Clinical and Genetic Determinants of Subclinical Coronary Atherosclerosis in Healthy Asian People

Coronary artery disease (CAD) is a formidable global health challenge, accounting for substantial mortality and morbidity.¹ Among the multimodal management strategies for this burden, early detection and primary prevention schemes are of paramount clinical and public health importance. Early detection of CAD may allow for lifestyle changes or pharmacological interventions, such as aspirin, lipid-lowering, and blood pressure-lowering therapies, which are effective in reducing risk of cardiovascular events.²

Traditionally, risk prediction for CAD has predominantly relied on established cardiovascular risk factors such as age, sex, blood pressure, lipid levels and smoking status.³ However, evidence is emerging that the genetic make-up of an individual may further improve risk prediction.⁴ Genome-wide association studies (GWAS) offer possibilities to enhance our understanding of underlying biological mechanisms in CAD and may provide objective risk stratification.⁵ However, data are lacking, particularly in Asian people.

Coronary artery calcium (CAC) scoring using CT is a useful and simple tool to assess the presence of coronary atherosclerosis with high predictive value for future cardiovascular events.^6–8 We evaluated the accuracy of traditional clinical as well as genetic prediction models for subclinical coronary atherosclerosis using CAC in a cohort of Asian people without known cardiovascular disease.

Methods

Study Design and Population

SingHEART is a prospective, population-based cohort study of healthy adults living in Singapore. Its design has been described and published previously.^9,10 In brief, SingHEART recruits asymptomatic adults aged 21–69 years without known cardiovascular disease and/or diabetes and aims to evaluate the development of future cardiovascular disease. Participants in our study were included from October 2015 until November 2021. They were excluded if they had any prior cancer, autoimmune or genetic disease, psychiatric illness, asthma, chronic lung disease or chronic infectious disease.

Data on demographics, socioeconomic status, medical history, lifestyle, diet and exercise and quality of life were obtained through standardised questionnaires. Blood investigations included full blood count, renal and liver function, fasting lipids and fasting blood glucose. Furthermore, a resting ECG, a 24-hour ambulatory blood pressure measurement and a 5–7 days activity and sleep tracker (Fitbit Charge HR, Fitbit Inc., San Francisco, CA, USA) were performed. Written informed consent was obtained from all participants and the study was approved by the institutional ethics review board.

Coronary Artery Calcium Scoring

All subjects aged >30 years underwent a CAC assessment at baseline. Electron beam CT scanning using contiguous 3-mm slices during a single breath hold was performed by a 320-slice CT scanner (Canon Medical Systems) with a tube voltage 120kVp, tube current 200–400 mA and gantry rotation time 350 ms. Using ECG triggering with RR-gating, scans covered a single heartbeat. CAC scores were calculated using the Agatston method using a Vitrea Workstation (Canon Medical Systems).¹¹ A score of 0 indicated the absence of calcified plaque, 1 to <10 minimal plaque, 10 to <100 mild plaque and ≥100 moderate-to-severe plaque.

Genome Sequencing and Bioinformatics Analysis

Whole-genome sequencing (WGS) and secondary analysis of sequencing data according to the Genome Analysis Tool Kit best practices was carried out as described previously.¹² From the resulting Variant Call Format file, genotypes of candidate single nucleotide polymorphisms (SNPs) described previously were extracted for incorporation in further analysis.^13–15 These studies were selected after a systematic PubMed search focusing on SNPs associated with CAC in Asian populations published until June 2023. Choi et al. conducted a GWAS in an asymptotic Korean cohort, also using CAC as the endpoint.¹³ Lu et al. performed GWAS in a large Han Chinese cohort with established CAD.¹⁴ Iribarren et al. validated a large number of previously discovered SNPs in a multi-ethnic cohort of patients with CAD events.¹⁵ The full list of SNPs tested and the corresponding references are found in Supplementary Table 1.

Statistical Analysis

Continuous variables with normal distribution are presented as mean ± SD; continuous variables with a skewed distribution are presented as median (interquartile range). Binary and categorical variables are presented as numbers (percentages). Comparisons of population characteristics were tested using Student’s t-test, Mann–Whitney U-test or χ2 test, as appropriate. Clinical, genetic and combined logistic regression models to predict moderate-severe CAC (CAC score ≥100) were made, as well as for different cut-offs of CAC (>0, ≥10). For the prediction models, univariate logistic regressions were performed, after which variables with a p-value <0.1 were included in multivariate analysis. Backward stepwise regression was then performed to make the final prediction models. The performance of the models was assessed using the Akaike’s information criterion (AIC), area under the curve (AUC) with DeLong test and Brier score. A sensitivity analysis for the models was also performed excluding activity variables that may not be readily available routinely (number of daily steps and minutes asleep). A two-sided p-value <0.05 was considered significant. Statistical analyses were performed using Stata version 18.0 MP (StataCorp).

