Predicting non-participants in a national health check-up programme using machine learning

GreyBay Institute senior researchers contributed to the paper Using machine-learning approaches to predict non-participation in a nationwide general health check-up scheme. The study addresses a core public-health problem in Japan's national health check-up and health-guidance system: how to identify people most likely not to attend screening so that outreach resources can be targeted more efficiently.

The analysis used a local-government database covering 7,290 adults aged 40-74 who had participated in at least one general health check-up between 2012 and 2015. The team built four predictive models based on XGBoost, random forest, support vector machines, and logistic regression, and compared them with a simple heuristic rule that assumed people who attended in one year were more likely to attend again the next year.

Machine-learning models clearly outperformed the heuristic approach in identifying future non-participants. XGBoost performed best with an AUC of 0.829 (95% CI: 0.806-0.853), compared with 0.821 for random forest, 0.812 for support vector machines, 0.816 for logistic regression, and only 0.683 for the heuristic baseline.

These findings show that data-driven models can identify likely non-participants much more accurately than rule-of-thumb targeting based only on prior attendance. That gives local governments and public-health agencies a more precise way to select intervention targets and allocate follow-up effort.

The paper also sheds light on which factors help predict attendance behaviour, creating a methodological foundation for more tailored health-promotion policy. Beyond health checks, the same framework could be extended to other public-health programmes that depend on sustained participation.

Overall, the study illustrates how machine learning can support public-health management and service optimisation, while also reflecting GreyBay's continuing interest in the intersection of data science and health policy. Predictive modelling offers a practical route toward broader and fairer coverage of preventive services.

Publication:
Using machine-learning approaches to predict non-participation in a nationwide general health check-up scheme.
Akihiro Shimoda, Daisuke Ichikawa, Hiroshi Oyama.
Computer Methods and Programs in Biomedicine, 2018.
DOI: 10.1016/j.cmpb.2018.05.032