SCORE2 cardiovascular risk score for cognitive impairment in women aged 40 years and above: a pragmatic triage signal in primary care — a single-center family medicine feasibility study with machine learning support

Abstract

Background Screening-defined cognitive impairment (CI; MMSE ≤23) in midlife women is under-recognized in primary care. SCORE2 is a 10-year cardiovascular disease risk score already routinely calculated in many primary-care settings. We evaluated whether SCORE2 provides a clinically useful triage signal for CI; secondarily, a parsimonious machine learning (ML) model was developed from routinely-collected data to support the SCORE2 signal. Methods In this single-center, cross-sectional feasibility study, 89 women aged 40–68 years (52 hypertensive, 37 normotensive controls) were analyzed. CI was defined as a Mini-Mental State Examination (MMSE) total score ≤23 (validated Turkish cut-off). The association between SCORE2 (high-risk-region coefficients) and CI was assessed by χ2 for trend, Spearman correlation, logistic regression, and ROC analysis. A parsimonious 5-variable ML model (age, BMI, systolic blood pressure, HT duration, folate) was evaluated by 10-times-repeated 5-fold cross-validation with bootstrap optimism correction, calibration, and Decision Curve Analysis. Reporting followed TRIPOD+AI 2024. Results SCORE2 showed a clear dose–response association with CI: prevalence rose from 21.2% (Low) to 50.0% (Moderate), 62.5% (High), and 60.0% (Very High) (χ2 trend p = 0.006; Spearman ρ = 0.37, p < 0.001). Each 5% absolute SCORE2 increase conferred OR 3.65 (95% CI 1.67–7.98, p = 0.001); High vs. Low OR 6.21 (1.85–20.86, p = 0.003). SCORE2 alone discriminated CI with AUC 0.742 (0.62–0.85). Overall CI prevalence was 36.0% (32/89): 53.8% in hypertensives vs. 10.8% in controls (relative risk 4.98; NNS = 2.3). Prevalence rose monotonically across cardiometabolic strata (normotensive→Stage 2 HT: 11.1% → 60.6%; normal-BMI → obese: 13.8% → 59.0%; cumulative score 0 → 4: 7.7% → 66.7%; all p ≤ 0.002). The supportive parsimonious ML model achieved AUC 0.804 ± 0.103 (optimism-corrected 0.808; Brier 0.176; Hosmer-Lemeshow p = 0.397); incremental discrimination over SCORE2 was modest (ΔAUC +0.062). A bedside nomogram derived from the parsimonious model is provided as an exploratory tool pending external validation. Conclusion SCORE2 — already routinely calculated in primary care — provides a clinically useful triage signal for cognitive impairment at no additional resource cost; women at high cardiovascular risk may benefit from formal cognitive testing. The parsimonious ML model mechanistically supports the SCORE2 signal. As a single-center feasibility study, multi-center prospective external validation is required before clinical deployment.