Keywords

History, Utility, and Sex Biases in Clinical Risk Scores for CAD

Historically, cardiovascular disease (CVD) risk assessment was initiated during the middle of the twentieth century spurred on by the spike in deaths in the USA due to coronary artery disease (CAD) that rose from ~10% of all deaths in 1900 to ~40% by 1960 [1]. This disease epidemic resulted in the Framingham Heart Study (FHS) being founded in 1947 by the US National Heart Institute in order to study and uncover causes of CVD in a relatively stable and well-defined community of mostly European ancestry [1]. Over a decade later in 1960, results from the FHS resulted in the initial concept of coronary risk factors (CRFs, see Fig. 38.1 for a comparison of their predictive performance in Framingham males and females), including, for example, high blood pressure, high cholesterol, history of smoking, type 2 diabetes (T2D), age, and sex that are observable in the preclinical phase before the disease occurs and are predictive of future CVD outcomes. After this initial success in providing some basic translational disease indicators, research over the following 20 years provided deeper understanding of the multiplicative nature of CRFs and the development of a cardiovascular risk prediction algorithm called the Framingham risk score (FRS) pioneered again by Framingham investigators. The first Framingham risk prediction algorithm to predict CAD (FRS-CAD) was designed in 1998 to combine individual patient’s risk factor information and produce a 10-year absolute risk score for angina, myocardial infarction (MI) , or death due to CAD in order to help clinicians decide whether therapy should be initiated [2]. In 2001, the Third Report from the Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III) recommended a modified 10-year Framingham risk score for CAD (ATP III-FRS-CAD) combined with a web-based calculator for use in patients exhibiting two or more risk factors to decide on thresholds for treatment of low-density lipoprotein (LDL) cholesterol [3, 4]. It was estimated that the use of these risk prediction scores increased the average life expectancy of CAD patients by 3 years in the USA [5].

Fig. 38.1
figure 1

Variation in lifetime coronary artery disease (CAD) risk depending on coronary risk factors (CRFs) in Framingham men and women. Each CRF was divided into quintiles (only highest, medium, lowest shown in Kaplan-Meier plots). Dashed lines are 95% CIs. Values shown in plots: Harrell’s C-index difference (ΔC) and significance test (assessed using a correlated jackknife test) between two models (with/without each CRF); net reclassification improvement (NRI) and associated p-value. SBP systolic blood pressure, TC total cholesterol; HDL, high-density lipoprotein cholesterol; T2D, type 2 diabetes. p-values: *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

While the FRS has often been the standard against which other subsequently developed CAD risk prediction tools are measured [2, 6,7,8,9], the general applicability of Framingham risk equations to modern populations has been debated over the last 10–15 years. This was partly because the FRSs are based on an ethnically and geographically limited, male-dominated sample when cardiovascular risk profiles were less well developed and also because many recent studies in diverse populations suggest that the FRS poorly classifies risk, particularly in women [10,11,12,13,14,15]. This is likely partly due to sex differences in progression and outcomes of CAD. For example, approximately 20% of all coronary events in women occur in the absence of major CRFs [16], and many women with CRFs do not experience coronary events [17]. As a consequence, the Reynolds Risk Score (RRS) for women was developed in 2007 that included C-reactive protein and parental history of premature CAD [18], and a revised Framingham risk score (FRS-CVD) that included increasingly common atherosclerotic cardiovascular disease (ASCVD ) events (i.e., stroke) was developed in 2008 [13]. The RRS for women [18] tested 35 potential risk factors on 24,558 initially healthy US women, and the resulting accuracy of the clinical risk score was greatly improved demonstrating the benefits of a sex-specific algorithm. This was further supported by a study in 2008 where the RRS for men was tested and optimized in 10,724 US men, which also significantly improved global cardiovascular risk prediction [19]. While we largely discuss US-based risk scores here, other non-US-based consortia have also developed other risk prediction algorithms (all simply include sex as a covariate) including, for example, the German Prospective Cardiovascular Munster Heart Study (PROCAM ) [20], the European Systematic Coronary Risk Evaluation (SCORE ) system [21], and QRISK2 [22].