Results

Population Characteristics

A total of 845 individuals with available CAC scoring were included. Of these, 238 (28.2%) had any CAC and 74 (8.8%) had a CAC score of at least 100. Most (94.0%) were Chinese. Clinical characteristics are depicted in Table 1. Notably, individuals with high CAC scores were older, more often men, and had higher blood pressure, higher blood glucose and lipid levels and worse kidney function (all p<0.05).

Predictors of Moderate to Severe Coronary Artery Calcium (Score ≥100)

Independent predictors for moderate-severe CAC are depicted in Table 2. On multivariate analysis, significant clinical predictors were higher age, male sex, higher blood pressure, higher blood glucose levels, more steps, longer sleep and a positive family history for ischaemic heart disease. From the 46 SNPs that were previously reported to be related to the presence of CAD, three SNPs were found to be independently associated with severe CAC: rs515135, rs2047009 and rs10757272. In the combined routine and genetic model, all routine predictors remained significantly associated with severe CAC, as well as two SNPs: rs9632884 and rs515135.

Comparison of Model Performance

Performance metrics for the three models for moderate-severe CAC are shown in Table 3. The clinical prediction model had an AIC of 323.5, AUC of 0.89 (95% CI [0.85–0.93]) and Brier score of 0.0570. The genetic prediction model showed an AIC of 444.3, AUC of 0.64 (95% CI [0.58–0.71]) and Brier score of 0.083. Compared with the genetic prediction model, the clinical prediction model showed a better AIC (>2 difference), better AUC (p<0.001) and better Brier score. Lastly, the combined model yielded an AIC of 279.8, AUC of 0.91 (95% CI [0.87–0.94]; p=0.0026) and Brier score of 0.0570. When comparing the combined model with the clinical prediction model, the combined model had a better AIC (>2 difference) and better AUC (p=0.0026) but equal Brier score.

Models Excluding Activity Variables

In a clinical model without the Fitbit-derived variables (sleep, steps and flights of stairs), results were similar. Higher age (OR 1.19; 95% CI [1.14–1.24]; p<0.001), male sex (OR 3.93; 95% CI [2.07–7.44]; p<0.001), higher systolic blood pressure (OR 1.03; 95% CI [1.01–1.06]; p=0.002) and family history of CAD (OR 2.33; 95% CI [1.13–4.80]; p=0.002), as well as higher aspartate transaminase (OR 1.04; 95% CI [1.01–1.07]; p=0.005) were independently predictive of moderate-severe CAC. The performance of this model was moderately similar to the clinical model with activity variables (AIC 353.9; AUC 0.89, 95% CI [0.85–0.92]; Brier score 0.0606). All these variables remained significant in a combined model with the addition of two SNPs (rs515135, rs1333042); all p<0.05. The performance of this model also was moderately similar to the combined model with activity variables (AIC 310.9; AUC 0.90, 95% CI [0.86–0.93]; Brier score 0.0635).

Other Cut-offs of Coronary Artery Calcium Scores

The performance metrics of the models for the other cut-offs for CAC are presented in Supplementary Materials Tables 2–15. In brief, results were similar with age, male sex, systolic blood pressure and blood glucose being constantly independent predictors, but with poorer performance metrics than for the ≥100 CAC score cut-off.

Discussion

CAD, a leading cause of death in Asia, represents a key target for establishing strategies aimed at early detection and prevention.¹⁶ In this study, we developed and evaluated three prediction models for moderate-severe CAC in a healthy Asian cohort. Our main findings were as follows: moderate to severe CAC was present in one in twelve people; these people had well-known risk factors for CAD; the clinical model was superior to the genetic model to predict a high CAC; and a combined routine and genetic model had the best performance, although the incremental improvement was minor.

We found that older males had a higher proportion of severe CAC, consistent with studies in other (ethnic) populations.^17,18 Interestingly, higher fasting blood glucose levels were associated with CAC, even in our population without diabetes. Indeed, it has been shown that hyperglycaemia is an independent risk factor for cardiovascular events, further fuelling the importance of early recognition and treatment of diabetes.¹⁹ Unique to the SingHEART study, we investigated several variables derived from activity trackers (Fitbit). More daily steps were found to be associated with increased CAC. Although counterintuitive, it is well established that physical activity may elevate CAC scores.^20,21 However, this increased CAC severity may not lead to higher cardiovascular mortality, as it has been established that increased physical activity consistently decreases risk of CAD events.²² From the Fitbit data, we also found that longer sleep was independently associated with severe CAC. Although this is a novel finding, the association between poor sleep or sleep irregularity and atherosclerosis has been previously described.^23,24 The quality and type of sleep as well as further validation of the sleep data was not available in this study and will be the work of further investigation. Nevertheless, in a sensitivity analysis excluding such activity variables that may not be so readily available, similar results were obtained. We do emphasise that these findings are hypothesis-generating and we cannot rule out all confounders that might have influenced the results. These novel variables should be further validated.

