Tailoring cardiovascular risk prediction to females.

TA is supported by the Medical Research Council https://doi.org/10.13039/501100000265 , Precision Medicine Grant (MR/W006804/1). MdB and DMK are supported by an Intermediate Basic Science Research Fellowship and Research Excellence Award from the British Heart Foundation https://doi.org/10.13039/501100000274 (FS/IBSRF/23/25161, RE/24/130012). JAM received funding from Wellcome Fellowship 209589/Z/17/Z.

The development of CVD varies by sex and age, therefore cardiovascular risk estimation systems should incorporate both sex and age interactions with risk factors to improve ASCVD risk estimates. As the incidence of non-ASCVD in females continues to rise, it is crucial to adopt a more holistic approach to risk assessment that extends beyond the traditional ASCVD outcomes. The key risk factors for cardiovascular disease in females include traditional, female-prevalent disease, female-specific and social factors. Female-specific factors can be further divided into pregnancy- and non-pregnancy-related. There has been limited research for non-pregnancy-related factors such as heavy menstrual bleeding. This may be partly due to an insufficient number of younger females and follow-up periods which are not long enough to capture cardiovascular events. Furthermore, derivation cohorts used to derive the risk estimation systems often include older females, and most risk estimation systems can only be used in females from 30 to 40 years of age. QR4 is the only cardiovascular risk estimation system that can be used in young females from 18 years of age, which includes two female-specific risk factors. External validation of QR4 is essential, and further research is needed to evaluate the clinical added value of incorporating female-specific factors into cardiovascular risk estimation systems. Finally, alongside improving overall research practices and the coding of female-specific factors in healthcare records, future research must include populations from low- and middle-income countries, as well as smaller minority groups, to ensure improvement in the global prevention of cardiovascular disease in females.

Cardiovascular disease (CVD), which encapsulates atherosclerotic and non-atherosclerotic outcomes, remains a major global public health challenge and is one of the leading contributors to disability-adjusted life years and death in females worldwide. The estimated age-standardised prevalence and mortality of CVD in females is 6,403 cases per 100,000 (95% confidence interval (CI), 6,079–6,740) and 204 cases per 100,000 (95% CI, 181–222), respectively ( Vogel et al. 2021 , Vervoort et al. 2024 ). The recent State of the Art Review on global burden of cardiovascular disease in women ( Vervoort et al. 2024 ), the British Cardiovascular Society ( Tayal et al. 2024 ) and the Lancet Women and Cardiovascular Disease Commission ( Vogel et al. 2021 ) collectively highlighted the need to address cardiovascular health in females through improving: i) research into sex-specific mechanisms in the pathophysiology of cardiovascular disease, ii) understanding of sex-specific factors that may increase risk, iii) the recognition of the effects of socioeconomic deprivation globally, and iv) interventions that reduce risk of CVD in females. One individual-level method for reducing cardiovascular risk is the use of cardiovascular risk estimation systems. These systems rely on traditional risk factors to identify high-risk individuals and offer preventative therapies aimed at lowering their risk of future cardiovascular events. However, the performance of these systems has been shown to be suboptimal. Most cardiovascular risk estimation systems have a concordance index of around 0.7, meaning that there is a 70% probability that a person who experiences a cardiovascular event will have a higher risk score than someone who does not ( Pennells et al. 2019 ). In a study involving individual participant data from 360,737 individuals, Pennells et al. found both over- and under-prediction when testing four commonly used risk estimation systems ( Pennells et al. 2019 ). Therefore, novel strategies to improve the accuracy of cardiovascular risk estimation are urgently needed. Given that several female-prevalent and female-specific risk factors have been shown to elevate cardiovascular risk in females, incorporating these factors into risk estimation systems may enhance the accuracy of cardiovascular risk estimates for females. In this review, we provide insights into how sex differences may or may not affect traditional risk factor associations with CVD. We provide an overview of the range of cardiovascular risk factors beyond traditional risk factors that need to be considered in females, from female-prevalent disease risk factors to female-specific risk factors (pregnancy- and non-pregnancy-related), and we discuss the effect of social determinants of health on cardiovascular risk ( Fig. 1 ). We review clinically applied cardiovascular risk estimation systems and discuss the integration of female-specific factors within these systems. Finally, we discuss the role of novel approaches and future directions to refine risk prediction in females in order to prevent CVD. Since this review primarily focuses on biological sex, we use the terms females and males throughout for clarity. Cardiovascular risk factors in females.