Most recently, the American Heart Association (AHA) and the American College of Cardiology (ACC) developed a new ASCVD risk score (AHA-ACC-ASCVD) in 2013 (see Fig. 38.2 below for an indication of its performance in Framingham males and females) derived from four (including FHS data) racially and geographically diverse prospective cohort studies [23,24,25,26] that utilize the same traditional CRFs as the original FRSs while offering tailored equations for white and African-American men and women in order to improve and guide ASCVD risk-reducing therapy [26]. However, problems still reside in this most contemporary risk score as it overestimates risk in independent cohorts [27, 28]. Moreover, there is a certain level of misclassification of most clinical risk scores developed to date depending on sex. For example, DeFilippis et al. [29] compared performance of the AHA-ACC-ASCVD risk score with four other risk prediction equations (FRS-CAD, FRS-CVD, ATP III-FRS-CAD, RRS) utilizing the Multi-Ethnic Study of Atherosclerosis (MESA ) study, which is a community-based, sex-balanced, multiethnic cohort consisting of 54% women, 42% European, 26% African-American, 20% Hispanic, and 12% Chinese. The authors found that the performance of the five risk scores to discriminate between those who did or did not have an ASCVD or CAD event was fairly similar. However, the FRS-CAD, FRS-CVD, ATP III-FRS-CAD, and AHA-ACC-ASCVD overestimated CAD and ASCVD risk by 37–154% in men. In comparison, the RRS estimated ASCVD and revascularization events in men relatively accurately (9% discordance). In women, the FRS-CVD accurately estimated risk (8% discordance), while the FRS-CAD, ATP III-FRS-CAD, and AHA-ACC-ASCVD overestimated risk (46–67% discordance) and the RRS underestimated (−21% discordance) the rate of events [29].

Fig. 38.2
figure 2

Lifetime risk of coronary artery disease (CAD) depending on individual variation in the AHA-ACC-ASCVD clinical risk score in Framingham men and women. AHA-ACC-ASCVD scores were divided into quintiles (highest, medium, lowest shown in Kaplan-Meier plots). Dashed lines are 95% CIs. For Framingham males, integration of the AHA-ACC-ASCVD score into a simple CAD risk prediction model (first five principal components on genotypes) improved 10-year risk prediction (Harrell’s C-index, C-index difference (ΔC) between two models assessed using a correlated jackknife test) by 3.4% (p < 0.001) with a net reclassification improvement (NRI) score of 0.45 (p < 0.001). For females, addition of the same score did not significantly improve 10-year C-index but did significantly improve the NRI (0.43, p < 0.001). P-values: *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

In summary, none of the most well-known ASCVD and CAD clinical risk prediction scores in the USA are highly accurate for both men and women [29]. More generally, CRFs underlying clinical risk scores provide only modest discrimination and cannot fully capture underlying risk [30]. This has clinical relevance for the current 10-year predicted risk threshold of 7.5% in the USA: over- or underestimation is likely to result in individuals with lower or higher risk receiving or not reaching the cutoff value for preventative treatment (statins), respectively [31,32,33]. Further testing and refinement of clinical risk scores, tailored to distinct populations and sex, is needed. While the RRS and AHA-ACC-ASCVD developed between 2007 and 2013 offer sex-specific clinical risk prediction algorithms, we now know that many more sex differences exist in the progression of CRFs and cardiovascular outcomes between men and women [34,35,36,37] (discussed below) that need to be accounted for in order to improve prediction of CAD and broader CVD. With the ever-increasing availability of large clinical and biobank-scale data as well as inclusion of genetic risk predictors (discussed below), these will lead to new more powerful clinical risk prediction equations.

GWAS and Sex-Specific Genetic Effects on CAD

Typical of common diseases, individual risk for CAD is modulated by the interplay between lifestyle and genetic factors [38]. Some of the first evidence that CAD had a genetic component was through early clinical observations [39], twin [40, 41] and cohort [42, 43] studies, which found increased CAD risk depending on whether there was direct or indirect family history of the disease and also estimated the heritability of CAD as ~50%. Genome-wide methods estimate CAD heritability between 40% and 50% [44]. Collectively these seminal studies laid the groundwork that helped drive the development of modern genetic tools to better define the underlying genetic architecture of CAD and begin to translate these findings into practice.

