Predictors of quality of life in OCD: Machine-derived features beyond expert consensus.
Quality of life (QoL) varies substantially among individuals with obsessive-compulsive disorder (OCD), but the clinical and contextual factors most strongly associated with QoL remain unclear. A better understanding of these factors may improve assessment, treatment planning, and service organization.
3152 adults with lifetime OCD completed a validated measure of QoL along with clinical and sociodemographic measures. A 22-feature expert-informed model, based on prior scientific and clinical knowledge, was compared with data-driven models using recursive feature elimination and gradient-boosted trees. Model performance was evaluated using 5-fold cross-validation.
Data-driven models modestly outperformed the expert-informed model (RMSE 7.23 vs. 7.56; R2 0.55 vs. 0.51). Although several predictors overlapped, machine-derived models identified predictors not selected by experts, including subjective social rank, anhedonia, and inequality-adjusted national human development. Across models, depression severity, subjective social rank, and anxiety severity were the strongest predictors of QoL. Reduced data-driven models using 12-18 features retained comparable predictive performance and significantly outperformed the expert model.
Expert knowledge captured much of the variance in QoL in OCD, but data-driven approaches improved predictive accuracy and identified novel factors. QoL was most strongly predicted by depression severity, anxiety severity, and social context rather than OCD severity alone, suggesting that clinical assessment and treatment planning should routinely consider broader affective and psychosocial difficulties alongside OCD symptoms.
3152 adults with lifetime OCD completed a validated measure of QoL along with clinical and sociodemographic measures. A 22-feature expert-informed model, based on prior scientific and clinical knowledge, was compared with data-driven models using recursive feature elimination and gradient-boosted trees. Model performance was evaluated using 5-fold cross-validation.
Data-driven models modestly outperformed the expert-informed model (RMSE 7.23 vs. 7.56; R2 0.55 vs. 0.51). Although several predictors overlapped, machine-derived models identified predictors not selected by experts, including subjective social rank, anhedonia, and inequality-adjusted national human development. Across models, depression severity, subjective social rank, and anxiety severity were the strongest predictors of QoL. Reduced data-driven models using 12-18 features retained comparable predictive performance and significantly outperformed the expert model.
Expert knowledge captured much of the variance in QoL in OCD, but data-driven approaches improved predictive accuracy and identified novel factors. QoL was most strongly predicted by depression severity, anxiety severity, and social context rather than OCD severity alone, suggesting that clinical assessment and treatment planning should routinely consider broader affective and psychosocial difficulties alongside OCD symptoms.
Authors
Thorsson Thorsson, Trujillo Trujillo, Romero Romero, IrreƱo-Sotomonte IrreƱo-Sotomonte, Borda Borda, Zepeda-Burgos Zepeda-Burgos, Martinez Mallen Martinez Mallen, Moyano Moyano, Anderberg Anderberg, MuƱoz MuƱoz, Moore Moore, Frederick Frederick, Berrones Berrones, Cruz Cruz, Onyeka Onyeka, Wiese Wiese, Wagner Wagner, , , Crowley Crowley, Storch Storch, Cervin Cervin
View on Pubmed