Besides routine risk prediction, we developed genetic prediction models based on 46 previously described SNPs. Evidence is increasing that genetic prediction models may add incremental value over routine risk prediction, yet WGS studies in healthy Asian people are sparse. Given the sample size of our study, it was not possible to conduct our own GWAS analysis, therefore we relied on previously reported results. It should be emphasised that some of these studies were performed using CAD events as outcome, instead of CAC score, as these studies are extremely sparse in Asian populations. From three relevant studies, we analysed the reported SNPs and found that three of these SNPs remained independently associated with severe CAC.^13–15 The study by Choi et al. was most similar to our approach, yet only identified one significantly associated SNP.¹³ The SNP rs515135 was previously found to be associated with CAD in a Chinese population.²⁵ The SNP is located on chromosome 2p23–24 on the APOB gene. This gene is functionally important for the removal of LDL cholesterol from the circulation.²⁶ SNP rs2047009 was also independently associated with severe CAC. It is located on locus 10q11 on the CXCL12 gene.²⁷ High blood levels of CXCL12 have been found to be associated with atherosclerosis and adverse events in patients with ischaemic heart disease.²⁸ SNP rs10757272 was found to be independently associated with severe CAC in asymptomatic Koreans.¹³ The SNP is located on the CDKN2B-AS1 gene in chromosome 9p21.3. This locus has been found to be associated with greater burden of obstructive coronary artery disease.²⁹ Lastly, SNP rs9632884 remained independently associated with severe CAC in the combined model. Located on chromosome 9p21, it is not located in a gene but was found to be associated with CAD in white people.³⁰ In a Chinese population, the SNP was associated with higher lipid levels.³¹

To compare the performance of logistic regression models, we compared three different metrics. The goodness of fit was compared using the AIC, where we found that the combined model was the best performing of the three models (>2 difference). For discriminatory ability, the AUC was compared. The AUCs of both the clinical and combined model were high, with an approximate AUC of 0.89 and 0.91 respectively, with the combined model having a statistically significantly better AUC, although the magnitude of this significance clinically may be marginal. Contrarily, the Brier score, a measure of accuracy, was similar for clinical and combined models. The genetic model consistently performed the least well on all metrics. This stresses that besides genetic factors, environmental and social factors may also play a significant role in the pathogenesis of CAD.³² Although adding SNPs to the clinical variables improved the AIC and AUC, the question remains whether this is a clinically relevant finding. With the clinical model already having a good performance, it may potentially not be cost-effective to employ an expensive, non-selective genetic screening strategy in the general population. Interestingly, even in the combined model, a positive family history for ischaemic heart disease remained independently associated with severe CAC despite there being SNPs in the model. There may be other genetic determinants not yet fully elucidated. The role of a more targeted genetic screening strategy will be the work for further research. Nevertheless, the importance of strategies targeting the significant modifiable risk factors found such as high blood pressure and high blood glucose levels should be emphasised.

Study Limitations

Several limitations exist. Firstly, as part of the cross-sectional design of this sub-study, clinical outcomes are not available. However, as part of the SingHEART study, this cohort will be followed-up for long-term events, which will subsequently be reported. It should be noted that the prevalence of at least moderate CAC was relatively low, which introduces optimistic bias and risk of overfitting. Secondly, bias may be introduced as these participants were volunteers from the general population. As such, they might have more favourable health-seeking behaviour, as reflected by the low prevalence of smoking. Thirdly, most of our cohort was Chinese and the results may not be generalisable to other Asian or Western ethnicities. Lastly, while the CAC score is a useful surrogate for coronary atherosclerosis, it does not exclude presence of noncalcified, or even obstructive, plaque.³³ In addition, the interpretation of severity of CAC scores may be complicated by differences in sex and ethnic groups.¹⁷ Ideally, the results of the present study should be validated in an external, independent cohort.

Conclusion

In this healthy Asian cohort, a clinical risk prediction model for moderate-severe subclinical coronary atherosclerosis performed better than a genetic risk prediction model. Adding genetic predictors to the clinical model slightly increased the predictive power. The focus of primary prevention in CAD should remain on modifying lifestyle factors.

Click here to view Supplementary Material.