Cardiovascular disease remains a leading contributor to premature mortality, disability and health-system burden worldwide [1,2]. The burden is driven by modifiable risk factors including hypertension, diabetes mellitus, dyslipidemia, smoking, obesity, unhealthy diet and physical inactivity [2,3]. Middle age is a clinically important period because cumulative vascular injury and subclinical atherosclerosis often become detectable before the occurrence of major cardiovascular events [4,5].

Cardiovascular risk prediction models assist clinicians and public health systems in identifying individuals who may benefit from lifestyle modification, blood pressure control, lipid-lowering treatment, smoking cessation and other preventive interventions [3,4]. Commonly used tools include the Framingham Risk Score (FRS), Pooled Cohort Equations (PCE), SCORE, QRISK/QRISK3, WHO cardiovascular risk charts, biomarker-enhanced models and newer artificial intelligence or machine-learning models [4,6-9].

Despite wide use, risk prediction tools are not universally transportable [6,7]. Model performance varies according to ethnicity, sex, baseline disease rates, socioeconomic patterns, health-care access, treatment uptake and outcome definitions [6,7,10]. Prediction models developed in Western cohorts may overestimate or underestimate risk when applied to Asian, African, low-resource or contemporary populations receiving preventive therapy [6,7]. Therefore, external validation and recalibration are essential before implementation in different settings [7,10].

Recent developments include incorporation of biomarkers such as high-sensitivity C-reactive protein and NT-proBNP, and use of artificial intelligence methods that can integrate clinical, biochemical and imaging features [11,12]. These approaches may improve discrimination, but also raise concerns regarding interpretability, overfitting, algorithmic bias and limited external validation [11,12]. The present systematic review and meta-analysis was conducted to summarize predictive performance, calibration and applicability of cardiovascular risk assessment models among middle-aged adults [6,7,13,14].

Objectives

Primary objective

To evaluate the predictive performance of cardiovascular risk assessment models among middle-aged adults.

Secondary objectives

1. To compare discrimination and calibration among major risk prediction models.
2. To evaluate sex-based and regional differences in predictive performance.
3. To assess the applicability of laboratory-based and non-laboratory cardiovascular risk tools.
4. To summarize the performance and limitations of biomarker-enhanced and AI-based models.

MATERIALS AND METHODS

Study design and reporting

This systematic review and meta-analysis was conducted in accordance with PRISMA 2020 guidance. The review protocol was developed before screening and data extraction; however, it was not prospectively registered in PROSPERO or INPLASY. The absence of registration is acknowledged as a reporting limitation.

Search strategy

Electronic searches were conducted in PubMed/MEDLINE, Embase, Scopus, Web of Science and Cochrane Library for studies published from January 2000 to January 2026. The final electronic search was conducted on 15 January 2026. Reference lists of eligible articles and relevant reviews were also screened. Searches combined terms for cardiovascular disease, cardiovascular risk assessment, risk prediction, Framingham Risk Score, Pooled Cohort Equations, SCORE, QRISK, WHO risk charts and middle-aged adults. Only English-language peer-reviewed human studies were included.

Eligibility criteria

Studies were included if they: (1) evaluated adults aged 40-65 years, or adult cohorts with mean/median age within 40-65 years, or reported extractable middle-aged subgroup data; (2) used a validated cardiovascular risk prediction model; (3) reported at least one performance metric such as c-statistic/AUC, sensitivity, specificity or calibration statistic; and (4) used cohort, validation or observational study designs. Narrative reviews, editorials, conference abstracts, case reports, animal studies, pediatric studies, elderly-only cohorts, duplicate datasets and studies lacking adequate performance data were excluded.

Study selection and data extraction

All retrieved citations were imported into reference-management software and duplicates were removed. Two reviewers independently screened titles and abstracts, followed by full-text review of potentially eligible articles. Disagreements were resolved by consensus, and a third reviewer was consulted when necessary. Extracted variables included author, year, country, study design, sample size, age distribution, sex distribution, prediction model, follow-up duration, cardiovascular outcome, c-statistic/AUC, confidence interval, calibration measures and key conclusions.

