Updated Review Targets Gaps in Cardiovascular Risk Prediction for Women
Cardiovascular risk assessment remains uneven for women, even as cardiovascular disease continues to drive morbidity and mortality worldwide. This PLOS ONE protocol sets out to update the evidence base on prediction models for women, with a specific focus on whether female-specific risk factors meaningfully improve risk estimation beyond the variables embedded in traditional scores. The rationale is clinically familiar: women experience sex-related differences in symptoms, pathophysiology, prognosis, and post-event outcomes, yet many widely used tools still treat risk as if it behaves the same across sexes.
The review is designed to capture primary studies published from July 27, 2017 through May 2026, extending the earlier Baart et al. synthesis. Investigators will search MEDLINE and EMBASE for development studies, validation studies, and incremental value analyses in women or in models that explicitly include sex as a predictor or effect modifier. The protocol is registered in PROSPERO under CRD420261420766 and follows PRISMA-P guidance.
Women with a history of pregnancy complications may move through routine follow-up with a substantial but underrecognized cardiovascular burden.
Female-Specific Predictors Move to the Center of the Search Strategy
The review framework reaches beyond standard population models and explicitly includes female-specific predictors that often fall outside conventional cardiovascular calculators. Those variables include early menopause, premature menarche, polycystic ovary syndrome, pre-eclampsia, gestational diabetes, gestational hypertension, preterm delivery, fetal growth restriction, endometriosis, and perinatal depression. This is not a narrow academic exercise. These exposures may identify vascular injury, endothelial dysfunction, metabolic risk, and inflammatory burden years before a coronary event, creating an opportunity to intervene during windows when women are still connected to care, particularly after pregnancy.
Eligibility criteria are broad enough to reflect real-world prediction science. Studies will qualify if they report on at least one multivariable model, score, or tool estimating future arterial cardiovascular events in women, including coronary heart disease, myocardial infarction, stroke, cardiovascular death, or congestive heart failure. Female-only derivation cohorts, sex-inclusive models that incorporate sex as a predictor or effect modifier, external validation studies in female-only or sex-stratified samples, and incremental value papers all fall within scope. Statistical models, machine learning approaches, and artificial intelligence–based methods are all eligible if they meet the remaining criteria.
What Will Be Counted, Compared, and Excluded
The protocol is careful about what will not enter the synthesis. Studies focused exclusively on venous disease will be excluded, as will cross-sectional validation papers and models developed in or validated only within disease-specific cohorts such as diabetes, chronic kidney disease, or HIV. The one major exception is obstetric and female-specific populations, including cohorts defined by hypertensive disorders of pregnancy, which remain eligible because they directly inform the cardiovascular risk profile of women.
The review also restricts inclusion to English- or French-language original research. Letters, conference abstracts, and conference presentations will be left out. Reference lists of eligible studies will be screened for additional reports, but no extra databases, trial registries, or author contact plans are built into the search strategy. That choice keeps the update aligned with the earlier review while preserving continuity in the database base and search architecture.
Model Performance Will Be Judged Beyond Discrimination Alone
Data extraction will go well beyond whether a model exists. The team plans to record sample size, number of outcome events, study design, population characteristics, geography, outcome definition, prediction horizon, modeling method, number and type of predictors, missing data handling, and internal or external validation procedures. Performance measures will include discrimination, calibration, and reclassification. Clinical usability will also be captured through variables such as the number of predictors, whether the equation is available, and whether an online calculator exists.
This matters because a model can look appealing on paper while failing at the bedside. In cardiovascular prevention, discrimination tells only part of the story; calibration determines whether the risk estimate matches the patient in front of the clinician. A poorly calibrated score can push low-risk women into overtreatment or leave higher-risk women without the intensity of prevention they need. For female-specific populations, where event rates, baseline hazards, and exposures differ from male-dominant derivation cohorts, calibration may be as clinically important as c-statistics.
PROBAST+AI Will Be Used to Judge Bias in Modern Prediction Studies
Risk of bias assessment will use PROBAST+AI, the prediction-model tool adapted for studies that may involve artificial intelligence or more traditional statistical learning. Two reviewers will work independently, with disagreements resolved by group consensus. That approach is especially relevant in a field where model development papers can differ sharply in variable selection, outcome ascertainment, validation strategy, and transparency around the final model equation.
The protocol also specifies that findings will be synthesized descriptively, with quantitative pooling reserved for situations where at least three external validation studies examine the same model using comparable outcomes, horizons, and performance measures. Even then, meta-analysis is not automatic. Random-effects pooling will be considered for c-statistics and calibration measures, with heterogeneity assessed using I2 and τ2 and, when possible, a 95% prediction interval. The plan to transform c-statistics to the logit scale before pooling is a technical choice aimed at stabilizing estimates across studies that may differ in population mix and event frequency.
A Long-Running Prevention Problem Hinges on Better Risk Recognition
The clinical problem driving this review is straightforward: if the risk model underestimates women, prevention starts late and often too weakly. Adverse pregnancy outcomes can function as early vascular warning signs, and early menopause may signal an altered hormonal and cardiometabolic trajectory. When those risk enhancers are invisible to the calculator, clinicians may miss the chance to intensify blood pressure control, lipid management, glycemic surveillance, and lifestyle intervention during a period when risk can still be modified.
The protocol is already in motion. Title and abstract screening has begun, and the investigators anticipate completing screening, extraction, and bias assessment by January 2027, with synthesis and manuscript preparation targeted for May 2027. No funding supported the protocol, and no competing interests were declared. No datasets were generated in the current study, because this is a review protocol rather than a completed analysis.
The next evidence review will not answer whether every female-specific predictor belongs in every calculator. It will determine which models have actually earned a place in women’s cardiovascular prevention, and which still overpromise on paper while leaving clinical risk insufficiently measured.