Exploring the application of Artificial Intelligence in palliative care and its practical, technical and ethical considerations: a scoping review

Exploring the application of Artificial Intelligence in palliative care and its practical, technical and ethical considerations: a scoping review | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Exploring the application of Artificial Intelligence in palliative care and its practical, technical and ethical considerations: a scoping review Thanarpan Peerawong, Tharin Phenwan, Sitthichok Chaichulee, Kamonrat Tangudomkit This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8924351/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract Background Palliative care improves the quality of life of patients with life-limiting conditions and their families; however, global access remains constrained by workforce shortages and late referrals. Artificial Intelligence (AI) has been proposed as a scalable solution for optimising the identification of needs, supporting clinical decision-making, and enhancing care delivery. However, real-world evidence of the application of AI in palliative care remains sparse, particularly regarding its impact on quality of life, quality of care, and associated practical, technical and ethical challenges. Methods A scoping review was conducted following the Joanna Briggs Institute methodology and the PRISMA-ScR checklist. Five databases (the ACM Digital Library, CINAHL, Cochrane Central, PubMed and Web of Science) were searched. Studies reporting the use of AI to facilitate or enhance palliative care delivery in adults were eligible. Four reviewers independently screened the records, and two reviewers extracted the data using Covidence software. A narrative synthesis was then performed. Results Fifteen studies, published between 2021 and 2025, were included. Fourteen originated from Global North settings (USA 5, Germany 2, Japan 2, Taiwan 2, UK 1, Spain 1 and Cyprus 1) and one from Iran. Conceptually, AI applications fall into three domains: (1) early identification of palliative care needs, (2) symptom assessment and management and (3) clinical decision support for care conversations. Fifteen studies (100%) reported or discussed quality of care outcomes, most commonly prognostic performance, usability and referral/conversation rates, and only two (13.3%) directly addressed quality of life. Effectiveness was consistently positive, with four randomised controlled trials demonstrating superiority over usual care in referrals, advance care planning, pain control and quality of life domains. Practical barriers were centred on workflow integration and resource demands, while technical limitations include data quality, generalisability, and interpretability. Ethical discourse is underdeveloped, with major gaps in the principles of AI governance. Conclusions AI shows potential to improve prognostic accuracy, trigger earlier involvement of palliative care specialists and support symptom management. However, this evidence is geographically skewed, methodologically immature and ethically underdeveloped. Future research must prioritise diverse global settings, patient-reported quality of life outcomes, participatory co-design and systematic ethical governance to ensure equitable implementation. Digital health technologies Artificial Intelligence Machine learning Digital tools Scoping review Figures Figure 1 Figure 2 Background Palliative care improves the quality of life of patients with life-limiting conditions and their families by addressing their physical, psychological, social and spiritual needs [ 1 ]. Early delivery of palliative care has been shown to reduce symptom burden, decrease unwanted aggressive interventions at the end of life and improve both patients’ and carers’ outcomes [ 2 , 3 ]. Despite these benefits, access remains limited worldwide, particularly due to workforce shortages, late referrals and resource constraints in both high- and low-income settings [ 4 , 5 ]. Artificial intelligence (AI) and related technologies have emerged as potential tools to address these gaps [ 6 , 7 , 8 ]. AI systems can analyse electronic health records, wearable sensor data or conversational voice data to identify unmet palliative care needs, predict prognosis, support symptom monitoring and facilitate conversations about serious illnesses [ 7 , 8 ]. However, evidence of real-world applications in palliative care remains limited, with most studies focusing on retrospective electronic health record data rather than prospective use with patients or caregivers in diverse clinical contexts. Few studies have evaluated the impact of AI on core palliative care outcomes, including patient quality of life (QoL) and quality of care (QoC). Moreover, although recent reviews have highlighted practical, technical and ethical challenges, comprehensive mapping