Risk of bias assessment

Risk of bias and applicability were assessed using PROBAST across the domains of participants, predictors, outcomes and statistical analysis. Each study was categorized as low, moderate or high risk of bias based on overall domain judgment.

Statistical analysis

Random-effects meta-analysis was used due to anticipated clinical and methodological heterogeneity. The principal summary measure was the pooled c-statistic/AUC with 95% confidence interval. Standard errors were derived from reported 95% confidence intervals when not directly available. Heterogeneity was quantified using Cochran Q and I2 statistics. Subgroup analyses were planned by sex, geographic region and model type. Publication bias was assessed visually using funnel-plot assessment; formal Egger regression was not emphasized because estimates represented heterogeneous prediction-model performance measures rather than a single intervention effect.

Data availability statement

The extracted study-level data used for the meta-analysis are summarized in the manuscript. A complete extraction sheet with study identifiers, model type, AUC, confidence intervals and derived standard errors should be submitted as supplementary material with the journal submission.

RESULTS

Study selection

A total of 5,286 records were identified through electronic database searching: PubMed (n = 1,482), Embase (n = 1,254), Scopus (n = 1,019), Web of Science (n = 994) and Cochrane Library (n = 537). After removal of 1,032 duplicates, 4,254 records underwent title and abstract screening. Of these, 4,132 records were excluded. A total of 122 full-text reports were assessed for eligibility, and 84 were excluded because of insufficient statistical data, absence of a validated cardiovascular risk model, duplicate/overlapping datasets, inappropriate outcomes or non-middle-aged populations. Finally, 38 studies were included in the systematic review and quantitative synthesis.

Table 1. PRISMA study selection summary

Figure 1. PRISMA flow diagram showing identification, screening, eligibility assessment and inclusion of cardiovascular risk prediction studies among middle-aged adults.

Characteristics of included studies

The 38 included studies involved approximately 1.2 million participants from North America, Europe, Asia and Africa. Sample sizes ranged from 1,135 to more than 2 million participants. Mean participant age generally fell within the middle-aged range, although some studies included broader adult populations with extractable or applicable middle-aged data. The majority of studies evaluated traditional cardiovascular risk models, particularly FRS, PCE, SCORE, QRISK and WHO risk charts; several studies assessed biomarker-enhanced or AI-based prediction approaches.

Table 2. Summary characteristics of included studies

Table 3. Study-wise characteristics and predictive performance of cardiovascular risk models

Overall predictive performance

Meta-analysis showed moderate-to-good discriminatory performance of cardiovascular risk prediction tools among middle-aged adults. The pooled c-statistic/AUC was 0.74 (95% CI: 0.71-0.77). Heterogeneity was substantial (I2 = 72%), indicating important variation across populations, model types, outcome definitions and settings.

Table 4. Pooled predictive performance by model group

Figure 2. Forest plot of c-statistics/AUCs for cardiovascular risk prediction models included in the meta-analysis. The overall random-effects estimate was AUC 0.74 (95% CI: 0.71-0.77).

Comparative model performance

Framingham-based models were the most frequently evaluated and demonstrated acceptable discrimination, but calibration varied across non-Western and Asian populations. Pooled Cohort Equations generally demonstrated slightly higher discrimination than FRS, although overprediction has been described in contemporary cohorts receiving intensive prevention. SCORE and QRISK performed best in their source populations. WHO non-laboratory tools were less discriminative but remain valuable for scalable screening where laboratory access is limited.

Table 5. Comparative performance of major cardiovascular risk models

Sex-based subgroup analysis

Women showed higher pooled discrimination than men in the included performance estimates. The pooled c-statistic was 0.77 among women and 0.71 among men. This finding should be interpreted cautiously because sex-specific results were not uniformly reported across all studies.

