Clinical implications of the recalibrated CHA₂DS₂-VA score for women after ischemic stroke: a prospective cohort study

Baseline demographics, cardiovascular risk factors (hypertension, diabetes mellitus, coronary artery disease, obesity, hypercholesterolemia, smoking), and stroke history were systematically recorded. Stroke severity was assessed using the National Institutes of Health Stroke Scale (NIHSS) at admission and at discharge or death. Functional outcomes were evaluated at discharge using the modified Rankin Scale (mRS). Cardiovascular risk stratification was performed using the CHA₂DS₂-VASc score at admission. To simulate the newly proposed CHA₂DS₂-VA score, one point for female sex was subtracted in women. Median scores and interquartile ranges were calculated for both CHA₂DS₂-VASc and CHA₂DS₂-VA scores in male and female stroke patients with AF. Analyses characterize post-stroke risk distribution and secondary-prevention implications; they do not constitute validation for primary-prevention performance.

Continuous variables were reported as medians with interquartile ranges (IQR) or means with standard deviations (SD), depending on data distribution. Categorical variables were presented as counts and percentages. Group comparisons were performed using the Mann–Whitney U test for continuous variables and the chi-square or Fisher’s exact test for categorical variables. Differences in stroke severity (NIHSS), functional outcome (mRS), and risk scores (CHA₂DS₂-VASc, CHA₂DS₂-VA) between sexes were evaluated. A two-sided P-value of < 0.05 was considered statistically significant. Confidence intervals (CIs) were calculated for key differences. All statistical analyses were conducted using SPSS Statistics, version 28 (IBM Corp.). Multivariable logistic regression was performed to assess the independent association between sex and severe stroke presentation, defined as NIHSS ≥ 10. Covariates included age, atrial fibrillation, diabetes mellitus, coronary artery disease, and stroke subtype (e.g., cardioembolic vs. ESUS). Adjusted odds ratios (aOR) with 95% confidence intervals (CI) were reported. Additionally, propensity score matching (PSM) was conducted to balance baseline characteristics between male and female patients with atrial fibrillation. Patients were matched 1:1 using nearest-neighbor matching without replacement on the basis of age, diabetes, and coronary artery disease. Post-matching balance was assessed using standardized mean differences. NIHSS and mRS outcomes were compared between matched pairs using Wilcoxon signed-rank and McNemar tests, respectively. To address the modest sample size and potential age confounding, we performed sensitivity analyses. Age was modeled flexibly using restricted cubic splines, and a Sex×Age interaction was tested. Robustness was further examined using overlap weighting, Firth’s bias-reduced logistic regression, and bootstrap resampling (1,000 iterations). Age-stratified analyses, standardization to a common age distribution, and E-value calculations were performed to assess the impact of residual and unmeasured confounding. Model performance with and without sex was compared using ΔAUC, reclassification metrics (IDI, NRI), and calibration measures.

Among the 714 patients included in the cohort, 403 (56.4%) were male and 311 (43.6%) female. Transient ischemic attacks (TIAs) were the only subtype with female predominance (55.1%), a difference that reached statistical significance (p = 0.02; 95% CI, 2.1 to 17.6% points). In contrast, atherosclerotic strokes occurred more frequently in men (67.3% vs. 32.7%, p < 0.001), while cryptogenic and ESUS subtypes demonstrated no significant sex differences (ESUS: 38.8% female vs. 61.2% male, p = 0.42)(Table 1).

Regarding embolic risk factors, coronary artery disease was more common among men (30.0% vs. 22.0%, p = 0.04; 95% CI, 0.6 to 15.2% points). Atrial fibrillation (25.2% vs. 20.1%) and prior stroke or TIA (27.3% vs. 23.4%) were more frequently observed in women; however, these differences did not reach statistical significance (p > 0.05 for both comparisons).

Sex-specific variation in cardiovascular risk profiles was also evident. Obesity was more prevalent in men (50.2% vs. 39.9%, p = 0.03; 95% CI, 1.1 to 19.8), as was current smoking (30.4% vs. 20.1%, p = 0.01; 95% CI, 2.9 to 17.1). Conversely, diabetes mellitus (35.1% vs. 26.5%, p = 0.04) and hypercholesterolemia (42.1% vs. 34.3%, p = 0.05) were more frequently observed in women (Table 2).

