Predictive patterns of cesarean section utilisation under Indonesia’s National Health Insurance following tariff reform: a machine learning analysis of hospital claims

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Objectives To examine the predictive contributions of clinical, non-clinical and policy-related factors to cesarean section (C-section) utilisation under Indonesia’s National Health Insurance (JKN) and to assess whether the predictive relationships differed between the periods before and after the 2023 Indonesia Case-Based Groups (INA-CBG) tariff reform. We developed and temporally evaluated machine learning models to determine whether non-clinical and tariff-related variables provided additional predictive information beyond clinical risk factors. Design Retrospective observational study using Random Forest and Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression models applied to national hospital claims data from 2022 and 2023. Models were developed separately by year and evaluated within and across years to assess discrimination, calibration, predictor selection and temporal transportability. Setting Under JKN, coverage is limited to medically indicated C-sections, reimbursed through the INA-CBG payment system. In 2023, INA-CBG tariffs were recalibrated nationwide, increasing payment levels and introducing regional payment reclassification. Participants A total of 39 049 delivery episodes involving 38 535 women with recorded mode of delivery in JKN-affiliated referral hospitals between January 2022 and December 2023. Primary and secondary outcome measures The primary outcome was mode of delivery (C-section vs vaginal birth). Model discrimination was evaluated using the area under the receiver operating characteristic curve, with bootstrap CIs. Calibration was evaluated using decile-based calibration plots. Secondary analytical measures included Random Forest variable-importance rankings, predictors retained by LASSO logistic regression and differences in model performance across within-year and cross-year evaluations. Results C-section rates remained high and stable at approximately 66% across the study period. Models demonstrated consistent discriminatory performance across years. Clinical risk factors remained the strongest predictors of C-section use. However, non-clinical characteristics—including subsidised insurance status, inpatient class, referral pathways and private facility ownership—gained greater predictive prominence following the 2023 tariff reform. Cross-year evaluation showed reduced discrimination when models were applied to data from the alternate year. Conclusions Clinical risk remained the dominant predictor of C-section utilisation under JKN, while non-clinical, health-system and tariff-related variables provided additional predictive information. Cross-year analyses indicated modest temporal differences in predictive relationships and limited model transportability across the two study years. Future studies should use longitudinal and quasi-experimental designs incorporating procedure-specific tariff changes to evaluate causal mechanisms.