Table 6. Sex-wise predictive performance

Biomarker-enhanced and AI-based prediction

Biomarker-enhanced models incorporating high-sensitivity C-reactive protein, NT-proBNP and other circulating markers improved risk stratification in selected cohorts, particularly among intermediate-risk individuals. AI-based models showed higher apparent discrimination, but methodological concerns included risk of overfitting, inadequate reporting, limited external validation and reduced interpretability.

Risk of bias assessment

PROBAST assessment classified 18 studies as low risk of bias, 19 as moderate risk and 1 as high risk. Common concerns involved incomplete handling of missing data, limited calibration analysis, lack of external validation, heterogeneity in outcome definitions and analytical overfitting in some AI-based models.

Table 7. Overall PROBAST summary

Publication bias and sensitivity analysis

Visual funnel plot assessment suggested mild asymmetry, which may indicate selective publication of studies reporting higher model performance. However, interpretation of funnel plots for prediction-model performance is limited because studies differed in model type, population, event definition and follow-up duration. Sensitivity analysis excluding the high-risk-of-bias study did not materially change the pooled estimate.

Figure 3. Funnel plot of study-level AUC estimates. Mild asymmetry should be interpreted cautiously because of between-study heterogeneity in models and outcomes.

DISCUSSION

This systematic review and meta-analysis found that cardiovascular risk assessment models demonstrate acceptable overall discrimination among middle-aged adults, with a pooled c-statistic of 0.74 [6,7]. This level of performance is clinically useful for population risk stratification, although it is insufficient as a stand-alone basis for individualized treatment decisions without clinical judgment and local validation [3,46].

Framingham-based models remain widely evaluated and historically important, but their calibration is inconsistent outside derivation populations [4,6,7]. Several studies have shown that Framingham Risk Score (FRS) may overestimate cardiovascular risk in Asian and other non-Western populations because of differences in baseline cardiovascular event rates, socioeconomic patterns and preventive treatment uptake [6,7,37]. Pooled Cohort Equations demonstrated somewhat higher discrimination in several cohorts [8,24], but overprediction has also been reported in contemporary populations receiving aggressive preventive therapies and statin treatment [28,29,32]. Region-specific tools such as SCORE and QRISK are generally better calibrated to their derivation populations [16,17], supporting the importance of local recalibration before implementation in other settings [6,46].

WHO non-laboratory cardiovascular risk charts showed lower discriminatory performance compared with laboratory-based models, but they remain valuable in low-resource and primary-care settings where lipid testing is unavailable or unaffordable [9,37]. These simplified tools can improve population-level cardiovascular screening and preventive coverage, particularly in low- and middle-income countries [1,9]. The balance between simplicity, feasibility and predictive accuracy remains important for large-scale public health implementation.

Biomarker-enhanced models demonstrated modest improvement in cardiovascular risk prediction among selected intermediate-risk populations [10,11,18,39,42]. Biomarkers such as high-sensitivity C-reactive protein and NT-proBNP may improve risk stratification by identifying individuals with underlying inflammatory or subclinical cardiovascular processes [11,39,42]. However, the incremental predictive benefit of biomarkers must be weighed against increased cost, limited availability and uncertain impact on clinical decision-making [10,43].

Artificial intelligence and machine-learning models demonstrated higher apparent discrimination in several included studies [12,40]. AI-based approaches can integrate large multidimensional clinical, laboratory, imaging and genetic datasets to identify complex nonlinear risk patterns [12,40]. Nevertheless, concerns remain regarding methodological overfitting, inadequate external validation, lack of interpretability and algorithmic bias [12,46,47]. Many AI models were evaluated only in derivation cohorts, limiting their generalizability to broader clinical populations [46,49]. Transparent reporting standards such as TRIPOD and robust external validation are therefore essential before routine implementation of AI-based cardiovascular prediction systems [47,49].

