Predicting multidimensional quality of life among autistic people and people with intellectual disability in Saudi Arabia: An explainable machine-learning study.

BACKGROUND: Quality of life (QoL) among autistic people with or without co-occurring intellectual disability (ID) and people with ID without autism is multidimensional and shaped by interacting individual and contextual factors. METHODS: A multi-region cross-sectional study included 275 adults aged 18-65 years with autism, ID, or co-occurring autism and ID recruited through disability-service, vocational rehabilitation, adult support, and community networks across Riyadh, Jeddah, and Dammam, Saudi Arabia. Eligibility required a documented diagnosis established by a qualified healthcare professional. QoL was assessed using the Arabic WHOQOL-BREF across four domains. Multivariable regression examined adjusted associations with 18 prespecified predictors. Random Forest, XGBoost, and support vector regression were evaluated using five-fold cross-validation within the training set and an untouched 20% held-out test set. SHAP was used to interpret the best-performing tree-based model for each domain, and K-means clustering explored multidimensional QoL profiles. RESULTS: Adjusted associations varied across QoL domains, with disability severity and family support showing the most consistent relationships. Machine-learning models provided stronger held-out prediction than linear regression, with performance differing by domain: Random Forest performed best for Physical Health (R² =.62) and Social Relationships (R² =.50), whereas XGBoost performed best for Psychological Health (R² =.68) and Environment (R² =.58). SHAP analyses revealed distinct domain-specific predictive patterns, with disability severity ranking highest for Physical Health, family support for Psychological Health and Social Relationships, and household income for Environment. Exploratory clustering further identified four multidimensional QoL profiles: High Well-Being (26.2%), Environmental Challenges (32.0%), Psychological Strain (24.0%), and Social Isolation (17.8%). CONCLUSIONS: QoL heterogeneity extended beyond diagnostic status and differed across physical, psychological, social, and environmental dimensions. Integrating regression, explainable machine learning, and person-centered profiling revealed complementary patterns, highlighting disability severity, family support, socioeconomic resources, and geographic context as salient to multidimensional QoL and service planning.