At-Home Deaths and Regional Factors in Japan
Clinical Snapshot
PICO Framework
| P — Population | Municipalities across Japan (municipal-level aggregate data), encompassing the general population at end-of-life |
| I — Intervention | Exposure: regional factors including geographic variables (latitude, longitude), population-related variables (population density, population change rate), medical resources (number of clinics providing end-of-life care), economic factors, and causes of death |
| C — Comparator | Municipalities with lower prevalence of at-home deaths compared to those with higher prevalence; no formal comparator group — this is a cross-sectional ecological analysis |
| O — Outcomes | Primary outcome: prevalence of at-home deaths at the municipal level in Japan; secondary: ranked importance of regional factors associated with at-home death prevalence using SHAP values from an XGBoost regression model |
Bottom Line
This Japanese ecological study used XGBoost machine learning with SHAP explainability to rank regional factors associated with municipal-level at-home death prevalence (median 12.7%). Geographic variables — latitude and longitude — emerged as the most important predictors, followed by clinic-based end-of-life care provision, population density, and population change rate. Higher at-home death rates were concentrated in major metropolitan areas. The study's primary value lies in its policy-informing potential: it identifies clinic-based end-of-life care as the most modifiable predictor, suggesting that expanding such services in geographically underserved municipalities could increase at-home death rates in line with patient preferences. However, the ecological design is a fundamental limitation — municipal-level associations cannot be extrapolated to individual patients (ecological fallacy), causal inference is not possible, and individual-level confounders remain uncontrolled. The absence of confidence intervals and model performance metrics limits statistical appraisal. For Australian clinicians and policymakers, the methodology offers a transferable framework for analysing geographic determinants of place of death using national open data, with direct relevance to rural and remote palliative care planning under the National Palliative Care Strategy.
Key Findings
P Value: Not reported
Effect Size: SHAP value rankings indicate: latitude (rank 1), number of clinics providing end-of-life care (rank 2), longitude (rank 3), population density (rank 4), population change rate (rank 5). Quantitative effect sizes in interpretable units not reported.
Primary Outcome: Prevalence of at-home deaths across Japanese municipalities; median prevalence was 12.7%. Higher prevalence observed in large metropolitan areas (Tokyo, Osaka, Nagoya).
Nnt Or Sensitivity: Not applicable — ecological cross-sectional study. Model performance metrics (R², RMSE, AUC) not reported in the abstract. No NNT, sensitivity, or specificity estimates provided.
Confidence Interval: Not reported
Clinical Application
The findings suggest that geographic and population-related factors — many of which are not directly modifiable by clinicians — are the strongest predictors of at-home death prevalence. The most actionable finding is the importance of clinic-based end-of-life care provision (rank 2), which is a modifiable healthcare system variable. Increasing the density of clinics providing end-of-life care in underserved regions is a feasible policy lever. Geographic factors (latitude, longitude) likely serve as proxies for unmeasured cultural, infrastructural, or climatic variables that require further investigation before targeted interventions can be designed. Direct application to Australia is limited. However, the study's methodology — using national open data to identify regional determinants of at-home death — is highly relevant to Australian health policy. Australia similarly faces a gap between public preference for home death and actual rates (approximately 14% of Australians die at home, compared to expressed preferences of 50-70%). The RACGP supports home-based palliative care, and the Australian Government funds the Palliative Care Quality Network and community palliative care services. The PBS funds medications for symptom management in home-based palliative care. The TGA has approved relevant palliative medications. Australian equivalents of the modifiable factor identified (clinic-based end-of-life care) include GP-led palliative care, Silver Chain, and community nursing services. The finding that geographic remoteness (proxied by latitude/longitude in Japan) is a key predictor resonates strongly with Australia's rural and remote health disparities. The RACGP's Standards for General Practices and the National Palliative Care Strategy 2018 both emphasise equitable access to end-of-life care across geographic regions. Health planners, policymakers, and community medicine specialists working at the municipal or regional level in Japan. Not directly applicable to individual clinical decision-making. Potentially informative for any healthcare system seeking to understand geographic determinants of place of death.
Abstract
In Japan, at-home deaths are not always popular, yet many people hope to spend their end-of-life at home, which makes this a socio-medical issue. Regional factors are associated with at-home deaths, but their relative importance has not been known. The present study investigated the relationship between at-home deaths and regional factors and ranked their importance at the municipal level in Japan. We used national open data on at-home deaths, population, medical resources, causes of death, economic factors and geographic factors as well as built the prediction model for at-home deaths using Extreme Gradient Boosting (XGBoost) regression. The Shapley Additive Explanations (SHAP) value was applied to rank the importance of regional factors and explain the XGBoost model. As a result, the median prevalence of at-home deaths across municipalities was 12.7%. A higher prevalence of at-home deaths was observed in large metropolitan areas such as Tokyo, Osaka, and Nagoya. SHAP values revealed that latitude was the most important factor in rank, followed by the number of clinics providing end-of-life care, longitude, population density, and population change rate. In conclusion, geographic and population-related factors were ranked highly among regional factors associated with at-home deaths. These findings may be informative for community planning to realize at-home deaths nationwide.
References
- 1.Shimizu, N., & Kotani, K. (2026). At-home deaths and regional factors in Japan. The Tohoku Journal of Experimental Medicine. Advance online publication. https://doi.org/10.1620/tjem.2025.J094
Related Research
Scandinavian journal of pain
Multimodal deep learning fusion for automatic pain detection in cancer patients
26 July 2026
Current treatment options in oncology
Telemedicine and Virtual Reality Technologies in Pain Management for Cancer Patients: A Structured Narrative Review
11 July 2026
Journal of medical Internet research
Attitudes and Needs of Health Care Providers Toward Artificial Intelligence-Assisted Pediatric Palliative Care: Mixed Methods Study
5 July 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service