Analysis of factors associated with inpatient reimbursement outcomes under the universal coverage scheme in Thailand

Abstract

Hospitals under Thailand’s Universal Coverage Scheme (UC) face financial challenges due to discrepancies between Diagnosis-Related Group (DRG) reimbursements and actual treatment costs. This study aimed to identify factors associated with hospital financial outcomes and evaluate the performance of Machine Learning (ML) models in predicting profit or loss for Nakhon Pathom Hospital. A retrospective analysis of 67,115 inpatient records (fiscal years 2023–2024) compared Random Forest and Extreme Gradient Boosting (XGBoost) algorithms. The Random Forest model demonstrated the best performance (Accuracy 84.5%, Receiver Operating Characteristic Area Under Curve (ROC AUC) 0.788). Length of Stay (LOS) was the most influential factor, followed by drug and non-drug costs. The model demonstrated strong capability in identifying loss cases. (True Negative Rate 95.8%). These findings indicate that Machine Learning, particularly the Random Forest algorithm, can serve as an effective decision-support tool for hospital financial management, enabling proactive risk mitigation and improved resource utilization within Thailand’s national reimbursement system.

References

Khiaocharoen et al., "Cost by Diagnosis Related Group: Outputs from the First Phase Cost per Disease Project," HISPA Compendium, vol. 1, no. 8, pp. 113–129, 2023. doi: 10.14456/hispa.2023.8

งานประกันสุขภาพ โรงพยาบาลโชคชัย, “ศูนย์จัดเก็บราย ได้คุณภาพ,” โรงพยาบาลโชคชัย, นครราชสีมา, ประเทศไทย, 2565. [ออนไลน์]. เข้าถึงได้จาก: https://www.uckkpho.com/wpcontent/uploads/2022/05/07.ศูนย์จัดเก็บรายได้ มีคุณภาพ.pdf [เข้าถึงเมื่อ: 28 ก.ย. 2568].

Kulkarni, S. S. Ambekar, and M. Hudnurkar, "Predicting the inpatient hospital cost using a machine learning approach," International Journal of Innovation Science, vol. 13, no. 1, pp. 87–104, 2021, doi: 10.1108/IIS-09-2020-0175

Thongpeth, A. Lim, A. Wongpairin, T. Thongpeth, and S. Chaimontree, "Comparison of linear, penalized linear and machine learning models predicting hospital visit costs from chronic disease in Thailand," Informatics in Medicine Unlocked, vol. 26, Art. no. 100769, 2021, doi: 10.1016/j.imu.2021.100769

Siwapraprapakorn, "Data analytic and predictive model for analysis of causes for a loss from diagnosis related groups (DRGs) payment of pediatric inpatient at Sunpasitthiprasong hospital," Journal of the Thai Medical Informatics Association, vol. 1, pp. 12–20, 2021.

Langenberger, T. Schulte, and O. Groene, "The application of machine learning to predict high-cost patients: A performance-comparison of different models using healthcare claims data," PLoS ONE, vol. 18, no. 1, p. e0279540, Jan. 2023, doi: 10.1371/journal.pone.0279540

Stone, R. Zwiggelaar, P. Jones, and N. Mac Parthaláin, "A systematic review of the prediction of hospital length of stay: Towards a unified framework," PLOS Digital Health, vol. 1, no. 4, p. e0000017, Apr. 2022, doi: 10.1371/journal.pdig.0000017

Jain, M. Singh, A. R. Rao, and R. Garg, "Predicting hospital length of stay using machine learning on a large open health dataset," BMC Health Services Research, vol. 24, Art. no. 860, 2024, doi: 10.1186/s12913-024-11238-y

Deniz, A. Şengül, Y. Aydemir, J. Ç. Emre, and M. H. Özhan, "Clinical factors and comorbidities affecting the cost of hospital-treated COPD," International Journal of COPD, vol. 11, pp. 3023–3030, Dec. 2016, doi: 10.2147/COPD.S121637

Haidar, R. Vazquez, and G. Medic, "Impact of surgical complications on hospital costs and revenues: retrospective database study of Medicare claims," Journal of Comparative Effectiveness Research, Art. no. e230080, Jun. 2023, doi: 10.57264/ cer-2023-0080

M. Sowjanya and O. Mrudula, "Effective treatment of imbalanced datasets in health care using modified SMOTE coupled with stacked deep learning algorithms," Applied Nanoscience, vol. 13, pp. 1829–1840, 2023, doi: 10.1007/s13204-021-02063-4