Objective: We aimed to identify the population in which encouraging participation in the general health check-up would be helpful using a prediction model based on a machine-learning method. A secondary analysis of data obtained from the health promotion program using a randomized controlled design, aimed at improving participation in the general health check-up, was performed.
Methods: The retrospective analysis was conducted using data from a health promotion program in the Fukuoka branch of Japan Health Insurance Association, Japan, between November 2015 and March 2016. Subjects were extracted from dependents (family members) of insured persons aged 40-74 years who had participated in general health check-up at least once in the past five years (2010-2014). Subjects were divided into two groups; the intervention group received a printed reminder saying “you are due to general health check-up” through mail, while the control group received nothing. The participation rates of both groups for each participation probability group (participation probability was calculated by the prediction model) were assessed after 4 month follow-up.
Results: The numbers in the intervention group and in the control group were 1,911 and 3,294, respectively. Regarding the prediction model, the AUC value for test data was 0.668 (95%CI: 0.635–0.701). With regard to the effectiveness of the intervention for each probability group, there was a significant difference between the groups only for the moderate participation probability groups as follows: 30-39% (P=0.005), 40-49% (P=0.003), 50-59% (P=0.004) and 60-69% (P=0.039). Conclusion: The intervention with printed reminder was effective for improving participation of general health check-up among the group with moderate participation probability. The targeting using the results of prediction model was useful for identifying appropriate intervention targets.
Conclusions: More studies are needed to assess the cost and benefits of adopting a system like this and then the appropriate actions could be taken.