Science

Ayesha Z. Sherzai, Emily L. Cauble, Emma S. Spielfogel, Joshua Zebadiah Willey, Yian Gu, Nikolaos Scarmeas, James Vincent Lacey, Dean Sherzai, Mitchell S.V. Elkind, and Sophia S. Wang

Background and Objectives
Mediterranean diet (MeDi) has been linked to lower incidence of cardiovascular and neurodegenerative diseases, and overall mortality, in several prospective studies. There are limited data, however, regarding the relationship between MeDi and stroke subtypes. We hypothesized that MeDi would be associated with a lower incidence of total, ischemic, and hemorrhagic stroke.

Methods
The California Teachers Study is a prospective cohort study that comprises 133,477 women who were educators and administrators enrolled in 1995–1996 and followed since. We identified incident strokes using linked California state hospitalization data and national death records from 1996 to 2020. The MeDi adherence score (range 0–9, higher score indicating better adherence) was calculated based on participants’ response to the Block food frequency questionnaire at study baseline. Multivariable Cox proportional hazard models adjusted for sociodemographic, lifestyle, and vascular risk variables were constructed to assess the association (hazard ratios and 95% CIs, HR 95% CI) between MeDi score and risk of stroke and its subtypes.

Results
A total of 105,614 participants were eligible and included in the final analytic cohort (mean age 52.5 ± 13.8 years). Over the follow-up period (average follow-up time was 20.5 years), there were 4,083 incident stroke events (3,358 ischemic; 725 hemorrhagic). In fully adjusted models for all stroke, ischemic, and hemorrhagic subtypes, there was a lower risk of stroke among those with MeDi scores of 6–9 compared with those with scores of 0–2 (all stroke HR 0.82, 95% CI 0.74–0.92; ischemic HR 0.84, 95% CI 0.75–0.95; hemorrhagic HR 0.75, 95% CI 0.58–0.97).

Discussion
Adherence to the MeDi is associated with lower risk of total, ischemic, and hemorrhagic stroke among women. Potential study limitations include recall bias, misclassification bias, and residual confounding, which would bias our results to the null.