Optimization of case-mix for per-diem payment of mental disorders based on the E-CHAID decision tree

Background Mental disorders have imposed a substantial economic burden on patients, families, and society in China. Although per-diem payment can effectively contain healthcare expenditures, the current per-diem payment standards for mental disorders remain relatively crude and fail to reflect differences in actual resource consumption among patients with different characteristics. Therefore, this study analyzes the factors associated with inpatient cost per diem among inpatients with mental disorders to provide evidence for developing a refined case-mix–based per-diem payment system suitable for mental health services in China. Methods A total of 25,510 inpatients with mental disorders admitted to two hospitals in Chongqing between 2020 and 2024 were included. Their demographic, clinical, and cost information were collected. Univariate and multivariable linear regression analyses were performed to identify the main factors associated with inpatient cost per diem, and an E-CHAID decision tree was applied to construct a case-mix classification model for per-diem payment in mental health services. Results The multivariable linear regression results showed that sex, age, insurance type, discharge mode, hospital level, readmission plan, inclusion in clinical pathways, surgical status, comorbidity status, number of hospitalizations, length of stay, and principal diagnosis were significantly associated with inpatient cost per diem. Using length of stay, hospital level, insurance type, and number of comorbidities as splitting variables, a total of 10 case-mix groups were generated. The grouping scheme demonstrated high reliability and robustness, with a reduction in variance (RIV) of 0.43 and coefficients of variation (CVs) below 1 across all groups (0.20–0.59). In addition, the per-diem payment standards for the 10 groups ranged from 120.06 to 658.16 yuan. Conclusion Length of stay, hospital level, insurance type, and number of comorbidities are key factors for grouping inpatients with mental disorders, and the construction of a refined case-mix–based per-diem payment system using the E-CHAID decision tree is a reasonable and appropriate approach.