Therapeutic Research

Abstract

Background：The number of cases of heart failure（HF）is increasing every year. In recent years, studies have reported on the use of machine learning to evaluate healthcare costs.　　Objective：This study aimed to use machine learning to identify factors that may influence the high healthcare costs of treating patients with HF. 　　Methods：We performed machine learning with medical insurance claims data to explore factors that have high generalizability and can be adjusted to reduce medical costs. The target dataset comprised patients with HF included in the Japanese insurance claims database Cross Fact® in the period from January 2015 to December 2020. 　　Results：Data of 15,043 patients were extracted；medical costs were high in 3484 （22.7％）of these patients and low in 11,559（77.3％）. Data on approximately 300 disease‒and procedure‒related factors were subjected to machine learning, and 25 factors of variable importance were extracted. Logistic regression analysis identified systemic atrophies primarily affecting the central nervous system and congenital malformations of the urinary system as being associated with high medical costs in patients with HF. The factor other forms of heart disease showed a trend toward being associated with lower medical costs. 　　Conclusion：A machine learning analysis of claims data identified factors associated with medical costs in patients with HF that are largely consistent with those found in previous studies. Further studies are needed to confirm the findings.