Prognostic Clinical Changes During the Last 3 Months of Life Among Nursing-home Residents: An Explainable Machine-Learning Study

OBJECTIVES: Timely recognition of the terminal phase among nursing-home residents is essential to align care goals. However, how dynamic changes during this period relate to mortality beyond baseline factors remains unclear. We aimed to examine baseline values and temporal changes in clinical and daily life conditions during the last 3 months of life. DESIGN: Retrospective cohort study using explainable machine learning with repeated stratified crossvalidation. SETTING AND PARTICIPANTS: Five nursing homes in the Kanto region of Japan. Included were 87 residents who died between August 2022 and June 2024 (cases) and 278 surviving controls (total n = 365). METHODS: We analyzed longitudinal data on clinical status, vital signs, physical, cognitive, and oral function, nutritional status, and bowel movements from routine electronic care records of 365 matched decedents and surviving controls. Extreme gradient boosting and SHapley Additive exPlanations (SHAP) values were used to describe how baseline indicators and temporal changes were associated with all-cause mortality. RESULTS: Models incorporating monthly change indicators from the last 3 months and weekly indicators from the last 4 weeks showed substantially better discrimination than the baseline-only model. With dynamic indicators included, short-term changes in vital signs, oral intake, and bowel movements had the largest SHAP contributions, whereas activities of daily living, body mass index, and background clinical characteristics contributed mainly through baseline levels measured 3 months before death. Time-specific SHAP analyses indicated a 2-phase structure: bowel-movement changes 2 to 3 months prior were more strongly related to mortality, whereas rapid deterioration in oral intake and vital signs became dominant from the final month to week. CONCLUSIONS AND IMPLICATIONS: Mortality was associated with long-term frailty and short-term physiological decline. Characterizing these monthly and weekly patterns using routine assessment systems may support earlier recognition of end-of-life transition and inform care planning without increasing staff burden.

Authors

Institutions

Publication Details

Journal
Journal of the American Medical Directors Association
Published
2026-11-12
DOI
https://doi.org/10.1016/j.jamda.2026.106514
Primary Topic
Palliative Care and End-of-Life Issues
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Prognostic Clinical Changes During the Last 3 Months of Life Among Nursing-home Residents: An Explainable Machine-Learning Study

Minoru Etoh, Noriko Yamamoto‐Mitani, Keiichi Shimatani, Sakiko Fukui et al.
Journal of the American Medical Directors Association
Palliative Care and End-of-Life Issues
article

Prognostic Clinical Changes During the Last 3 Months of Life Among Nursing-home Residents: An Explainable Machine-Learning Study

Minoru Etoh, Noriko Yamamoto‐Mitani, Keiichi Shimatani, Sakiko Fukui, Maiko Noguchi‐Watanabe, Jun'ichi Kotoku, Kazuhiro Yoshiuchi
article en

Abstract

OBJECTIVES: Timely recognition of the terminal phase among nursing-home residents is essential to align care goals. However, how dynamic changes during this period relate to mortality beyond baseline factors remains unclear. We aimed to examine baseline values and temporal changes in clinical and daily life conditions during the last 3 months of life. DESIGN: Retrospective cohort study using explainable machine learning with repeated stratified crossvalidation. SETTING AND PARTICIPANTS: Five nursing homes in the Kanto region of Japan. Included were 87 residents who died between August 2022 and June 2024 (cases) and 278 surviving controls (total n = 365). METHODS: We analyzed longitudinal data on clinical status, vital signs, physical, cognitive, and oral function, nutritional status, and bowel movements from routine electronic care records of 365 matched decedents and surviving controls. Extreme gradient boosting and SHapley Additive exPlanations (SHAP) values were used to describe how baseline indicators and temporal changes were associated with all-cause mortality. RESULTS: Models incorporating monthly change indicators from the last 3 months and weekly indicators from the last 4 weeks showed substantially better discrimination than the baseline-only model. With dynamic indicators included, short-term changes in vital signs, oral intake, and bowel movements had the largest SHAP contributions, whereas activities of daily living, body mass index, and background clinical characteristics contributed mainly through baseline levels measured 3 months before death. Time-specific SHAP analyses indicated a 2-phase structure: bowel-movement changes 2 to 3 months prior were more strongly related to mortality, whereas rapid deterioration in oral intake and vital signs became dominant from the final month to week. CONCLUSIONS AND IMPLICATIONS: Mortality was associated with long-term frailty and short-term physiological decline. Characterizing these monthly and weekly patterns using routine assessment systems may support earlier recognition of end-of-life transition and inform care planning without increasing staff burden.

Journal of the American Medical Directors AssociationVol. 27(12)
Institute of Science Tokyo (JP), Institute for Open and Transdisciplinary Research Initiatives (JP), The University of Tokyo (JP), Teikyo University (JP), The University of Osaka (JP)
Tokyo Institute of Technology, Japan Society for the Promotion of Science
Good health and well-being
Palliative Care and End-of-Life Issues
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon