artery bypass grafting (n = 17), and fatal CAD with evidence of CAD before death (n = 7) [18]. The participants’ baseline characteristics are shown in Table 1. Case patients were older, more likely to be intravenous drug users, current smokers, diabetic, or dyslipidemic or to have a family history of CAD, and their ART exposure was longer.

Table 1.

Characteristics of Case Patients and Controls at Matching Dates

	Case Patients or Controls, No. (%)a
Characteristic	Case Patients (n = 269)	Controls (n = 567)
Male sex	235 (87.4)	500 (88.2)
Age, median (IQR), y	54 (48–62)	53 (47–62)
HIV acquisition mode
Heterosexual	70 (26.0)	162 (28.6)
MSM	132 (49.1)	295 (52.0)
IDU	57 (21.2)	96 (16.9)
Other	10 (3.7)	14 (2.5)
Smoking status
Current	135 (50.2)	246 (43.4)
Past	81 (30.1)	173 (30.5)
Never	53 (19.7)	148 (26.1)
Cocaine use
Recentb	10 (3.7)	22 (3.9)
Ever	22 (8.2)	50 (8.8)
Family history of CAD	44 (16.4)	61 (10.8)
Diabetes mellitus	47 (17.5)	37 (6.5)
Hypertension	83 (30.9)	163 (28.8)
Dyslipidemia	177 (65.8)	264 (46.6)
Receiving ART	247 (91.8)	481 (84.8)
HIV RNA <50 copies/mL (undetectable) during ART	221 (82.2)	453 (79.9)
Total years on ART, median (IQR)	10.9 (6.6–15.8)	6.0 (2.5–10.9)
Duration of observation,c median (IQR), y	11.8 (8.1–17.4)	11.2 (7.6–17.2)
Currently taking abacavir, n (%)	84 (31.2)	113 (20.0)
ART exposure >1 y
Lopinavir	79 (29.4)	100 (17.6)
Indinavir	65 (24.2)	53 (9.4)
Darunavir	45 (16.7)	60 (10.6)
Stavudine	117 (43.5)	87 (15.3)
CD4 cell count, median (IQR), cells/μL
At matching date	490 (353–722)	526 (376–688)
During observation time	459 (323–618)	470 (356–585)
Nadir	169 (71–257)	205 (126–318)
CD4 cell count nadir <50 cells/μL	50 (18.6)	58 (10.2)
Previous AIDS	69 (25.6)	121 (21.3)
HCV seropositivity	121 (21.3)	75 (27.9)
CMV seropositivity	234 (87.0)	458 (80.8)

Abbreviations: ART, antiretroviral therapy; CAD, coronary artery disease; CMV, cytomegalovirus; HCV, hepatitis C virus; HIV, human immunodeficiency virus; IDU, intravenous drug use; IQR, interquartile range; MSM, men who have sex with men.

^aAll data shown apply to the matching date and represent no. (%) of participants, unless otherwise indicated.

^bRecent defined as within 6 months before matching date.

^cDuration from registration in the Swiss HIV Cohort Study until the matching date, and, for controls, until the first regular, twice-yearly follow-up visit after the matching date.

PRS Results

After P value thresholding, the CAD-PRS included 607 895 SNPs, and the longevity-PRS included 4 independent SNPs after clumping (Supplementary Table 1). There was no evidence of correlation between CAD-PRS and longevity-PRS (Pearson correlation r = 0.06).

Univariable Analyses

Probability of CAD Events

Case patients had higher clinical and higher genetic CAD risk than controls, as indicated by the asymmetric distribution of case patient among the quintiles; that is, their distribution was skewed toward the fifth (most unfavorable) quintile compared with the other quintiles of clinical risk (Figure 1A), CAD-PRS (Figure 1B), longevity-PRS (Figure 1C), and meta-PRS (Figure 1D). CAD event probability was significantly associated with clinical risk (test for trend, P < .001), CAD-PRS (P < .001), longevity-PRS (P = .001), and meta-PRS (P < .001) (Figure 2).

Figure 1.

A–D, Distribution of clinical risk factors and polygenic risk scores (PRSs) in 567 controls without coronary artery disease (CAD) events (white bars) and in 269 case patients with CAD events (gray bars). We divided study participants into 5 quintiles according to their individual clinical and PRSs and show here the number, percentage, and 95% confidence intervals (CIs) of participants in each quintile. A, Distribution of case patients and controls according to quintiles of clinical risk; there were 7 case patients (4.2%) versus 160 controls (95.8%) controls in the first (most favorable) quintile, 15 (9%) versus 152 (91%) in the second, 37 (22.2%) versus 130 (77.8%) in the third, 82 (49.1%) versus 85 (50.9%) in the fourth, and 128 (76.2%) versus 40 (23.8%) in the fifth (most unfavorable) quintile. B, Distribution of case patients and controls according to quintiles of CAD PRS; there were 42 (25.9%) case patients versus 120 (74.1%) controls in the first quintile, 46 (26.7%) versus 126 (73.3%) in the second, 42 (23.3%) versus 138 (76.7%) in the third, 66 (38.6%) versus 105 (61.4%) in the fourth, and 73 (48.3%) versus 78 (51.7%) in the fifth quintile. C, Distribution of case patients and controls according to quintiles of longevity-PRS; there were 36 (24.2%) case patients versus 113 (75.8%) controls in the first quintile, 35 (31%) versus 78 (69%) in the second, 40 (29.4%) versus 96 (70.6%) in the third, 92 (32.1%) versus 195 (67.9%) in the fourth, and 66 (43.7%) versus 85 (56.3%) in the fifth quintile. D, Distribution of case patients and controls according to quintiles of combined CAD-PRS and longevity-PRS (meta-PRS); there were 35 (20.1%) case patients versus 139 (79.9%) controls in the first quintile, 49 (28.7%) versus 122 (71.4%) in the second , 51 (29%) versus 125 (71%) in the third, and 58 (33.9%) versus 113 (66.1%) in the fourth quintile, and 76 (52.8%) versus 68 (47.2%) in the fifth quintile.

Figure 2.

Odds ratios (ORs) for coronary artery disease (CAD) events (with 95% confidence intervals [CIs]), according to clinical risk factors and different polygenic risk scores (PRSs). Results show univariable and multivariable conditional logistic regression of associations with CAD events, for 269 case patients and 567 controls. Compared with the first (most favorable) quintile of clinical risk, participants in the second, third, fourth, and fifth (most unfavorable) quintiles had multivariable ORs that remained similar, irrespective of which PRS we adjusted for (left column). After adjustment for clinical risk factors, the CAD-PRS and meta-PRS remained significantly associated with CAD events, but not the longevity-PRS (right column).

Probability of CAD Events According to Clinical Risk Factors

Regarding traditional risk factors, CAD was associated with age, family history of CAD, current smoking, diabetes mellitus, dyslipidemia, CMV seropositivity, and hepatitis C virus seropositivity. Compared with participants in the first (most favorable) quintile of traditional risk, those in the second, third, fourth, and fifth (most unfavorable) quintiles had univariable CAD-ORs of 2.63 (95% confidence interval [CI], 1.36–5.09), 3.96 (2.09–7.52), 3.68 (1.95–6.95) and 8.70 (4.82–15.68), respectively.

Regarding HIV-associated risk factors, CAD was associated with current use of abacavir, cumulative exposure to lopinavir, indinavir, darunavir, stavudine, and CD4 cell count nadir, but not with HIV viral load or CD4 cell count at the matching date, or with cocaine use (Supplementary Table 2). Compared with participants in the first quintile for HIV-associated risk, those in the second, third, fourth, and fifth quintiles had univariable CAD-ORs of 1.63 (95% CI, .80–3.31), 3.65 (1.80–7.42), 4.04 (2.09–7.82), and 8.71 (4.93–15.39), respectively. Compared with participants in the first quintile for clinical risk (ie, traditional and HIV-related risk factors combined), those in the second, third, fourth, and fifth quintiles had univariable CAD-ORs of 2.31 (95% CI, .91–5.89), 6.50 (2.73–15.49), 17.18 (6.94–42.56), and 17.82 (8.19–38.76), respectively.

Probability of CAD Events According to PRSs

Compared with the first (most favorable) CAD-PRS quintile, participants in the second, third, fourth, and fifth (most unfavorable) quintiles had CAD-ORs of 0.99 (95% CI, .59–1.63), 0.79 (.48–1.31), 1.79 (1.11–2.89), and 2.93 (1.78–4.82), respectively. Compared with the first longevity-PRS quintile, participants in the second, third, fourth, and fifth quintiles had CAD-ORs of 1.31 (95% CI, .76–2.27), 1.17 (.69–1.99), 1.43 (.91–2.25), and 2.28 (1.39–3.76), respectively. Compared with the first meta-PRS quintile, participants in the second, third, fourth, and fifth quintiles had CAD-ORs of 1.49 (95% CI, .88–2.51), 1.63 (1.00–2.69), 1.91 (1.16–3.15), and 4.02 (2.43–6.66), respectively.