NS has consulted for and/or received speaker honoraria from Abbott Laboratories, AbbVie, Amgen, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Hanmi Pharmaceuticals, Janssen, Menarini-Ricerche, Novartis, Novo Nordisk, Pfizer, Roche Diagnostics and Sanofi, and received grant support paid to his university by AstraZeneca, Boehringer Ingelheim, Novartis and Roche Diagnostics outside the submitted work. JAM has provided consultancy advice (all paid to institution; no personal remuneration) to Gedeon Richter. DMK has received honoraria from Roche Diagnostics and Abbott Diagnostics outside the submitted work. All other authors have no interests to declare.

The use of a cardiovascular risk estimation system is integral to the primary prevention of CVD. A derivation cohort that includes individuals free from cardiovascular disease at baseline is used for model development. Current cardiovascular risk estimation systems are developed using data from electronic health records, observational cohort(s) or a combination of both. Ideally, contemporary target population data from the country or region are used to optimise performance in common practice. All clinically applied risk systems are sex-specific and include risk factors chosen by developers to further optimise model performance ( Table 2 ). Cardiovascular risk estimation systems are usually externally validated on an independent dataset, based on a population with similar characteristics. Risk factors included in clinically applied cardiovascular risk estimation systems for females. DM, diabetes mellitus; SBP, systolic blood pressure; BMI, body mass index; eGFR, estimated glomerular filtration rate; HbA1c, haemoglobin A1c; ACR, albumin-creatinine ratio; CRP, C-reactive protein; MI, myocardial infarction; IHD, ischaemic heart disease; TIA, transient ischaemic attack; CHD, coronary heart disease; CABG, coronary artery bypass graft; AAA, abdominal aortic aneurysm. Commonly applied cardiovascular risk estimation systems for females are: QR4 ( Hippisley-Cox et al. 2024 ), ASSIGN V.2.0 ( Welsh et al. 2025 ), SCORE2 ( SCORE2 Working Group and ESC Cardiovascular Risk Collaboration 2021 ), SCORE2-OP ( SCORE2-OP Working Group and ESC Cardiovascular Risk Collaboration 2021 ), PREVENT ( Khan et al. 2024 ) and Reynolds ( Ridker et al. 2007 ). Differences in characteristics between derivation cohorts and selection of risk factors included in the models do exist ( Tables 2 and 3 ). QR4 and ASSIGN V.2.0 are developed and intended for use in the UK and Scottish populations, respectively. PREVENT and Reynolds are based on observational data from the US, and SCORE2 and SCORE2-OP were developed for use across Europe. SCORE2 and SCORE2-OP attempted to account for differences in baseline cardiovascular risk between countries by categorising them into low, medium, high and very high-risk regions, based on World Health Organisation ASCVD mortality data. Cohort characteristics of clinically applied cardiovascular risk estimation systems. ARIA, Atherosclerosis Risk in Communities; CARDIA, Coronary Artery Risk Development in Young Adults; CHS, Cardiovascular Health Study; FHS, Framingham Heart Study; JHS, Jackson Heart Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; CRIC, Chronic Renal Insufficiency Cohort; RBS, Rancho Bernardo Study; CPRD, Clinical Practice Research Datalink; IQR, interquartile range; SD, standard deviation. With regards to risk factors, diabetes is not included in SCORE2 or SCORE2-OP, as a specific tool for people with diabetes has been developed (SCORE2-Diabetes) ( SCORE2-Diabetes Working Group and the ESC Cardiovascular Risk Collaboration 2023 ). The QR4 risk model is the only model that includes body mass index. PREVENT, ASSIGN V.2.0 and QR4 include social deprivation status, with QR4 also adding ethnicity to the risk score. The developers of PREVENT decided not to include ethnicity as a predictor for ASCVD, reasoning that differences associated with ethnicity are primarily driven by social determinants of health rather than ethnicity itself. PREVENT does include antihypertensive and lipid-lowering medication as predictors for ASCVD risk, as well as laboratory parameters for renal function and diabetes. The Reynolds risk score includes C-reactive protein measurements. QR4 includes corticosteroids, antipsychotic medications, an additional 13 disease-specific risk factors and is the first and only tool to include the female-specific risk factors pre-eclampsia and postnatal depression ( Hippisley-Cox et al. 2024 ). Another key difference between risk systems is how they incorporate the effect of age on risk factors and the selection of CVD outcomes. Both QR4 and PREVENT incorporate age interactions in their risk models, whereas SCORE2 and SCORE2-OP are age-specific models designed for younger and older individuals, respectively. Although all risk scores include fatal and non-fatal ischaemic heart disease, coronary artery disease, myocardial infarction, stroke and cerebrovascular disease, some also account