Historically, linkage analysis studies on families with a predisposition to early-onset CAD provided some initial insights into Mendelian or monogenic-like drivers of CAD. Linkage analysis can be an effective means of utilizing large multigenerational families to identify highly penetrant variants responsible for disease. Familial hypercholesterolemia defined by a heritable pattern of increased cholesterol and premature CAD was first described in 1938 [45], and in 1985 a deletion in the LDLR gene was found to be causative [46] demonstrating that a discrete mutation in a single gene predisposed to CAD. Subsequent family-based studies have identified further mutations in APOB, PCSK9, and ABCG5/ABCG8 [47,48,49,50] that contribute to familial hypercholesterolemia, but beyond this condition that predisposes to CAD, the use of family studies to identify drivers of CAD has been limited. This was especially clear in linkage studies performed on families with more common presentations of CAD that failed to detect any susceptibility loci [51, 52]. Therefore, despite the tendency to cluster in families, these earlier studies highlighted the complex polygenic nature of CAD.

After the sequencing of the human genome in the late 1990s and early 2000s, which resulted in ~1.5 million single nucleotide polymorphisms (SNPs), researchers used small-scale candidate gene studies (i.e., using previously identified potentially causative variants) to identify genes involved in disease; however, this underpowered approach with small sample sizes often resulted in false-positive findings [53, 54]. Many of the larger-scale candidate gene studies on CAD (or stroke) performed in the mid-late 2000s pooled data on both sexes and added sex as a covariate [55,56,57,58,59,60] without consideration for potential sex-specific effects. However, in 2002, a Japanese case-control study that tested variants predictive for MI found differences in the significance of those between males and females when they were analyzed separately [61], supporting the presence of sex-specific genetic risk factors for CAD. In 2008, a case-control study of over 14,000 Finnish individuals where authors tested for sex-specific effects on 46 candidate genes for CAD and CVD found that while some variants conferred risk in both sexes, others showed significant effects only in males (e.g., rs2069840 in IL6 for CVD) and females (e.g., SNPs in SELENOS for stroke) with a larger number of sex-specific effects for women than men [62]. These studies showed some of the first evidence for sex differences in the effects of known variants in genes underlying CAD and CVD.

Subsequent further sequencing with the availability of cheaper genotyping chips (designed to capture common variation in populations) and careful documentation by the International HapMap Consortium of ~3.1 million SNPs in 270 individuals from four ethnic backgrounds in 2007 [63, 64] provided the necessary resources for initial genome-wide association studies (GWAS ). The first CAD GWAS was published in 2007, and during that time, three independent groups reported variants at the 9p21 locus with the risk allele significantly increasing risk of CAD in European populations by approximately 30% [65,66,67]. Following those initial discoveries, progressively larger sample sizes have been used to define many variants with relatively small effects that contribute significantly to CAD. Over the last 5 years, large international consortia have utilized massive sample sizes from mixed ethnic backgrounds in GWAS meta-analyses for CAD and have identified ~60 distinct genetic loci significantly linked to CAD [68,69,70,71,72,73]. Six more common variant loci were identified in 2017 [74], and rare variant (captured with whole exome/genome sequencing, i.e., not included on genotyping chips) association studies have identified at least nine genes in which mutations significantly alter CAD risk [75].

The majority of common CAD variants identified to date have a minor allele frequency of >5% in most worldwide populations, are located in nonprotein-coding regions of the genome highlighting their impact on CAD risk through gene regulation [73, 76], only cause modest increases in CAD risk (i.e., typically <20% change in risk per allele), collectively explain ~30–40% of CAD heritability [72, 73], and just under half (~40%) modulate CAD risk by influencing CRFs such as cholesterol, T2D, and hypertension [68, 72, 77] suggesting that the mechanisms whereby a majority of these loci mediate CAD risk remain unknown.