Multivariable Analyses

CAD Probability According to Clinical Risk Factors

CAD events remained associated with age, dyslipidemia, diabetes, CMV seropositivity, current use of abacavir, cumulative exposure to indinavir, darunavir, and stavudine (Supplementary Table 2). The effect size of clinical risk factors was similar when they were unadjusted for genetic background and when we adjusted them for CAD-PRS, longevity-PRS, or meta-PRS (Figure 2, left column). For example, participants in the fifth versus the first clinical risk quintile had adjusted CAD-ORs of 19.31 (95% CI, 8.74–42.66), 17.57 (8.04–38.41), and 17.41 (7.91–38.31) with adjustment for CAD-PRS, longevity-PRS, or meta-PRS, respectively. There was no evidence for interactions with formal testing using likelihood ratio tests (all P > .4).

CAD Probability According to PRS

With adjustment for clinical risk, and compared with the first (most favorable) CAD-PRS quintile, participants in the second, third, fourth, and fifth (most unfavorable) quintiles of CAD-PRS had multivariable CAD-ORs of 1.24 (95% CI, .67–2.30), 0.76 (.40–1.42), 1.82 (1.00–3.00), and 3.17 (1.74–5.79), respectively (Figure 2, right column). With adjustment for clinical risk, and compared with the first longevity-PRS quintile, participants in the second, third, fourth, and fifth longevity-PRS quintiles had multivariable ORs of 1.41 (95% CI, .74–2.71), 0.97 (.51–1.85), 1.13 (.66–1.93), and 1.61 (.89–2.91), respectively. With adjustment for clinical risk, and compared with the first meta-PRS quintile, participants in the second, third, fourth, and fifth meta-PRS quintiles had multivariable ORs of 1.31 (95% CI, .71–2.42), 1.40 (.77–2.56), 1.62 (.89–2.96), and 3.67 (2.00–6.73), respectively.

CAD Variability Explained by Clinical Risk Factors and PRSs: Final Multivariable Model

CAD variability explained by traditional and HIV-related risk factors is shown in the Supplementary Results. The area under the ROC curve (AUC) for clinical risk factors was 0.851. The ROC AUC for CAD-PRS, longevity-PRS, and meta-PRS was 0.688, 0.645, 0.688, respectively. The ROC AUC was improved when clinical and genetic risk factors were combined; that is, the ROC AUC for the full clinical model plus CAD-PRS, longevity-PRS, or meta-PRS was 0.870, 0.855, or 0.868, respectively (Figure 3). Results were similar when we applied pseudo-R² values rather than ROC AUC: 0.309 for the full clinical model, 0.060 for CAD-PRS, and 0.353 for the full model plus CAD-PRS, respectively (Figure 3).

Figure 3.

Coronary artery disease (CAD) event variability explained by clinical risk factors and different polygenic risk scores (PRSs), including CAD-PRS, longevity-PRS, meta-PRS, full model of clinical risk factors without considering any PRS (continuous model and model including quintiles of clinical risk), and full model plus CAD-PRS, longevity-PRS, and meta-PRS. White bars represent variability based on pseudo-R² test; gray bars, variability based on area under the receiver operating characteristic curve (ROC AUC) values.

Addition of TL to the Multivariable Models

ROC AUC and pseudo-R² values were further improved when we added TL; that is, for the full clinical model plus CAD-PRS plus TL, the full model plus longevity-PRS plus TL, and the full model plus meta-PRS plus TL, the ROC AUC was 0.876, 0.864, and 0.876, respectively, and the pseudo-R² was 0.371, 0.338, 0.366, respectively. We found no evidence of a correlation between longevity-PRS and TL (Spearman rank correlation P = .7).

DISCUSSION

Our study investigating clinical and genetic CAD prediction in Swiss PLWH of European descent has 2 main findings. First, an unfavorable genetic background independently increases CAD event risk 3.17-fold, with application of an individual PRS based on CAD-associated SNPs, and this was increased 3.67-fold with the addition of longevity-associated SNPs to the CAD-PRS in a combined meta-PRS. Second, we provide a combined estimate of the impact of traditional and HIV-related risk factors (clinical risk) and show that the highest clinical risk category was associated with a 17.4–19.3-fold increased CAD risk. Thus, while clinical risk factors clearly explained a larger proportion of CAD variability than genetic background, clinical and genetic models independently predicted CAD events, and a combined clinical plus genetic model afforded the best CAD prediction.