is lacking [ 7 , 8 , 9 ]. This is particularly concerning, given the vulnerability of the population and the additional ethical considerations that arise when AI interfaces directly with human decision-making in end-of-life care [ 10 , 11 , 12 , 13 ]. With AI gaining rapid global momentum in healthcare, there is an urgent need to systematically capture the emerging evidence of its role in palliative care delivery, its effectiveness in improving QoL and QoC, and its associated practical, technical and ethical implications. Of particular concern is the risk that AI, developed predominantly in high-resource settings, may exacerbate existing inequities when applied globally. Therefore, this scoping review aims to: map existing evidence on the application of AI in palliative care in a real-world context. examine its reported effectiveness and influence on the quality of life and quality of care. identify implicit and explicit practical, technical and ethical considerations, together with proposed mitigation strategies. Review questions What types of AI have been implemented to improve palliative care delivery? Which health-related outcomes in the QoL and QoC domains have been measured or discussed? How effective are these AI applications in relation to the measured or discussed outcomes? What practical, technical and ethical considerations have been identified, and what mitigation strategies have been proposed? For this review, relevant definitions from the United Kingdom Research Integrity Office (UKRIO) were used [ 14 ] p2: AI is an umbrella term for a range of algorithm-based technologies that solve complex tasks by carrying out functions that previously required human thinking. Machine learning (ML) is a subset of AI that learns from data without being explicitly programmed. This learning can be unsupervised or human instructed/’human-in-the-loop’ (reinforcement learning)’. Natural language processing (NLP) refers to processes that enable machines to understand, generate and manipulate human language. Generative AI is an ‘AI that can create new text, images, audio, video, and code’. An example includes ChatGPT, which is ‘a large language model (LLM) designed to receive text cues or prompts (inputs) and generate outputs using NLP. Multimodal AI refers to the use of multiple AI systems to process different types of data simultaneously. Methods A scoping review was conducted following the Joanna Briggs Institute (JBI) methodology for scoping reviews and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews checklist [ 15 ]. The review protocol was registered in the Open Science Framework platform [ 16 ]. Eligibility criteria Studies were included based on inclusion and exclusion criteria (Table 1 ). The Population, Concept and Context framework was used to formulate the criteria and review questions [ 17 ]. Table 1 Inclusion and exclusion criteria Inclusion criteria Exclusion criteria Population Articles that discuss: - Adults receiving palliative care - Family carers of adults receiving palliative care - Healthcare professionals - Children Concept - Any application of AI to facilitate or enhance palliative care delivery (e.g., prognostication, symptom management, decision support, communication) - Articles not related to palliative care provision - Theoretical discussions, legal/ethical/theoretical debates without any application Context - Any context of care (acute hospitals, long-term care facilities, community) - Global – Types of Evidence - Quantitative studies - Qualitative studies - Mixed methods - Grey literature - Published in English - Reviews, editorials, protocols - Articles published in other languages Search strategy Four team members, supported by an academic librarian, searched the Association for Computing Machinery Digital Library, the Cumulative Index to Nursing and Allied Health Literature, Cochrane Central, PubMed and Web of Science databases between September and October 2025. A combination of keywords, Boolean operators and truncations was applied to maximise the retrieval of relevant studies (see Supplementary File 1). Grey literature was eligible; however, searches were limited to the databases listed above. Screening and selection The search results were uploaded to Covidence, a review management software programme for semi-automated duplicate removal. Duplicates were further removed manually. Initially, four reviewers performed a 25-title pilot screening against the inclusion and exclusion criteria and discussed any ambiguities. Subsequently, a dual-screening process was applied. Four reviewers were randomly assigned to screen the articles independently. The lead reviewer was responsible for the final decision on whether to include or exclude records in cases of discrepancy. Titles with unclear or absent abstracts were included in the full-text review. A similar process was then