2023 Volume 50 Issue 5 Pages 439-446
Objective: Lifestyle disorders are chronic diseases that require long-term primary prevention measures. In the short term, changes in health checkup results are often small and subjective symptoms are rare, making it difficult to create an opportunity to engage in primary prevention. If the risk estimation score of lifestyle disorders (hereinafter referred to as "JPM score") can be reported to examinees immediately at the time of health checkups in a simple and cost-effective way, the same can be used for specific health guidance and other purposes.
Methods: In this study, 51,771 people who underwent health examination in 2011 and 2021 were included in the creation of the "JPM score" prediction formula. The prediction formulae were generated by a logistic regression model (LRM). Machine learning models (MLMs) were used for performance comparisons.
Results: The 3-year averages of accuracy score, area under the curve, and Matthews correlation coefficient for LRM were 77.2%, 0.823, 0.407, respectively, and for MLMs were 77.2–78.4%, 0.796–0.842, 0.406–0.451, respectively.
Conclusions: No major differences were observed among models. LRM, which is computationally less expensive and easier to provide a basis for generating advisory comments, can be used to implement the "JPM score" in our core healthcare system.