for hypertensive disease and abdominal aortic aneurysms (SCORE2, SCORE2-OP), non-atherosclerotic diseases such as heart failure (SCORE2, SCORE2-OP, PREVENT), arrhythmias (SCORE2, SCORE2-OP) and interventions such as coronary angioplasty (ASSIGN V.2.0, Reynolds). The incidence of non-atherosclerotic disease in females is rising ( Conrad et al. 2024 ), therefore it is crucial to adopt a more holistic approach to risk assessment that extends beyond ASCVD outcomes to other CVD. ( Tschiderer et al. 2023 ) reviewed nine studies that incorporated female-specific factors such as hypertensive disorders of pregnancy and gestational diabetes into risk estimation systems for primary prevention of CVD. Of these, only one study included factors unrelated to pregnancy, such as age at menarche or menopause, menopausal status and hormone replacement therapy. In addition, a recent UK Biobank study explored 13 female-specific factors, six of which were non-pregnancy related ( Doust et al. 2024 ). In all these studies, the addition of female-specific factors resulted in little or no improvement in model discrimination or reclassification. One possible reason for this could be that some female-specific factors, particularly those related to pregnancy, are closely linked to traditional risk factors or may be heightened by them. As a result, the inclusion of traditional risk factors might already capture the increased ASCVD risk associated with certain female-specific factors. Another explanation could relate to the characteristics of the derivation cohorts used to develop the models. Most risk estimation systems were developed using cohorts of older females, with mean ages typically ranging from 49 to 53 years ( Table 3 ), while the UK Biobank study included a slightly older group with a median age of 58 years ( Doust et al. 2024 ). However, female-specific risk factors are more prevalent in younger women; as a result, the data used to build existing models may not adequately capture these factors. QR4 is the only risk estimation system that used a younger population for its derivation cohort, with a mean age of 39 years. This tool was developed for a broader population aged 18–84 years, whereas other models are intended for those aged 30 and older. With the largest sample size for model derivation – 9,976,306 participants – QR4 may offer more robust insights. In contrast, the relatively small sample sizes in other studies could explain the lack of improvement in model performance observed elsewhere. Given these uncertainties, external validation of QR4 in diverse populations is crucial for enhancing our understanding of female-specific risk factors and improving CVD risk prediction. Finally, while metrics such as discrimination, reclassification and calibration offer valuable insights into model performance at population level, they do not fully capture the clinical value necessary to determine whether incorporating female-specific factors can clinically meaningfully impact the prevention of CVD in females. Further research is needed to assess the clinical added value of including female-specific factors in risk models, with a focus on metrics that extend beyond traditional performance indicators (e.g. decision curve analysis). Multiple strategies have been proposed to enhance cardiovascular risk prediction, such as the addition of cardiac biomarkers or the development of polygenic risk scores. Recent data on cardiac biomarkers support N-terminal pro-B-type natriuretic peptide, growth differentiation factor-15 and cardiac troponins as the most promising biomarkers for refining cardiovascular risk estimates in both females and males ( Welsh et al. 2023 ). However, further research is needed to evaluate the clinical impact of using risk assessment tools that include cardiac biomarkers. The combination of a polygenic risk score with the QRISK2 risk estimation system has shown promising results, identifying a greater proportion of individuals who went on to have a major cardiovascular event with the combined system compared to QRISK2 alone ( Samani et al. 2024 ). This additive approach appeared to perform particularly well in younger individuals, with a strong relative increase in those aged between 40 and 54 years, and was similarly beneficial for both females and males. With the introduction of more advanced statistical modelling, such as machine learning, it has become feasible to incorporate more complex modalities (e.g. electrocardiograms and cardiac imaging) and large quantities of routinely collected healthcare data into cardiovascular risk estimation systems ( Adedinsewo et al. 2022 ). Machine learning approaches offer the potential for more accurate risk estimation compared to standard statistical methods, as they can analyse large, complex datasets to identify patterns that may indicate a higher future risk of ASCVD. However, using data that lack information on female or minority groups, or suffer from poor-quality female-relevant healthcare data, may exacerbate