applied to the full-text review, starting with a pilot screening against the inclusion and exclusion criteria, followed by a dual-screening process of randomly assigned articles, with the lead reviewer resolving discrepancies. The reasons for exclusion were documented at the full-text stage. Data extraction and synthesis A pilot extraction file was created using the Covidence software. The lead reviewer then extracted the relevant data from the pilot extraction file, and the process was iterative, ensuring that the review questions were fully answered. Once finalised through team discussions, two reviewers extracted the remaining data until completion. The remaining reviewers verified the completeness of the extracted data. The extracted data included: Author(s); Year of publication; Country of origin; Aims/objectives; Population and sample size; Concept; Methodology/methods used; Context of the study; Types of AI used; AI development and testing process, if relevant; Outcome measures/discussed; Influences/effects of AI used on quality of life and quality of care; Practical considerations; Technical considerations; Ethical considerations. Study authors were not contacted for clarification or to obtain missing data. No formal critical appraisal of individual study quality was undertaken, which is consistent with the JBI scoping review guidelines [ 17 ]. Synthesis of results Due to the anticipated heterogeneity in study designs and outcomes, findings were narratively synthesised and presented in tables and figures [ 18 ]. The included studies were categorised according to their shared characteristics, and the findings were related to the review questions. Patient and public involvement There was no direct patient or public involvement in this review. Results Overall characteristics of the included studies The databases were searched through October 2025. In total, 814 records were identified. Of these, 204 duplicates were removed, and 610 titles and abstracts were screened. In total, 144 articles were included in the full-text assessment. Of these, 129 articles were excluded, leaving 15 studies for review (see Fig. 1). Figure 1. PRISMA flow diagram Study characteristics The included 15 studies were published between 2021 and 2025 and spannedmultiple countries: United states of America [ 19 , 20 , 21 , 22 , 23 ] (N = 5); Germany [ 24 , 25 ] (N = 2); Japan [ 26 , 27 ] (N = 2); Taiwan, China [ 28 , 29 ] (N = 2); Cyprus [ 30 ] (N = 1); Iran [ 31 ] (N = 1); UK [ 32 ] (N = 1), and one study conducted in six countries [ 33 ] (Brazil, Italy, Greece, Portugal, Scotland and Spain), indicating global interest in the topic. Only one study originated from the Global South (Iran) [ 31 ] and demonstrated potential epistemic inequality. Most studies were conducted in acute hospitals [ 20 , 21 , 22 , 23 , 24 , 25 , 26 , 28 , 29 , 31 , 32 ] (N = 11). This was followed by multiple settings [ 30 , 33 ] (N = 3) and community settings [ 19 , 27 ] (N = 1). None of the patients were admitted to long-term care facilities. Regarding study designs, 14 studies employed quantitative designs. One study used a quasi-experimental design [ 20 ]. Four studies were randomised controlled trials (RCTs) [ 21 , 22 , 23 , 30 ]. Two studies used a prospective cohort design [ 28 , 29 ], while one study used a retrospective cohort design [ 26 ]. Other designs included vignette-based evaluations [ 24 , 33 ], cross-sectional studies [ 27 , 32 ], device validation studies [ 25 ], observational predictive modelling studies [ 31 ], and qualitative focus group studies [ 19 ]. In terms of conditions, six studies focused on advanced cancer [ 20 , 22 , 26 , 28 , 29 , 30 ] and one study focused on any stage of cancer [ 21 ]. One study included patients with unspecified palliative conditions [ 25 ]. One study included vignettes of patients with gynaecological cancer [ 24 ], while another focused on vignettes of patients with a poor prognosis [ 33 ]. Two studies focused on multiple palliative conditions [ 19 , 23 , 27 ], one on chronic obstructive pulmonary disease (COPD) [ 31 ], and one on individuals with dementia [ 32 ]. The sample sizes of the included studies were generally small (range 3–20,506, median 112). Twelve studies primarily involved palliative care patients (with or without family carers as secondary participants) or used palliative care patient records [ 20 , 21 , 22 , 23 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 ]. Three studies focused primarily on healthcare professionals (HCPs) as the end users of AI tools [ 19 , 24 , 33 ]. The inclusion of patients, families and HCPs reflects the interdisciplinary nature of palliative care. However, a true participatory design, in which patients or caregivers are co-researchers rather than passive data sources, was absent. Even in studies using voice recordings [ 27 ] or wearables [ 28 – 30 ], patients