Analysis of medical costs and two-model prediction for patients with severe mental disorders in Gansu Province, China

Front Public Health. 2026 Feb 23;14:1715899. doi: 10.3389/fpubh.2026.1715899. eCollection 2026.

ABSTRACT

BACKGROUND: The economic burden of severe psychiatric disorders presents a major global public health challenge, particularly in regions with underdeveloped healthcare systems. Analysing medical costs is essential for optimizing resource allocation and improving patient outcomes.

AIMS: This study provides the first comprehensive analysis of medical expenditures for severe mental disorders in Gansu Province, China, and compares the predictive performance of the Bayesian Regression Model based on Gaussian Processes with Random Forest regression for outpatient and inpatient costs.

METHODS: This retrospective analysis utilized data from the Gansu Provincial Healthcare Security Administration, covering 284,447 outpatient and 8,962 inpatient cases diagnosed between 2021 and 2023. Data distribution was assessed using the Kolmogorov-Smirnov test, and group comparisons were conducted using chi-square and Mann-Whitney U tests. Medical costs were predicted using the Bayesian Regression Model based on Gaussian Processes and Random Forest regression models.

RESULTS: Between 2021 and 2023, the average costs per outpatient visit and inpatient admission were US$77.29 and US$922.86, respectively. The median outpatient cost declined annually from US$65.98 in 2021 to US$46.84 in 2023, whereas the median inpatient cost in 2023 exceeded that of 2021 and 2022 (p < 0.001). In the prediction of outpatient costs, the Bayesian regression model based on Gaussian processes performed slightly better than the Random Forest model; however, the predictive ability of both models was quite limited, with a very low proportion of cost variance explained (Bayesian regression: R ² = 0.3977, 95% CI: 0.03918-0.4022; Random Forest: R ² = 0.0620, 95% CI: 0.0586-0.0653). Random Forest demonstrated markedly superior performance in predicting inpatient costs (R ² = 0.7741, 95% CI: 0.7013-0.7982), significantly outperforming Bayesian regression (R ² = 0.3405, 95% CI 0.3802-0.4098).

CONCLUSION: Outpatient costs continued to decline, while inpatient costs increased significantly. In predicting outpatient costs, the Bayesian regression model based on Gaussian processes performed relatively well but its overall predictive capability remained limited; the Random Forest model demonstrated superior performance in predicting inpatient costs. The study suggests that in underdeveloped regions, data-driven cost analysis should be prioritized to optimize the allocation of mental health resources and alleviate the economic burden.

PMID:41810298 | PMC:PMC12969060 | DOI:10.3389/fpubh.2026.1715899