existing inequalities for females. Machine learning cannot overcome these deficiencies, and the creation of ‘black box’ algorithms makes it difficult to detect biases, such as those arising from failing to account for confounders or overtraining of models ( Adedinsewo et al. 2022 ). Despite these advances in analytical methods, improving the standardisation and recording of female-specific risk factors and diagnoses in routine healthcare records remains essential for the effective prevention of ASCVD in females. Another key aspect of commonly used cardiovascular risk estimation systems is their static nature. Most clinical practice guidelines recommend a cardiovascular risk assessment in individuals who are 40 years of age or older. With advancements in digitalisation and electronic health record systems, there is a potential to shift this dogma towards a dynamic risk estimation system that can be updated over time. This would allow for multiple risk assessments throughout a female’s life, such as during healthcare interactions such as pregnancy, enabling better targeting of young women who might otherwise be missed. However, we do feel it is essential to carefully consider the health benefits and costs of any risk assessment strategy recommended in practice. The key recommendations and suggestions for next steps in cardiovascular health for females are provided in Table 4 . It is important to consider the data used to develop cardiovascular risk estimation systems. All current risk estimation systems use observational data for development. Studies that include only participants from observational cohorts often introduce selection bias, as they typically involve the healthiest individuals from the least deprived areas. Using electronic healthcare records helps reduce selection bias, and there are methods available to address the main issue of missing data. Large, general population datasets based on healthcare records can improve representativeness, ensuring sufficient inclusion of younger females, ethnic minorities and female-specific risk factors in derivation cohorts. Key recommendations for future research in cardiovascular disease in females. In addition, research practices need to be enhanced. Consistency in defining cardiovascular events, including non-ASCVD conditions, extending follow-up periods to capture risk in younger females and standardising the reporting of both absolute and relative risk will provide more robust evidence for the prevention of CVD in females. Steps are being undertaken to improve the recording of female-specific factors outside of pregnancy, such as the introduction of a reporting system for normal and abnormal uterine bleeding ( Munro et al. 2018 ). Issues with ethnicity data are being highlighted, strengthening the importance of accurate coding of self-identified ethnicity being incorporated routinely into observational studies ( Pineda-Moncusí et al. 2024 ). However, for these systems to add value, it is important to provide education to healthcare professionals and encourage consistent use of coding systems. In addition, natural language processing tools are being developed to identify uncoded data, which could enhance the use of healthcare record data, particularly in primary care. Novel techniques, such as machine learning approaches, may also be applied to incorporate additional markers such as cardiac biomarkers and multimodal imaging data. In addition to female-specific factors, the complex relationship between obesity and cardiovascular risk factors in females requires further study. Obesity is not only associated with diabetes, hypertension and abnormal lipids, but also with female-specific risk factors such as pre-eclampsia, gestational diabetes, polycystic ovary syndrome and gynaecological cancers. Furthermore, the increased cardiovascular risk associated with obesity is observed across ethnicities. With the rising prevalence of obesity, particularly among younger populations ( Sattar et al. 2023 ), advancing obesity research is crucial to reducing the burden of ASCVD in females. Finally, it is important to understand cardiovascular risk in under-researched populations, in minority groups, the transgender population and also globally in low- and middle-income countries. The global burden of cardiovascular disease is increasingly shifting to low- and middle-income countries. Though observational data from countries such as China may be used to identify risk factors in these populations ( Li et al. 2020 ), data-driven approaches used in high-income countries may not be directly applicable. Resource-limited countries may therefore need alternative strategies, and increased funding is needed to improve cardiovascular health in females in these regions.

TA, JAM and DMK conceived and designed this review. TA did the search and selected the studies for inclusion. TA drafted the manuscript and MdB, AA, NS, JAM and DMK revised it critically for important intellectual content. All authors approved the final manuscript.