Since the first CAD GWAS was performed a decade ago (including previous candidate gene studies), researchers to date have typically not tested for sex-specific effects despite earlier findings (particularly in candidate gene studies) that these exist. Most studies have pooled men and women (adjusting for sex as a covariate) and typically account for population structure by including the first 5–10 principal components derived on genotypes from the sample. For example, one of the most recent GWAS meta-analyses of CAD [73] that used 60,801 CAD cases and 123,504 controls from a mix of males and females from different ancestral backgrounds (77% European, South (13% India and Pakistan) and East (6% China and Korea) Asian, Hispanic and African-American ~4%) included sex as a covariate. However, a slightly earlier study [72] published by the same consortium with a similar sample performed subgroup analyses by sex, and while they observed no higher risk trends for males vs females for any of the 46 genome-wide significant CAD risk loci identified in the main analysis (with sex added as a covariate), they did find one new locus (rs16986953 in gene desert, 1.3 Mb away from APOB) that reached genome-wide significance in males [72] but not females. Other non-CAD large-scale GWAS studies have revealed significant sex differences in contributing genetic variants [78,79,80] demonstrating that these are present in many polygenic traits and diseases.

A small subset of genome-wide CAD studies focusing on women exist but have often been limited by sample sizes with insufficient power to test for association [81]. For example, in 2012 Orozco et al. [82] tested for sex-specific associations in the Wellcome Trust Case Control Consortium of 399/1527 cases and 1492/1446 controls, for women and men, respectively, and found no CAD variants (previously identified from GWAS ) were significant in women or men. More recent studies with larger sample sizes have found some convincing sex-specific genetic effects on CAD. Goodarzynejad et al. [83] in a sex-stratified study discovered a SNP in SCARB1 (plasma membrane receptor for HDL) was associated with CAD in women but not men. A SNP in CPS1 has also been found significantly associated with CAD in women but not men [84]. Liu et al. [78] found a SNP on 9p21 that showed a large (p = 1.38 × 10–8) male-specific association with CAD, but not females (p = 0.167). Taylor et al. [80] found sex-specific SNPs associated with risk factors (HDL, LDL) for CAD. This supports earlier findings from candidate gene studies that uncovered sex-specific genetic effects on CAD. Other recent GWAS have found autosomal variants contributing to complex traits in a sex-specific manner (e.g., adiposity [85]). Therefore, the investigation of sex-specific effects on CAD with larger better-powered studies is warranted. Resolving the full complement of shared and sex-specific CAD genetic loci will help provide a better understanding of the pathological pathways in progression and thus improve individually tailored treatments for this disease.

Notably, GWAS to date have almost exclusively focused on autosomal variation [86], even though the X chromosome is included on all current microarrays, and as a result, the role the X chromosome plays in CAD (and most complex diseases) remains largely unknown. This is mainly due to difficulty in accounting for analytical problems arising from the X’s unique mode of inheritance and expression (i.e., dosage compensation in females), but recent analytical tools to perform XWAS [87] could begin to resolve this. The Y chromosome might also bring further understanding of sex-based differences in CVD development and outcomes [37]. For example, significant associations between Y chromosome variants and blood pressure have been found in different populations [88, 89], and some studies have suggested that blood pressure in men is largely determined by paternal (and not maternal) blood pressure status [90]. In males, the Sry locus of the Y chromosome contributes to hypertension through regulation of tyrosine hydroxylase, whereas in women, estrogen modulates the activity of this enzyme [36]. Therefore, modification of blood pressure, a key risk factor for CAD, is partly regulated by different genetic loci interacting with environmental stimuli that vary between males and females (i.e., changes in estrogen with pregnancy and menopause, discussed below). For an in-depth discussion of sex-specific differences in CVD related to the sex chromosomes, see Sampson et al. [37], Kling et al. [36], and Regitz-Zagrosek and Kararigas [35]. A recent study has also shown CVD is more prevalent among patients with pathogenic mtDNA mutations [91]. Given the seemingly incomplete autosomal nuclear genetic picture of CAD, variation in sex chromosomes and mtDNA could represent part of the missing heritability of CAD.

