Prediction of hospital expenses for stroke patients under DRG/ DIP reform based on XGBoost integrated algorithm

Abstract

Objective: To analyze the value of XGBoost, support vector machine and BP neural network in predicting the hospitalization
cost of stroke patients, and select the best model through comparative analysis, so as to help hospitals rationally allocate the medical
resources of the disease and make it have scientifi c basis. Methods: The IDC code from 2017 to June 2024 was selected, which corresponds
to the ICDM160-163 stroke disease code. A total of 18107 valid data after excluding outliers were searched, and SPSS21.0 software was
used to conduct the research with the help of Excel database. The hospitalization cost of stroke patients was analyzed and predicted. Three
major models of XGBoost, support vector machine and BP neural network were established with the help of Python, and the optimal model
was selected through comparison and comparison, so as to accurately predict the hospitalization cost of the patient. Through in-depth study,
it was found that the average annual hospitalization cost reached 32079 yuan. The R-square and MAPE of the three models were 0.896
and 0.196, respectively. 0.721, 0.211; 0.824, 0.218. The above data show that among the three models, XGBoost has the highest prediction
accuracy. Conclusion: Compared with BP neural network, XGBoost and support vector machine have obvious advantages in the prediction
related research, and have higher estimation accuracy and reliability. The prediction of the cost, so that the relevant managers have a decision
basis, not only is conducive to ensuring the quality of medical care, but also can guide them to take the initiative to control the cost, in order
to reasonably regulate the medical behavior, greatly improve the utilization rate of hospital resources.