Machine-learning-based cost prediction models for inpatients with mental disorders in China

Area of Science:

• Health Economics
• Medical Informatics
• Psychiatry

Background:

• Rising prevalence of mental disorders increases healthcare expenditures.
• Accurate prediction of inpatient costs is essential for health insurance reimbursement.
• Per-diem payment models necessitate advanced methods for predicting average daily hospitalization costs (ADHC).

Purpose of the Study:

• To develop and evaluate machine learning models for predicting ADHC in mental health inpatients.
• To identify key factors influencing ADHC to refine insurance reimbursement strategies.
• To explore the utility of advanced algorithms in managing healthcare costs for mental disorders.

Main Methods:

• Utilized data from 5070 hospitalized mental disorder patients in Jinhua, China.
• Employed six machine learning algorithms to predict ADHC.
• Evaluated model performance using 5-fold cross-validation and bootstrap methods.

Main Results:

• The random forest (RF) model achieved the best performance (R-squared = 0.6417).
• Top factors influencing ADHC included medical institution level, age, functional, and cognitive classifications.
• Models with the top 11 factors showed performance comparable to using all variables.

Conclusions:

• Machine learning, especially the RF algorithm, significantly improves ADHC prediction accuracy for mental health patients.
• Findings support the establishment of more equitable insurance payment standards.
• The study aids in optimizing resource allocation within clinical practice for mental health care.