Genetic and Genomic Risk Prediction in CAD

While many genetic loci for CAD have now been identified, individually these are not clinically useful. For example, based on the most recent meta-analysis [73], the 9p21 locus remains the strongest genetic risk factor for CAD boasting an impressive significance of p = 2.29 × 10−98; however, the odds ratio (OR) for the effect allele is a modest 1.21 [92]. This represents the basis for recently developed genetic risk scores (GRS) that aim to improve CAD risk prediction. By combining GRS with or without CRFs (often referred to as modifiable risk factors because of the large environmental component to their variation), this should provide more accurate risk stratification of individuals [33, 93]. GRS are calculated essentially by counting the number of risk alleles (adjusted for effect size) inherited for each individual in a given population and provide a quantitative measure of genetic risk [94]. While family history might naturally serve as a substitute for genetic risk, individual risk loci and GRS predict CAD independent of family history, and the association between family history and CAD tends not to be attenuated after the inclusion of GRS [95]. This suggests that the proportion of heritability captured by current GRS is not currently high enough to erode the predictive power of family history and that family history of CAD partially reflects common environment effects.

The first attempt to include genetic markers for CVD prediction was in 2008 where variants associated with cholesterol were used to predict cardiovascular events [96] and extended later with nine significant independent risk variants for CAD [68]. These initial studies showed that individuals in the top GRS quintile had a twofold increased risk of CAD compared to those in the lowest GRS quintile [68] and paved the way for the exploration of various GRS based on different combinations of genetic loci for CAD in different populations [96,97,98,99,100,101,102,103,104,105]. For example, Mega et al. [106] utilized a GRS based on 27 CAD genetic risk variants (GRS27) and showed that genetic risk is independent of CRFs. Shortly after, Tada et al. [95] incorporated these same 27 with additional 23 CAD risk variants to form a 50-variant GRS (GRS50) and tested its performance on 23,595 individuals from the Malmo Diet and Cancer Study. They found their GRS50 further significantly improved CAD discrimination (i.e., C-index: measures how well a model discriminates individuals with(out) the outcome of interest; values range from 0.5 (performs no better than random) to 1.0 (perfectly distinguishes outcomes)) and reclassification (estimates improvement in performance of risk prediction model after additional risk variable is included). A recent study in 2016 by Abraham et al. [107] advanced the GRS by taking account of 49,310 small-effect SNPs predicted to underlie polygenic CAD to construct a genomic risk score (GRS49k), and tested this on 12,676 Finnish men and women (Cox regression stratified by sex). The GRS49k confirmed the increased predictive value of including a large number of SNPs, by improving CAD risk prediction largely independently of CRFs (including family history) and significantly improved 10-year CAD risk prediction when it was combined with the ATP III-FRS-CAD or AHA-ACC-ASCVD clinical risk scores.

These studies show risk stratification provided by GRS is superior to clinical risk scores and the inclusion of many genetic risk variants significantly improves prediction of CAD outcomes. This could be particularly important in asymptomatic individuals without major risk factors. Because DNA is essentially stable across the lifetime of individuals, genetic risk can be obtained at birth. GRS could therefore be particularly useful for very early (i.e., young adults) intervention, which is especially important for a disease like CAD that begins to develop very early in life and is a result of lifetime modifiable exposures (diet, smoking, exercise) interacting on a background of low to high genomic risk. Adherence to a healthy lifestyle will be particularly important for young individuals with a high CAD GRS. For example, a recent analysis of high GRS individuals found a 46% attenuation of CAD risk in those with a favorable versus unfavorable lifestyle [38], and in another study a 50% reduction in CAD risk was achieved through statin therapy in those with high GRS [106]. Another recent study importantly showed the ability of GRS not only to predict incident CAD but also recurrent CAD independent of all CRFs and family history [106]. This shows that GRS hold great potential to improve both primary and secondary prevention strategies for CAD.

