Exploration of the Relationships between Men's Healthy Life Expectancy in Japan and Regional Variables by Integrating Statistical Learning Methods.

A quantitative understanding of the relationship between comprehensive health levels, such as healthy life expectancy and their related factors, through a highly explanatory model is important in both health research and health policy making. In this study, we developed a regression model that combines multiple linear regression and a random forest model, exploring the relationship between men's healthy life expectancy in Japan and regional variables from open sources at the city level as an illustrative case. Optimization of node-splitting in each decision tree was based on the total mean-squared error of multiple regression models in binary-split child nodes. Variations of standardized partial regression coefficients for each city were obtained as the ensemble of multiple trees and visualized on scatter plots. By considering them, interaction terms with piecewise linear functions were exploratorily introduced into a final multiple regression model. The plots showed that the relationship between the healthy life expectancy and the explanatory variables could differ depending on the cities' characteristics. The procedure implemented here was suggested as a useful exploratory method for flexibly implementing interactions in multiple regression models while maintaining interpretability.