Machine Learning in Palliative Care: Scoping Review of Applications

Palliative care is increasingly recognized as essential for an aging population and rising life-limiting illnesses. Machine learning (ML) has been widely applied in this field, primarily for prognostication. However, recent literature suggests broader applications that may enhance patient-centered care and optimize system-level processes.

This study aimed to map and summarize the evolving landscape of ML applications in palliative care through a scoping review, identifying how studies extend beyond mortality prediction into new domains, while assessing explainability, equity, and implementation readiness.

We conducted a scoping review following the Arksey and O'Malley framework and PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines. Six databases (MEDLINE, PsycINFO, Embase, CINAHL, Scopus, and Web of Science) were searched from inception to April 15, 2021, with an update through February 9, 2026. Included studies were peer-reviewed primary studies applying ML to palliative care contexts. Each study was coded for explainable AI (XAI) methods, equity considerations, and implementation readiness. Two reviewers independently screened and extracted data. Synthesis combined descriptive statistics and inductive thematic analysis. Consistent with scoping review methodology, no formal risk-of-bias assessment was performed.

We included 121 studies (2015-2026) spanning 24 countries, with 69.4% (84/121) published from 2021 onward. The United States contributed the largest share (66/121, 54.5%), followed by Japan, Taiwan, and China (22/121, 18.2%). Cancer was the most commonly studied population (52/121, 43%). Supervised classification was the most common approach (84/121, 69.4%), followed by natural language processing and text mining (16/121, 13.2%). Six application domains were identified: mortality and survival prediction (51/121, 42.1%), health care use (25/121, 20.7%), symptom assessment and phenotyping (20/121, 16.5%), communication and natural language processing (16/121, 13.2%), clinical decision support and care quality (6/121, 5%), and other (3/121, 2.5%). Approximately half of the studies (61/121, 50.4%) used at least one XAI technique, most commonly feature importance rankings and SHAP (Shapley Additive Explanations) values. Among the 121 studies, equity in model performance was fully addressed in only 8 (6.6%) studies, partially in 8 (6.6%) studies, and not addressed in 105 (86.8%) studies. Two-thirds of studies (80/121, 66.1%) remained at the proof-of-concept stage, while 16.5% (20/121) achieved external validation and 17.4% (21/121) reached prospective deployment or clinical integration.

ML applications in palliative care are expanding beyond prognostication toward patient-centered uses, including symptom management, clinical decision support, and resource planning. The persistent gap in equity reporting (105/121, 86.8% did not report equity considerations) signals that the field risks developing tools that may not perform equitably across diverse populations. While half of studies now use XAI techniques, fewer than 1 in 5 studies (21/121, 17.4%) have reached clinical integration. Bridging this translational gap requires systematic attention to implementation science, equity auditing, and explainability reporting.