While the performance of GRS for CAD prediction is now being tested on less well-studied populations [98, 108,109,110,111], almost all GRS studies to date have not examined sex-specific effects in GRS. Of the few studies that have, the predictive performance and ability of GRS to stratify high- and low-risk individuals for CAD tends to perform much more poorly in women. For example, Paynter et al. [101] used a 101 SNP GRS based on GWAS loci significantly associated with CVD (or intermediate phenotypes) and found the GRS hazard ratio was initially weak (1.02, p = 0.006) but then not significantly associated with incident CVD in multivariable models in the Women’s Genome Health Study (WGHS ). Compare this to some of the most recent GRS predicting CAD, where hazard ratios for GRS are typically highly significant ranging from 1.27 to 1.74 [102, 103, 107] in multivariable models. A recent study demonstrated that the ability of the GRS49k and CRFs to stratify high- and low-risk individuals for CAD performed much better in males than females (e.g., see Fig. 38.3 and Fig. S11–S13 in Abraham et al. [107]). This is also supported statistically by subsequent analyses on these cohorts that highlight male-female differences for key risk prediction performance indicators (see Fig. 38.3 for estimates).

Fig. 38.3
figure 3

Differences in the ability of the genomic risk score (GRS49k) to stratify men and women into high and low coronary artery disease (CAD) risk groups. Plots show Kaplan-Meier estimates for cumulative risk up to age 75 in 12,676 FINRISK and 3406 Framingham Heart Study (FHS) men and women. GRS were grouped into quintiles (only 0–20, 40–60, 80–100% shown), with vertical bars showing the age at which each group attained a cumulative CAD risk of 10%. Dashed lines are 95% CIs (Sourced from Abraham et al. [107]). For Framingham males, integration of the GRS with a simple CAD risk prediction model (first five principal components on genotypes) or complex model (first five PCs, total and HDL cholesterol, systolic blood pressure, T2D, smoking, treatment for high blood pressure, family history of CAD) improved 10-year risk prediction (Harrel’s C-index, C-index difference between two models assessed using a correlated jackknife test) by 2.5% (p = 0.02) to 1.6% (p = 0.05), respectively. For Framingham females, integration of the GRS with simple and complex CAD risk prediction models did not improve 10-year risk prediction with C-index changing by 0.2% (p = 0.78) to 0.03% (p = 0.86), respectively. The same sample and methods as described in Abraham et al. [107] were used to obtain these estimates

Full size image

Less successful prediction and stratification in females is perhaps not surprising in these studies given the combination of genetic and phenotypic biases that currently exist in CAD research. Current GRS are based on variants from GWAS that insufficiently capture sex-specific genetic effects on CVD and CAD (particularly females, discussed above), and clinical definitions for CAD outcomes (i.e., progression, presentation, age at onset) are male-dominated, resulting in lower specificity and accuracy of CAD outcomes captured and predicted in females. Testing a sex-specific GRS for CAD might be appropriate. Sex-tailored GRS have been constructed in other fields for conditions and diseases other than CAD. For example, female-specific GRS have been tested in relation to natural menopause [112], polycystic ovary syndrome (PCOS ) [113], multiple sclerosis [114], and breast cancer [115], and a male-specific GRS has been tested for prostate cancer [116]. Indeed, developing sex-specific GRS can be useful when there are known differential sex differences in clinical phenotypes [117]. For CAD, the success of this will depend on better annotation of sex-specific variants identified in large GWAS that specifically search for sex-specific effects.

Improvement of Future Clinical and Genetic Risk Scores for CAD Through More Precise Definitions of Sex-Specific Effects

As is the case for many common human diseases, improving our understanding of the genetic architecture underlying CAD, which is essential for enhancing predictive performance of GRS, must also involve parallel improvements in the definitions of sex-specific differences in phenotypic presentations of CAD, which will reduce case/control misclassification and improve precision and risk stratification of males and females. The performance of clinical risk scores will also benefit from more refined CRFs that account for sex differences in their progression and interactions with environmental stimuli. As we discuss below, using exactly the same CAD and CRF definitions for males and females is not optimal in the era of precision medicine.