contributed physiological or conversational data but had no clearly documented role in tool design, threshold setting or interpretation of outputs. Therefore, most AI interventions remain clinician-centred, prioritising HCP workflow efficiency and clinical data extraction over direct patient agency (See Table 2 ). Table 2 Characteristics of the included studies Study Country Design Participants Context Conditions Bowles et al. [ 19 ] United States of America (USA) Qualitative study using focus group interviews 3 palliative care team members Home and community health provider 3 seriously ill patients as identified by the algorithm Gensheimer et al. [ 20 ] USA Quantitative interventional study with a controlled before-and-after design 1,251 patients with metastatic cancer Acute hospital, 4 oncology clinics Metastatic cancer Manz et al. [ 21 ] USA Randomised controlled trials (RCT) 20,506 cancer patients (41,021 encounters) Acute hospital, 9 tertiary and community-based oncology clinics Cancer (any stage) Kamdar et al. [ 22 ] USA RCT 112 patients with advanced cancer Acute hospital, outpatient palliative care clinic Advanced cancer Wilson et al. [ 23 ] USA RCT 2,544 patients Acute hospitals, 12 nursing units Multiple medical conditions Braun et al. [ 24 ] Germany Quantitative descriptive study with expert evaluation of case vignettes 5 experts in gynaecologic oncology (3) and palliative care (2) Unspecified, implied academic medical centres 10 patient vignettes of gynaecologic cancers with metastases receiving palliative care Griebhammer et al. [ 25 ] Germany Quantitative comparative clinical trial 34 and 19 palliative care patients in study arms I and II, respectively Acute hospital Life-threatening illnesses requiring hospitalisation Kawashima et al. [ 26 ] Japan Quantitative, retrospective cohort study 561 patients with advanced cancer Acute hospital Advanced cancer receiving chemotherapy Dong et al. [ 27 ] Japan Quantitative, cross-sectional study 100 home-based palliative care patients Community Multiple medical conditions Yang et al. [ 28 ] Taiwan Quantitative, prospective observational study 60 end-stage cancer patients Acute hospital End-stage cancer Liu et al. [ 29 ] Taiwan Quantitative, prospective observational cohort study 40 terminal cancer patients Acute hospital Terminal cancer Haleem et al. [ 30 ] Cyprus Quantitative RCT with a between-subject design 999 patients with advanced cancer Community-based cancer care organisation Advanced cancer Nejatifar et al. [ 31 ] Iran Quantitative, cross-sectional study 140 patients with chronic obstructive pulmonary disease (COPD) Acute hospital, respiratory clinic COPD with frailty indicators Sampson et al. [ 32 ] United Kingdom Quantitative, cross-sectional study 63 people with moderate and severe dementia Acute hospitals, 6 wards Dementia Blanes-Selva et al. [ 33 ] Italy, Brazil, Spain, Greece, Scotland, Portugal Mixed-methods study 21 palliative care healthcare professionals (HCPs), including 15 physicians and 6 nurses Multiple healthcare settings where HCPs use a web-based palliative care clinical decision support system 6 patient vignettes with a poor prognosis How AI is used Three types of AI have been conceptually identified, with some studies focusing on more than one category. Early identification of palliative care needs (N = 9) This concept is the dominant application, present in almost every year of publication and is particularly prominent from 2021 to 2023. These studies used ML or deep learning (DL) models applied to electronic health records, claims data, frailty indices or wearable actigraphy to predict mortality, deterioration or specialist palliative care requirements, with the aim of triggering earlier referral or advance care planning [ 19 , 20 , 21 , 23 , 26 , 28 , 29 , 31 , 33 ]. Symptoms assessment and management (N = 6) This concept became more prominent between 2024 and 2025, reflecting the emergence of novel, multimodal and sensor-based approaches. These include AI-based facial recognition for pain assessment in people with dementia [ 32 ], contactless radar heart-rate monitoring [ 25 ], voice recognition for patient-reported outcome extraction [ 27 ], multimodal deep-learning QoL estimation from wearables [ 30 ], digital therapeutic applications with AI triage for cancer pain [ 22 ], and evaluation of ChatGPT therapy suggestions [ 24 ]. Clinical decision support for care conversations (N = 4) This concept appeared consistently but less frequently across the period, often combining predictive models with behavioural nudges [ 20 , 21 ], generative LLMs to translate claims data into narrative summaries [ 19 ], or user-centred clinical decision support systems with explainability features [ 33 ]. Some studies contributed to more than one concept. For instance, Haleem (2025) combined multimodal wearable data with patient-reported