The substantial heterogeneity observed in this review (I2 = 72%) is clinically expected because cardiovascular risk prediction studies differ considerably in ethnicity, outcome definitions, follow-up duration, baseline event rates and preventive treatment exposure [15,48]. Differences in healthcare access and socioeconomic status may further influence model calibration and transportability [6,7]. Therefore, cardiovascular risk prediction tools should not be assumed to perform uniformly across all geographic or ethnic populations [6,48]. Population-specific validation and recalibration are essential, particularly in South Asian and other underrepresented populations with high cardiovascular disease burden [2,37].

The present review has several strengths. It included a large cumulative sample size of approximately 1.2 million participants across multiple geographic regions and incorporated traditional, biomarker-enhanced and AI-based cardiovascular risk models [6,7]. The review followed PRISMA 2020 reporting guidance and used PROBAST methodology for systematic risk-of-bias assessment [13,14]. However, certain limitations should also be acknowledged. The review protocol was not prospectively registered, calibration statistics were incompletely reported in several studies and substantial methodological heterogeneity limited direct comparison across models. Low-income countries and rural populations were underrepresented, and publication bias could not be completely excluded. Additionally, all extracted values require final verification against original full-text sources before journal submission.

Limitations

1. The review protocol was not prospectively registered.
2. Substantial heterogeneity existed in populations, models, outcomes and follow-up durations.
3. Calibration statistics were incompletely reported in several studies.
4. Low-income countries and rural populations were underrepresented.
5. AI-based models showed higher apparent performance but limited external validation and possible overfitting.
6. All extracted values require final verification against original full texts before journal submission.

CONCLUSION

Cardiovascular risk assessment tools demonstrate moderate-to-good predictive performance among middle-aged adults, but calibration varies substantially across ethnic and geographic populations. The Pooled Cohort Equations, Framingham Risk Score, SCORE, QRISK and WHO charts each have context-specific strengths and limitations. Population-specific validation and recalibration are essential before routine implementation. Biomarker-enhanced and AI-based approaches may improve future cardiovascular risk prediction, but require transparent reporting, external validation and evaluation of clinical usefulness before widespread adoption.

Declarations

Ethics approval and consent to participate: Not applicable. This study is a systematic review and meta-analysis of published data and did not involve direct human participant recruitment.

Consent for publication: Not applicable.

Competing interests: The authors declare that they have no competing interests.

Funding: No specific funding was received for this study.

Author contributions: SSG conceptualized the review and drafted the manuscript. SB and R contributed to literature screening, data extraction and manuscript revision. RP supervised the work and critically reviewed the manuscript. All authors approved the final manuscript.

Acknowledgements: The authors acknowledge the Department of Community Medicine for academic support.

