(OR07-05-26) Sex-Specific Dietary Networks Predict All-Cause Mortality and Life Expectancy in Canadian Adults: A Machine Learning-Based Population Study

Objectives: Diet is a leading modifiable determinant of premature mortality, yet conventional dietary assessment approaches have limited ability to capture conditional food–food dependencies or sex differences in dietary structure. This study aimed to (1) identify overall and sex-specific dietary networks and communities in Canadian adults using machine learning (ML)-based network analysis; and (2) examine their associations with all-cause mortality and life expectancy (LE) in sex-stratified analyses.

Methods: Dietary intake was assessed using 24-hour recalls from two nationally representative Canadian Community Health Survey–Nutrition cycles: 2004 (n = 15,835; linked to health administrative databases 2004–2017) and 2015 (n = 13,557). Food–food conditional dependencies were estimated using semiparametric Gaussian copula graphical models (SGCGM) with graphical LASSO regularization to build overall and sex-specific networks. Dietary communities were detected using the Louvain algorithm. We derived individual-level dietary community scores and assessed their associations with mortality using multivariable Cox proportional hazards models. LE was estimated using abridged life tables.

Results: SGCGM-derived sex-specific dietary networks revealed distinct communities: women had unique communities centered on whole fruit and eggs, whereas men had fast food and solid fat hubs. The overall network identified three major communities: Vegetable-Rich (VR), High-Sugary Beverage and Low Fruit (HSBLF), and High-Fat Breakfast (HFB). Higher VR scores were associated with 59% lower all-cause mortality in women (HR = 0.41; 95% CI: 0.26–0.63) and 48% lower in men (HR = 0.52; 95% CI: 0.33–0.81). Higher HSBLF scores were associated with 39% higher mortality in men (HR = 1.39; 95% CI: 1.03–1.88), with no significant association in women. LE gains at age 45 ranged 1.0–8.3 years in women and 0.8–6.1 years in men with favorable dietary community scores.

Conclusions: ML-based network analysis identified dietary communities with distinct sex-specific structures, and community scores were differentially associated with mortality and LE in women and men. These data-driven scores may support sex-tailored healthy eating strategies and help inform targeted clinical guidance.

Funding Sources: Canadian Institute for Health Research