Stroke severity at presentation, assessed by the National Institutes of Health Stroke Scale (NIHSS), was significantly higher in women compared to men (median NIHSS, 12 [interquartile range, 7–16] vs. 8 [IQR, 4–13]; p < 0.01). This sex difference in stroke severity was particularly pronounced in the cardioembolic stroke subgroup. Functional outcomes at discharge showed a non-significant trend toward worse disability in women (median mRS 4 vs. 3; p = 0.08). Regarding stroke subtype distribution, the majority of AF patients were classified as cardioembolic strokes (74.5%), with no significant sex differences in subtype classification (p = 0.45). Within the ESUS subgroup, women with AF presented with more severe strokes than men (median NIHSS, 7 vs. 3; p = 0.03), although this disparity was not evident at discharge (Fig. 1).

Bild vergrößern

In the overall stroke cohort, women presented with higher stroke severity than men (median NIHSS 8 [IQR 4–12] vs. 6 [IQR 3–10]; p = 0.04, Mann–Whitney U test). This sex difference was particularly pronounced in the cardioembolic subgroup (median NIHSS 12 [IQR 7–16] vs. 8 [IQR 4–13]; p < 0.01) and also evident in ESUS (median NIHSS 7 [IQR 5–12] vs. 3 [IQR 2–7]; p = 0.03). At discharge, NIHSS scores had improved in both sexes, and the between-sex difference was no longer statistically significant.

Among the 714 patients included in the cohort, atrial fibrillation (AF) was identified in 161 patients (22.5%). Of these, 81 patients (50.3%) were female, and 80 patients (49.7%) were male. Female patients with AF were older than their male counterparts (mean age, 78 years vs. 74 years; p = 0.03). Cardiovascular risk profiles differed by sex. Diabetes mellitus was more prevalent in women with AF (40.7% vs. 32.5%), whereas coronary artery disease was more common among men (28.8% vs. 17.3%; p = 0.04).

Median CHA₂DS₂-VASc scores differed significantly between sexes. Female stroke patients exhibited a higher median CHA₂DS₂-VASc score compared to male patients (5 [IQR, 4–6] vs. 3 [IQR, 2–4]; p < 0.001). After recalculating the score by removing the sex category to derive the CHA₂DS₂-VA score, the difference persisted, although it was attenuated (median CHA₂DS₂-VA score, 4 [IQR, 3–5] in women vs. 3 [IQR, 2–4] in men; p = 0.03).

The adjusted difference in median CHA₂DS₂-VA scores between women and men was 1 point (95% CI, 0.5 to 1.5 points). These findings indicate that female stroke patients retain a higher vascular risk burden even after the sex category is removed, underscoring the importance of comprehensive risk stratification beyond simple score recalculations (Table 3).

Among the 81 female stroke patients with atrial fibrillation, 11 (13.6%) had a premorbid CHA₂DS₂-VA score of ≤ 1. By contrast, only 1 of 80 male patients (1.3%) fell into this category (p = 0.006, Fisher’s exact test). In the female low-score subgroup, 8 of 11 patients (72.7%) presented with moderate to severe stroke (NIHSS ≥ 8), compared to 48.2% (34/70) in the remaining female AF cohort (p = 0.04, Fisher’s exact test). Cardioembolic stroke was the assigned etiology in 6 of 11 cases (54.5%) among low-score women, compared to 74.3% (52/70) in other women with AF (p = 0.19).

Female sex was associated with a significantly higher likelihood of falling below the anticoagulation threshold (OR 5.78, 95% CI 1.15–29.0; p = 0.03). NIHSS at admission in this low-score subgroup had a median of 9 (IQR 6–12), compared to 7 (IQR 4–11) in the rest of the female AF cohort (p = 0.04, Mann–Whitney U test).

To account for the observed age difference between male and female stroke patients with atrial fibrillation (mean age, 78 vs. 74 years; p = 0.03), an age-independent analysis of the CHA₂DS₂-VA score was performed. For this purpose, CHA₂DS₂-VA scores were recalculated excluding the age-related points (age 65–74 years and ≥ 75 years).

After removal of age as a scoring component, female patients continued to demonstrate significantly higher median CHA₂DS₂-VA scores compared to male patients (median, 3 [IQR, 2–4] vs. 2 [IQR, 1–3]; p = 0.04). The adjusted difference in scores between women and men was 1 point (95% CI, 0.3 to 1.7 points). These findings suggest that, beyond differences in chronological age, women with atrial fibrillation who experience stroke have a higher burden of vascular risk factors than men. This persistent disparity reinforces the importance of individualized risk assessment beyond standard scoring algorithms.