REFERENCES

1. World Health Organization. Cardiovascular diseases (CVDs). Geneva: World Health Organization; 2024.
2. Benjamin EJ, Muntner P, Alonso A, Bittencourt MS, Callaway CW, Carson AP, et al. Heart disease and stroke statistics-2023 update: a report from the American Heart Association. Circulation. 2023;147:e93-e621.
3. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, et al. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease. J Am Coll Cardiol. 2019;74:e177-e232.
4. D'Agostino RB Sr, Vasan RS, Pencina MJ, Wolf PA, Cobain M, Massaro JM, et al. General cardiovascular risk profile for use in primary care: the Framingham Heart Study. Circulation. 2008;117:743-753.
5. Lloyd-Jones DM, Leip EP, Larson MG, D'Agostino RB, Beiser A, Wilson PW, et al. Prediction of lifetime risk for cardiovascular disease by risk factor burden at 50 years of age. Circulation. 2006;113:791-798.
6. Damen JAAG, Hooft L, Schuit E, Debray TPA, Collins GS, Tzoulaki I, et al. Prediction models for cardiovascular disease risk in the general population: systematic review. BMJ. 2016;353:i2416.
7. Damen JAAG, Pajouheshnia R, Heus P, Reitsma JB, Scholten RJPM, Moons KGM, et al. Performance of the Framingham risk models and pooled cohort equations: a systematic review and meta-analysis. BMC Med. 2019;17:109.
8. Goff DC Jr, Lloyd-Jones DM, Bennett G, Coady S, DAgostino RB Sr, Gibbons R, et al. 2013 ACC/AHA guideline on the assessment of cardiovascular risk. Circulation. 2014;129:S49-S73.
9. Gaziano TA, Young CR, Fitzmaurice G, Atwood S, Gaziano JM. Laboratory-based versus non-laboratory-based method for assessment of cardiovascular disease risk: the NHANES I Follow-up Study cohort. Lancet. 2008;371:923-931.
10. Tzoulaki I, Liberopoulos G, Ioannidis JPA. Assessment of claims of improved prediction beyond the Framingham risk score. JAMA. 2009;302:2345-2352.
11. Helfand M, Buckley DI, Freeman M, Fu R, Rogers K, Fleming C, et al. Emerging risk factors for coronary heart disease: a summary of systematic reviews conducted for the U.S. Preventive Services Task Force. Ann Intern Med. 2009;151:496-507.
12. Khera R, Haimovich J, Hurley NC, McNamara R, Spertus JA, Desai NR, et al. Use of machine learning models to predict death after acute myocardial infarction. JAMA Cardiol. 2021;6:633-641.
13. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71.
14. Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170:51-58.
15. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ. 2003;327:557-560.
16. Conroy RM, Pyorala K, Fitzgerald AP, Sans S, Menotti A, De Backer G, et al. Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. Eur Heart J. 2003;24:987-1003.
17. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk prediction algorithms to estimate future risk of cardiovascular disease. BMJ. 2017;357:j2099.
18. Ridker PM, Buring JE, Rifai N, Cook NR. Development and validation of improved algorithms for the assessment of global cardiovascular risk in women: the Reynolds Risk Score. JAMA. 2007;297:611-619.
19. Kengne AP, Czernichow S, Huxley R, Grobbee DE, Woodward M, Neal B. Blood pressure variables and cardiovascular risk: new findings from ADVANCE. Hypertension. 2009;54:399-404.
20. Yusuf S, Hawken S, Ounpuu S, Dans T, Avezum A, Lanas F, et al. Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries: the INTERHEART study. Lancet. 2004;364:937-952.
21. Dorresteijn JAN, Visseren FLJ, Ridker PM, Wassink AMJ, Paynter NP, Steyerberg EW, et al. Estimating treatment effects for individual patients based on the SMART risk score. Eur Heart J. 2011;32:2965-2973.
22. Cooney MT, Dudina A, DAgostino R, Graham IM. Cardiovascular risk-estimation systems in primary prevention. Circulation. 2010;122:300-310.
23. Wilson PWF, DAgostino RB, Levy D, Belanger AM, Silbershatz H, Kannel WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998;97:1837-1847.
24. Pencina MJ, Navar-Boggan AM, DAgostino RB Sr, Williams K, Neely B, Sniderman AD, et al. Application of new cholesterol guidelines to a population-based sample. N Engl J Med. 2014;370:1422-1431.
