Female-Specific, Psychosocial Factors Better Predict Stroke Risk in Younger Women

Adding female-specific and psychosocial factors to traditional stroke prediction models may better identify women younger than age 50 years who are at risk for stroke, according to study findings published in Neurology.

Current prediction models do not use female-specific or psychosocial factors to identify risk for stroke in women younger than age 50 years.

To assess the benefit of adding these female-specific and psychosocial factors into stroke prediction models, researchers in the Netherlands conducted a population-based cohort study, obtaining data from the STIZON electronic health record (EHR) database between January 2007 and December 2020. The researchers included 409,026 women between the ages of 20 and 49 years, each of whom was seen for a minimum of 1 year during this time frame at a Dutch general practice that used STIZON.

Of these 409,026 women, stroke occurred in 2751 (0.67%) women prior to age 50 years at an incidence rate of 6.9 per 10,000 person-years (95% CI, 6.6-7.2). These strokes included both ischemic and hemorrhagic subtypes as well as fatal and nonfatal stroke outcomes.

The researchers subdivided the women into 3 age groups: 20-29, 30-39, and 40-49 years. 

The researchers applied current prediction models based on traditional cardiovascular risk factors for stroke, such as age, smoking status (either current or former tobacco user), hyperlipidemia, hypertension, and diabetes mellitus. They obtained this information using ICD-9 and ICD-10 diagnosis codes, medication prescription codes, and other relevant EHR documentation. Most of the women (>80%) in the study were missing biomarker data on serum cholesterol, blood pressure, and blood glucose, so the researchers were unable to use these measurements in their calculations.

First, the researchers assessed prediction model performance using both model discrimination (C-statistic) and calibration curve slope at 10 years of follow-up for each age group — both for the traditional prediction models and their updated prediction model that included female-specific and psychosocial risk factors.

For reference, a C-statistic value of 0.5 indicates that the model is no better at predicting outcomes than random chance, while scores over 0.7 indicate good models and scores over 0.8 indicate strong models.

The traditional prediction models demonstrated poor to moderate performance across all 3 age groups of women younger than age 50 years. The researchers calculated C-statistic scores for the traditional prediction models of 0.617, 0.615, and 0.585 for women aged 20-29, 30-39, and 40-49 years, respectively.

In contrast, the calibration curves for the traditional prediction models were good, resulting in a value of 0.949, 0.974, and 0.984 for women in their 20s, 30s, and 40s, respectively.

Next, the researchers developed a prediction model that accounted for the following female-specific factors derived from the literature:

• Migraine
• Gestational diabetes
• Preeclampsia
• Preterm birth (O vs 1 or more)
• Miscarriage (0 vs 1 or more)
• Stillbirth (0 vs 1 or more)
• Menstrual irregularity or primary ovarian insufficiency
• Female infertility
• History of hysterectomy
• Poor fetal growth or small for gestational age designations for the woman’s neonate
• Complications during birth (postpartum hemorrhage, intrapartum hemorrhage, umbilical cord complications)
• Use of hormone replacement therapy
• Combined hormonal contraceptive use

Psychosocial factors that the researchers included in their prediction model consisted of the following:

• Socioeconomic status derived from postal codes, occupation of inhabitants, education level, and income
• History of depression or psychotic disorders per ICD-9 or ICD-10 diagnosis codes and prescription medication codes

Adding these female-specific and psychosocial factors moderately improved model discrimination performance. In particular, the C-statistic values for the ages 30-39 and ages 40-49 groups increased by 0.019 and 0.029, respectively. Compared with reference models, these C-statistic improvements accounted for a 16.5% and 34.1% difference in discrimination performance. No significant changes were observed to the calibration curves with the addition of these factors.

“The addition of female-specific and psychosocial risk factors to traditional cardiovascular predictors improves the discriminatory performance of prediction models for women under age 50,” the researchers wrote. They concluded, “Our newly developed stroke risk age tool can help discuss stroke risk in clinical practice.”

There are several study limitations that warrant mention, ranging from quality-related problems with EHR-derived data (especially related to underreporting of clinical conditions) to incomplete registration of noncardiovascular deaths outside of the hospital setting in primary care EHRs.