Data-Driven Discovery of Plant-Based Dietary Subtypes and Their Association with Cancer, Cardiovascular diseases, and Type 2 diabetes in Postmenopausal Women: Insights from the Women's Health Initiative

Abstract

Background: Plant-based diets have been associated with lower risks of chronic diseases; however, not all plant-based diets confer equal health benefits. Data-driven methods provide an opportunity to empirically identify distinct plant-based dietary subtypes and examine their associations with health outcomes.

Objective: To categorize subtypes of plant-based dietary patterns using a data-driven approach and evaluate their associations with cardiovascular disease (CVD), type 2 diabetes (T2D), and cancer incidence among postmenopausal women in the Women's Health Initiative (WHI).

Design: Prospective cohort study using principal component analysis (PCA) and k-means clustering to empirically derive dietary subtypes.

Participants/setting: A total of 823 postmenopausal women from the WHI Observational Study who were healthy at baseline and reported low meat or poultry consumption during screening (1993 - 1998) were included and followed prospectively for incident chronic disease outcomes during WHI follow-up and extension periods (2023). A stratified comparison group of 823 higher-meat consumers also drawn from WHI Observational Study, was constructed using stratified random sampling on dietary pattern position, and key demographic variables.

Main outcome measures: Incident cases of CVD, T2D, and any type of cancer (except nonmelanoma skin cancer), verified through adjudicated clinical outcomes.

Statistical analyses performed: Cox proportional hazards models estimated associations with CVD, T2D, and cancer outcomes, adjusting for sociodemographic, lifestyle, and clinical factors.

Results: Three dietary clusters were identified: (1) Lower Meat; Fruit and Vegetable, (2) Lower Meat; Dairy and Solid Fat, (3) Lower Meat; Fish and Meat (classified as lower-meat based on screening but reporting relatively higher fish and moderate meat intake on the FFQ). No statistically significant associations were observed between any lower-meat dietary pattern and overall cancer incidence. Compared to higher-meat consumers, the Fish and Meat cluster was associated with a significant 67% lower risk of T2D (HR = 0.33, 95% CI: 0.14-0.81, p = 0.02). The Dairy and Solid Fat and Fruit and Vegetable clusters did not reach statistical significance for T2D risk (HR = 0.77, 95% CI: 0.34-1.75 and HR = 1.10, 95% CI: 0.46-2.61, respectively). For CVD, the Dairy and Solid Fat cluster was associated with a statistically significant higher risk (HR = 2.63, 95% CI: 1.05-6.56, p = 0.04), while results for the Fruit and Vegetable cluster did not reach statistical significance (HR = 0.47, 95% CI: 0.13-1.73, p = 0.26).

Conclusion: This combined PCA-clustering approach revealed meaningful heterogeneity within lower-meat dietary patterns, offering a novel framework for characterizing plant-based diets and improving precision in future diet-disease investigations.

Keywords: Plant-based diets; cancer; cardiovascular disease; diabetes; k-means clustering.