Abstract / Summary
Abstract Background Prognostic models for metabolic dysfunction-associated steatotic liver disease (MASLD) are largely ‘one-size-fits-all’ and do not collectively account for sex differences and social determinants of health (SDoH). To address this gap, we trained and developed sex-specific machine learning models that incorporated SDoHs to examine important predictors of all-cause mortality among adults with MASLD. Methods Using the Canadian Longitudinal Study on Aging, MASLD was defined as the presence of hepatic steatosis, at least one cardiometabolic risk factor, and low levels of sex-specific alcohol consumption (females: < 140 g/week, males: < 210 g/week). Three machine learning models were examined in sex and age-specific subgroups (middle-age: 45–64 years vs. older age: $$\:\ge\:$$ 65 years). Based on the literature, expert input, and data availability, 25, clinical, sociodemographic, and lifestyle predictors capturing upstream, intermediate, or surrogate factors associated with MASLD progression and mortality were considered. The cohorts were split into 75% training and 25% testing, and 5-fold cross-validation was used for hyperparameter tuning. Evaluation was performed in the held-out test set using the Concordance index (C-Index), integrated Brier score (iBS), time-dependent area under the curve (AUC(t)), and visual calibration plots. Important predictors were described based on global importance and directionality of Shapley additive explanation values. Results Of 30,097 participants, we identified 8,429 with MASLD (35.3% female) followed for a median of 7.67 years (IQR: 6.84, 8.45), of which 631 (7.5%) died. Using a random survival forest model, all-cause mortality risk prediction demonstrated meaningful stratification across the follow-up period (Females: C-Index 0.73, iBS 0.032, mean AUC(t) 0.78; Males: C-Index 0.79, iBS 0.035, mean AUC(t) 0.82). Risk prediction profiles for females and males differed. Alongside clinical factors such as albumin and cardiometabolic multimorbidity, female risk was impacted by income, education, and lifestyle, particularly in middle-age. Comparatively, metabolic parameters such as waist circumference, blood pressure, albumin, and body mass index were important predictors for males. Conclusion Our findings describe how mortality risk prediction profiles for people with MASLD vary by sex and age. This highlights the potential value of integrating SDoHs into sex-specific prognostication, bridging social and precision medicine to enable more equitable, individualized risk prediction in MASLD.