To evaluate the predictive performance of the conventional CHA₂DS₂-VASc score versus the revised CHA₂DS₂-VA score, we performed receiver operating characteristic (ROC) curve analyses with severe stroke (NIHSS ≥ 10) as the binary outcome (Fig. 2). The area under the curve (AUC) for the CHA₂DS₂-VASc score was 0.74 (95% CI, 0.68–0.80), compared to 0.71 (95% CI, 0.65–0.78) for the CHA₂DS₂-VA score. The difference in AUCs was not statistically significant (DeLong test p = 0.09), suggesting comparable discriminatory ability between the two scoring systems in this cohort. Sex-stratified analyses revealed a slightly higher AUC for CHA₂DS₂-VASc in women than in men (0.75 vs. 0.71), but this difference was not significant (p = 0.12). Both scores demonstrated similar calibration across risk strata. These findings suggest that while removal of female sex from the CHA₂DS₂-VASc score does not significantly impair its overall performance in predicting stroke severity, it may underestimate the cumulative risk burden among female patients with AF, particularly those with overlapping comorbidities..

Bild vergrößern

To assess whether sex was independently associated with stroke severity, we performed multivariable logistic regression with severe stroke (NIHSS ≥ 10) as the dependent variable. After adjusting for age, atrial fibrillation, diabetes mellitus, coronary artery disease, and stroke subtype, female sex remained an independent predictor of severe stroke (adjusted odds ratio [aOR], 1.54; 95% CI, 1.03–2.29; P = 0.03). Age and the presence of atrial fibrillation were also independently associated with severe stroke (aOR for age ≥ 75 years, 1.82; 95% CI, 1.12–2.95; P = 0.01; aOR for AF, 1.46; 95% CI, 1.01–2.11; P = 0.04). To reduce confounding by baseline vascular risk profiles, we performed 1:1 propensity score matching (PSM) on 72 women and 72 men with atrial fibrillation, matched for age, diabetes, and coronary artery disease. Standardized mean differences for matched variables were < 0.1, indicating good balance.

After matching, female patients continued to present higher stroke severity at presentation, with a median NIHSS of 11 (IQR, 7–16) compared to 8 (IQR, 5–13) in male patients (P = 0.02, Wilcoxon signed-rank test). Functional outcome at discharge, assessed by mRS, showed a non-significant trend toward worse outcomes in women (mRS ≥ 4 in 56.9% of women vs. 43.1% of men; P = 0.08, McNemar test). These findings suggest that the sex disparity in stroke severity among patients with AF persists even after accounting for age and comorbid risk factors.

To address both potential age confounding and the modest sample size, we performed several robustness checks. Flexible age modelling with restricted cubic splines confirmed the association between female sex and severe stroke (aOR = 1.48, 95% CI 1.05–2.10), with no evidence for a Sex×Age interaction (p = 0.41). Near-exact age matching and overlap weighting yielded consistent results (aORs 1.52–1.55, all p < 0.05). Age-stratified analyses (65–74, 75–84, ≥ 85 years) and standardization to a common age distribution demonstrated a persistent risk difference of 7.2% (95% CI 1.5–12.9).

This study has limitations. It was conducted at a single tertiary academic center, which may restrict generalizability. Despite comprehensive assessment of comorbidities, residual confounding from unmeasured factors such as socioeconomic status, frailty, or access to outpatient care cannot be excluded. Stroke severity and functional outcomes were evaluated only at discharge, without longer-term follow-up for recurrence, mortality, or quality of life. Detailed information on anticoagulation adherence, dosing, and quality was not available, which may have influenced outcomes independently of baseline risk scores. The overall cohort, particularly the atrial fibrillation subgroup, was modest, limiting statistical power and precision. To mitigate this, we applied small-sample–robust methods—including Firth’s bias-reduced logistic regression, bootstrap resampling, and influence diagnostics—all of which yielded consistent estimates. Because women in our cohort were substantially older, we also performed extensive sensitivity analyses with flexible age modelling, stratified and standardized comparisons, and propensity-based weighting, which consistently confirmed the association between sex and stroke severity. These complementary approaches support the robustness of our findings, although replication in larger, multicenter cohorts with longitudinal follow-up is warranted. Given the modest subgroup sample and potential age imbalance, all subgroup analyses—particularly those involving CHA₂DS₂-VA ≤ 1—should be regarded as exploratory.