ORIGINAL RESEARCH article
Front. Public Health
Sec. Environmental Health and Exposome
Forecasting Daily Bathtub-Drowning Mortality in Japan: A Comparative Analysis of Statistical, Machine Learning, and Deep Learning Approaches
Provisionally accepted
- Nara Medical University, Kashihara, Japan
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Background Japan reports the highest global mortality rate from drowning among older adults, predominantly owing to bathtub‑related incidents. Despite sustained public health interventions, this mortality has increased over several decades. Timely warnings advising older adults to avoid unsupervised bathing during high‑risk conditions may mitigate this issue; however, no nationwide forecasting model currently exists. Methods We integrated death certificate records from 1995–2020 (99,930 bathtub-drowning deaths) with meteorological, temporal, and demographic data across all 47 prefectures (446,359 prefecture‑days). Daily mortality counts were modelled using a distributed-lag non‑linear model (DLNM), extreme gradient boosting (XGBoost), and long short‑term memory network (LSTM). Data were partitioned chronologically into training (1995–2015), validation (2016–2018), and test (2019–2020) sets. Predictive accuracy was evaluated using root mean square error (RMSE) and mean absolute error (MAE), whereas feature importance was quantified via Shapley additive explanations. Results During the test period, DLNM, XGBoost, and LSTM exhibited comparable predictive performance (RMSE = 0.577, 0.574, 0.575; MAE = 0.345, 0.333, 0.347, respectively). The most important features across all models were daily mean temperature, prefectural population, and binary prefecture indicators. Restricting DLNM meteorological inputs to routinely forecasted variables—daily maximum and minimum temperatures—did not reduce predictive accuracy (RMSE = 0.577 [95% confidence interval, 0.566–0.590]; MAE = 0.344 [95% confidence interval, 0.340–0.349]). 3 Conclusions The DLNM-based framework provides a practical means of forecasting the daily bathtub-drowning deaths. Integration into routine meteorological broadcasts and mobile platforms may facilitate timely warnings, prompting older adults to avoid unsupervised bathing on high‑risk days, thereby reducing Japan's ongoing preventable bath‑related mortality.
Keywords: Bath-related deaths, Bathtub drowning, Meteorological factors, Predictive Modeling, machine learning, deep learning, older adults
Received: 29 Sep 2025; Accepted: 24 Nov 2025.
Copyright: © 2025 Tai, Obayashi, Yamagami and SAEKI. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Yoshiaki Tai, yoshiaki.t@naramed-u.ac.jp
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.