Firstly, because pathophysiology of CAD in men and women is quite different, grouping males and females and simply including sex as a covariate will reduce trait specificity and cause a level of bias in GWAS [118] from sex-based misclassification of CAD cases/controls. This bias is likely further compounded (especially for rarer genetic variants or variants with weaker yet significant effects) with the disproportionate ratio of men and women when they are pooled in GWAS . In general, analyzing sex differences in genetic associations for CAD risk is complicated because the age at onset is sex-dependent. For example, mortality from CAD tends to appear more commonly [119] and much earlier in males than females [120, 121]. Various factors have been proposed (e.g., psychological stressors, access and utilization for health care, inappropriate surveillance, and treatment guidelines for women) to contribute toward this mortality difference [36].

Progression of CAD shows some distinct differences by sex. For example, more severe structural/functional abnormalities in epicardial coronary arteries typify CAD progression in men, while more abnormal coronary reactivity, microvascular dysfunction, and plaque erosion/distal microembolism typify CAD progression in women [122]. This has led to the recent suggestion that the definition of ischemic heart disease (IHD) might be more appropriate for women than CAD [81]. Heart failure in women is often more associated with preserved ejection fraction (HFpEF), and if women suffer from left ventricular hypertrophy, they have a higher mortality risk compared to men [123]. Compared to women, men also present with more severe atherosclerosis in their coronary arteries. Correspondingly, MI typically occurs 10 years earlier in men [35]. Some recent studies suggest that most newly identified CAD GWAS loci mainly predict atherosclerosis [124]. Given that the pathophysiology of CAD can be roughly divided into three major processes (plaque formation (atherosclerosis), plaque rupture, thrombotic response to plague rupture [125]) and atherosclerosis tends to be more typical of male CAD progression [125], this suggests many newly identified CAD loci might be more predictive in males. This corresponds with differences of GRS to stratify high- and low-risk males and females for CAD outcomes [107] discussed in Sect. “Genetic and Genomic Risk Prediction in CAD” above.

Risk factors for CAD (i.e., cholesterol, BP, T2D) also show distinct differences between men and women [126,127,128,129,130,131,132] that often vary with age (see Fig. 38.4 for examples in Framingham males and females). For example, even though men often present with systemic hypertension at an earlier age, the incidence of pulmonary hypertension is greater in women than men [133,134,135]. Further, hypertension and T2D can be triggered by pregnancy and menopause [36]. For example, mothers experiencing pregnancy disorders such as preeclampsia and gestational diabetes subsequently have a much higher risk of hypertension and T2D [136,137,138,139,140,141,142]. Menopause also involves significant changes in estrogen levels, and estrogen modulates tyrosine hydroxylase, which is a key enzyme in the development of hypertension [36]. Correspondingly, premenopausal women have a reduced risk for hypertension and IHD compared to men, but this reverses after menopause [143, 144]. There is some evidence that women may be more protected from atherosclerosis before menopause [35], which is supported by the observation that women who have hormonal (estrogen) disturbances due to PCOS develop atherosclerosis and subsequent MI earlier in life [145, 146]. Men with a mutation in the estrogen receptor gene (ESR1) also have earlier CAD [147, 148].

Fig. 38.4
figure 4

Variation in coronary risk factors (CRFs) between the ages of 20 and 80, 1955–2005, for males and females in the Framingham Heart Study. Note the very different lifetime profiles of cholesterol for males and females. At around 60 years of age, total cholesterol (TC) tends to be much higher in females compared to males, and high-density lipoprotein (HDL) cholesterol levels in females tend to be generally higher in females compared to males. Plots show the 3-D surface accurately derived from many repeated individual measures (total cholesterol measured 37,317 times in 6605 males, 44,798 times in 7518 females; systolic blood pressure (SBP) measured 45,990 times in 6626 males, 58,690 in 7546 females; HDL cholesterol measured 17,207 times in 5334 males, 20,095 times in 6210 females) in a generalized additive model. Each plot shows example of individual repeated measures from three randomly chosen individuals

Full size image

In summary, more accurate classification of phenotypic sex differences in CAD risk factors, environmental stimuli that interact with these differently by sex, and sex differences in clinical presentations of CAD will in turn allow for further and more accurate identification of genes in female-focused and male-focused studies [149] and better predictive models that utilize GRS and CRFs . This will become increasingly important as we move further into the era of precision medicine.