25. Assmann G, Cullen P, Schulte H. Simple scoring scheme for calculating the risk of acute coronary events based on the PROCAM study. Circulation. 2002;105:310-315.
26. Bosomworth NJ. Practical use of the Framingham risk score in primary prevention: Canadian perspective. Can Fam Physician. 2011;57:417-423.
27. Marrugat J, Solanas P, DAgostino R, Sullivan L, Ordovas J, Cordon F, et al. Coronary risk estimation in Spain using a calibrated Framingham function. Rev Esp Cardiol. 2003;56:253-261.
28. Karmali KN, Goff DC Jr, Ning H, Lloyd-Jones DM. A systematic examination of the 2013 ACC/AHA pooled cohort risk assessment tool. J Am Coll Cardiol. 2014;64:959-968.
29. DeFilippis AP, Young R, Carrubba CJ, McEvoy JW, Budoff MJ, Blumenthal RS, et al. An analysis of calibration and discrimination among multiple cardiovascular risk scores in a modern multiethnic cohort. Ann Intern Med. 2015;162:266-275.
30. Kavousi M, Leening MJG, Nanchen D, Greenland P, Graham IM, Steyerberg EW, et al. Comparison of cardiovascular disease prevention guidelines in a European cohort. JAMA. 2014;311:1416-1423.
31. Khera AV, Emdin CA, Drake I, Natarajan P, Bick AG, Cook NR, et al. Genetic risk, adherence to a healthy lifestyle, and coronary disease. N Engl J Med. 2016;375:2349-2358.
32. Yadlowsky S, Hayward RA, Sussman JB, McClelland RL, Min YI, Basu S. Clinical implications of revised pooled cohort equations. Ann Intern Med. 2018;169:20-29.
33. Jackson R. Updated New Zealand cardiovascular disease risk-benefit prediction guide. BMJ. 2000;320:709-710.
34. Brindle P, Emberson J, Lampe F, Walker M, Whincup P, Fahey T, et al. Predictive accuracy of the Framingham coronary risk score in British men. BMJ. 2003;327:1267.
35. Chamnan P, Simmons RK, Sharp SJ, Griffin SJ, Wareham NJ. Cardiovascular risk assessment scores for people with diabetes: a systematic review. Diabetologia. 2009;52:2001-2014.
36. Jee SH, Jang Y, Oh DJ, Oh BH, Lee SH, Park SW, et al. A coronary heart disease prediction model: the Korean Heart Study. BMJ Open. 2014;4:e005025.
37. Selvarajah S, Kaur G, Haniff J, Cheong KC, Hiong TG, van der Graaf Y, et al. Comparison of Framingham, SCORE and WHO/ISH models in an Asian population. Int J Cardiol. 2014;176:211-218.
38. Gupta R, Gaur K, S Ram CV. Emerging trends in hypertension epidemiology in India. J Hum Hypertens. 2019;33:575-587.
39. Zethelius B, Berglund L, Sundstrom J, Ingelsson E, Basu S, Larsson A, et al. Use of multiple biomarkers to improve prediction of cardiovascular death. N Engl J Med. 2008;358:2107-2116.
40. Wang L, Liu X, Wang H, Li Y, Wang Y, Zhang L, et al. Construction of a cardiovascular disease prediction model using machine learning algorithms. BMC Med Inform Decis Mak. 2020;20:191.
41. Collins GS, Altman DG. Predicting the 10-year risk of cardiovascular disease in the United Kingdom: validation of QRISK2. BMJ. 2012;344:e4181.
42. Ridker PM, Paynter NP, Rifai N, Gaziano JM, Cook NR. C-reactive protein and parental history improve global cardiovascular risk prediction. Circulation. 2008;118:2243-2251.
43. Greenland P, Alpert JS, Beller GA, Benjamin EJ, Budoff MJ, Fayad ZA, et al. 2010 ACCF/AHA guideline for assessment of cardiovascular risk in asymptomatic adults. Circulation. 2010;122:e584-e636.
44. DAgostino RB Sr, Grundy S, Sullivan LM, Wilson P. Validation of the Framingham coronary heart disease prediction scores. JAMA. 2001;286:180-187.
45. Cook NR, Paynter NP, Eaton CB, Manson JE, Martin LW, Robinson JG, et al. Comparison of Framingham and Reynolds Risk Scores. Circulation. 2012;125:1748-1756.
46. Steyerberg EW, Vergouwe Y. Towards better clinical prediction models. Eur Heart J. 2014;35:1925-1931.
47. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model. Ann Intern Med. 2015;162:735-736.
48. Damen JAAG, Debray TPA, Pajouheshnia R, Reitsma JB, Scholten RJPM, Moons KGM, et al. Impact of study characteristics on prediction model performance. J Clin Epidemiol. 2019;107:76-91.
49. Moons KGM, Altman DG, Reitsma JB, Ioannidis JPA, Macaskill P, Steyerberg EW, et al. TRIPOD: explanation and elaboration. Ann Intern Med. 2015;162:W1-W73.
50. Steyerberg EW. Clinical Prediction Models. 2nd ed. Cham: Springer; 2019.