Our results confirm a previous CAD genetic report from 2013 that was based on a multinational (MAGNIFICENT) consortium of PLWH cohorts [14]. Importantly, we extend those results by showing improved CAD prediction when applying an individual PRS in this study, compared with a validated panel of 23 common SNPs (associated with CAD in the general population), as applied in the MAGNIFICENT study [14]. The inclusion of common variants of smaller effect sizes in PRSs, in addition to only the genome-wide significant variants, is now a well-established method to improve the predictive power of genetic risk scores [13, 32, 33]. The CAD effect of genetic background was not subtle; that is, an unfavorable PRS increased CAD-OR 3.17–3.67-fold, depending on which PRS we applied. In contrast, an unfavorable genetic background based on 23 SNPs in the MAGNIFICENT study [14] was associated with a 1.47-fold increased CAD-OR. Similarly, the CAD event variability explained was 6% (pseudo-R² test) for CAD-PRS and meta-PRS, compared with 2% for longevity PRS, and only 0.9% in the MAGNIFICENT study [14]. The CAD variability explained by traditional and HIV-related risk factors was higher (14% and 18%, respectively, and 31% in combination), emphasizing the importance of clinical risk factors for CAD event prediction. However, the best CAD prediction model overall was the combination of clinical risk factors plus meta-PRS (35% CAD event variability explained). Adding TL to the models increased CAD prediction further (37% CAD event variability explained).

We investigated the hypothesis of improved CAD prediction by applying a PRS based not only on SNPs associated with CAD but also on SNPs linked to longevity in large meta-analyses in the general population [15, 16]. There is some overlap between SNPs associated with longevity and SNPs associated with CAD and other aging traits, such as Alzheimer disease, diabetes, and cancer. This is supported by the fact that the longevity-PRS is dominated by the effect of rs429358, located within the APOE gene, a SNP that is also part of the CAD-PRS. The T allele of rs429358 has previously been associated with decreased triglyceride, decreased low-density lipoprotein, and increased high-density lipoprotein levels [34], while the C allele has been associated with an increased risk of Alzheimer disease [35]. However, we found no evidence of a correlation between CAD-PRSs and longevity-PRSs, providing the rationale for combining these scores into the meta-PRS, which modestly improved CAD prediction when compared with the CAD-PRS.

Our genetic results appear robust because we considered only SNPs that have been validated in large reference GWAS in the general population [13, 16]. We applied rigorous quality control to the genetic data, corrected for residual population stratification, and excluded population outliers. Additional strengths included our exploitation of prospectively recorded information in participants of the well-established SHCS, which allowed us to quantify and compare the CAD effects of all relevant (ie, clinical, HIV-related, and genetic) risk factors. Of note, genetic background predicted individual CAD risk independently of family history of CAD, consistent with findings of the MAGNIFICENT study [14] and findings in the general population [36, 37].

The contribution of genetic variation to common diseases such as CAD has been well studied in the general population, demonstrating a clinical value of genetic testing. Knowledge on how genetic risk factors contribute to HIV-related comorbid conditions remains limited. It was beyond the scope of our study to assess the clinical value of genetic testing (this will require prospective trials). Nonetheless, our findings suggest how an individual PRS might be applied in clinical HIV practice. The knowledge that an unfavorable genetic background independently increases the CAD event risk 3.67-fold in the 20% of PLWH in the fifth meta-PRS quintile may suggest paying even greater attention to the optimization of clinical risk factors, and, perhaps, instituting primary CAD prevention with statins in such individuals. In addition, applying different PRSs can inform the selection of PLWH at increased risk of attaining relevant end points in clinical trials.

Addition of TL to the model further improved CAD prediction, consistent with our group’s previous report [18]. Although aging is correlated with shorter TL, we found no evidence of a link between longevity-PRS and TL in our data set. Detailed pathway analyses based on genetic information, using the principle of mendelian randomization, can reveal causal relationships, provide pathogenic insights into CAD, and help avoid the risk of unknown confounding factors and reverse causation [38, 39]. Because of the limited study population size, our study was not powered for this type of genetic analysis.

Our study has additional limitations. We included only participants of European descent, because most GWASs of CAD have been conducted in populations of European descent. Our population was 87% male and relatively young; thus, results should only cautiously be extrapolated to female and elderly PLWH.

