Random-Forest Analysis of Factors in Hospitalization Costs Due to Physical Inactivity-Related Circulatory Diseases in Brazil

Physical inactivity represents one of the main modifiable risk factors for diverse chronic diseases globally, generating substantial economic burden in national health systems. Estimating hospitalization costs of diseases linked to physical inactivity entails the identification of factors influencing lifestyle choices to guide preventive strategies at population level. This study investigates factors associated with hospitalization costs due to physical inactivity-related circulatory diseases (PICD) in Brazil from 2008 to 2019. The investigation proposes adaptation of the Andersen–Newman framework for evidence-based decision-making processes in public health policy. Data on public sector costs of hospitalizations due to PICD, health infrastructure, and demographic, economic, and health characteristics for 5570 Brazilian municipalities between 2008 and 2019 were obtained from Brazilian government datasets. Random forest analysis and linear regression models were applied to the unbalanced panel data, based on a temporally ordered split for training and hyperparameter tuning with data from 2008 to 2017, and final testing of machine learning models with data between 2018 and 2019. The random forest model showed that 70.03% of variance in hospital-level PICD costs were explained by demographic, economic, health, and infrastructure characteristics of the Brazilian municipalities. Robust results from regression models using municipal fixed-effects specification indicated that PICD hospitalization costs were positively associated with cases of diabetes (β = 0.224; p < 0.001) and negatively associated with primary healthcare coverage (β = −0.057; p < 0.01). The findings indicate that municipalities showing higher proportion of female population (predisposing factor), public healthcare expenditures (enabling factor), and diabetes cases (need factor) may benefit from considering preventive health policies based on lifestyle-change strategies aimed at promoting physical activity. The adaptation of the Andersen–Newman framework into the observational–ecological study was suitable to assess predisposing, enabling, and need factors contributing to PICD hospitalization costs at population level. The study provides insights for health systems sustainable management through evidence-based decision-making processes in public policy, considering cost-escalation scenarios due to physical inactivity-related diseases.