In conclusion, PLWH may have a significantly increased CAD risk because of clinical risk factors, an unfavorable genetic background, or a combination of both. Our results suggest that an unfavorable genetic background may explain why certain PLWH with low clinical CAD risk have coronary events, even in the absence of established traditional or HIV-related CAD risk factors, and vice versa. Our analyses demonstrate an independent contribution of individual PRSs to explaining interindividual variation in CAD risk, when analyzed in the context of multiple traditional, HIV-related, and antiretroviral CAD risk factors. A combination of CAD-PRSs and longevity-PRSs modestly improved CAD prediction.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Author Contributions. Study design: I. C. S., C. W. T., B. L., N. A. K., P. R., R. K., H. F. G., J. F., and P. E. T. Data management, participant selection, case-control matching: B. L. Data acquisition: C. W. T., B. L., B. H., C. M., C. T., M. S., E. B., M. C., H. B., J. F., and P. E. T. Data analysis: I. C. S., C. W. T., B. L., J. F., and P. E. T. Drafting of the manuscript: I. C. S., C. W. T., B. L., and P. E. T. Critical review and revision of the manuscript: All authors.

Acknowledgments. The authors acknowledge the effort and commitment of investigators, study nurses, laboratory personnel, and participants.

Study group. The Swiss HIV Cohort Study (SHCS) includes the following members: A. Anagnostopoulos, M. Battegay, E. B., J. Boni, D. L. Braun, H. C. Bucher, A. Calmy, M. C., A. Ciuffi, G. Dollenmaier, M. Egger, L. Elzi, J. Fehr, J. F., H. Furrer (chairman of the Clinical and Laboratory Committee), C. A. Fux, H. F. G. (president of the SHCS), D. Haerry (deputy of “Positive Council”), B. H., H. H. Hirsch, M. Hoffmann, I. Hosli, M. Huber, C. R. Kahlert, L. Kaiser, O. Keiser, T. Klimkait, R. K., H. Kovari, B. L., G. Martinetti, B. Martinez de Tejada, C. M., K. J., Metzner, N. Muller, D. Nicca, P. Paioni, G. Pantaleo, M. Perreau, A. Rauch (chairman of the Scientific Board), C. Rudin (chairman of the Mother & Child Substudy), A. U. Scherrer (head of Data Centre), P. Schmid, R. Speck, M. Stockle, P. E. T., A. Trkola, P. Vernazza, G. Wandeler, R. Weber, and S. Yerly.

Financial support. This work was supported by the SHCS (project 836), the Swiss National Science Foundation (SNSF; grant 177499), and the SHCS Research Foundation.

Potential conflicts of interest. B. L. received personal fees from Kantonsspital Baselland (for consultancy and statistical analyses), Liestal, Switzerland, during the conduct of the study, and reports personal fees from Gilead for lectures and ViiV for advisory board service, outside the submitted work. E. B. reports grants/support from the SNSF to their institution during the conduct of the study. E. B. has received consulting fees from Gilead Sciences, MSD, ViiV Healthcare, Pfizer, and AbbVie and travel support from Gilead Sciences, MSD, and Pfizer, all paid to their institution and all outside the submitted work. M. C. reports grants/support from Gilead, MSD, and ViiV, payment for expert testimony from Gilead, MSD, and ViiV, and travel support from Gilead, all paid to their institution and all outside the submitted work. R. K. reports grants/support from Gilead Sciences, paid to their institution, outside the submitted work. H. F. G., outside this study, reports grants from the SHCS and the SNSF, during the conduct of the study; grants from the SHCS, the SNSF, the National Institutes of Health, Gilead (unrestricted research grant), and the Yvonne Jacob Foundation, all paid to their institution; personal fees as an advisor/consultant for Merck, ViiV Healthcare, and Gilead Sciences and a data and safety monitoring board member for Merck, paid to their institution; payments made to coauthor for consulting fees from Gilead and Merck; and payments made to coauthor for data and safety monitoring board/advisory board services from Gilead, Merck, and ViiV, outside the submitted work. P. E. T.’s institution reports unrestricted grants from Gilead and ViiV, outside the submitted work, and P. E. T. reports grants/support from the SHCS, paid to their institution, during the conduct of the study, and fees from Gilead and ViiV for advisory board service, paid to their institution, outside the submitted work. All other authors report no potential conflicts.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Author notes

I. C. S. and C. W. T. contributed equally to this work.