outcomes to estimate deep-learning QoL. This illustrates the growing convergence of predictive and symptom monitoring capabilities in newer applications (See Table 3 ). Table 3 Concepts identified and how AI is used. Concepts (N) Primary AI used How AI is used Studies Early identification of palliative care needs (9) - Gradient boosting models (XGBoost / GBM) - Machine learning (ML) risk prediction models - Deep learning (DL) sequence models (LSTM / BiLSTM) - Ensemble learning (Super Learner) - Predict mortality/survival risk (short- to long- term horizons) - Identify frailty or palliative-care needs - Trigger referrals/serious illness conversations - Prioritise advance-care planning [ 19 , 20 , 21 , 23 , 26 , 28 , 29 , 31 , 33 ] Symptoms assessment and management (6) - Generative large language models (LLMs) - Multimodal DL - Facial recognition ML - Radar-based ML - Voice recognition ML - Real-time pain/symptom triage - Contactless vital-sign monitoring - Automated patient-reported outcome measures extraction from voice - Multimodal quality-of-life estimation from wearables - Generate/evaluate symptom-management advice [ 22 , 24 , 25 , 27 , 30 , 32 ] Clinical decision support for care conversations (4) - Generative LLMs - Behavioural nudges - SHapley Additive exPlanations explainability - User-centred clinical decision support systems (CDSS) - Translate claims/ electronic health record data into narrative summaries - Trigger nudges (emails/lists) for conversations - Present predictions vs. clinician judgement in CDSS - Support shared decision-making [ 19 , 20 , 21 , 33 ] Figure 2 demonstrates how AI has been applied across palliative care domains. Figure 2. Heatmap of artificial intelligence (AI) applications across palliative-care domains AI algorithms and modalities Structure-data predictive modelling was used in seven studies [ 20 , 21 , 23 , 26 , 29 , 31 , 33 ]. These approaches use structured data to train prognostic classifiers (such as gradient boosting, tree-based models and ensemble approaches) to identify patients at higher risk of mortality or requiring specialist palliative input. Most studies have focused on algorithm development and validation [ 26 , 29 , 31 , 33 ], with only a few evaluating models integrated into care delivery [ 20 , 21 , 23 ]. Time series and multimodal DL were employed in three studies [ 28 , 29 , 30 ], primarily to handle continuous or high-dimensional data. This includes the use of Recurrent Neural Networks and Long Short-Term Memory networks to predict survival outcomes in end-stage cancer [ 28 ] and estimate health-related quality of life (HRQoL) from wearable physiology [ 30 ]. AI-enabled digital therapeutic applications have been reported in one study [ 22 ]. The AI component functions as an interactive decision-support logic to guide tailored self-management content and escalation pathways for pain control. Generative AI and LLMs were incorporated in two studies [ 19 , 24 ]. These were applied to unstructured text tasks, specifically generating narrative clinical summaries to support palliative care needs assessment [ 19 ] and to evaluate guideline conformity and the quality of therapy suggestions in a palliative setting [ 24 ]. Three studies used specialised technologies to facilitate symptom assessment [ 25 , 27 , 32 ]. These include facial recognition to support pain assessment in individuals with dementia [ 32 ], voice recognition to detect patient-reported outcomes using conversational audio [ 27 ] and contactless radar-based heart rate monitoring as a low-burden physiological modality [ 25 ]. Reported health outcomes QoL In this review, QoL outcomes were defined using validated QoL/HRQoL instrument scores. Two of the 15 studies (13%) directly measured QoL using dedicated instrument [ 22 , 30 ]. Both studies discussed AI as a potential tool for improving the QoL [ 22 , 30 ]. Although QoL is rarely a primary endpoint, several studies have incorporated QoL-related measures within modelling [ 26 , 27 , 28 ], but not as direct QoL outcomes: Kawashima (2024) used a patient-reported Pain Numerical Rating Scale (NRS) for modelling specialist palliative-care needs [ 26 ], Dong (2024) used Integrated Palliative Care Outcome Scale (IPOS) in their Patient-Reported Outcome Measures (PROM) framework [ 27 ], and Yang (2021) used Karnofsky Performance Status (KPS) for survival prediction [ 28 ]. Additionally, many studies on QoC outcomes have addressed the potential for improving QoL. Kawashima (2024) concluded that earlier identification of palliative-care needs via ML "has the potential to contribute to earlier palliative care" and thereby improve QoL whereas Bowles (2024) briefly mentions in the abstract that the narratives created from generative AI have "the potential to support clinical decision-making" and "improve quality of life" by enabling timelier needs assessment. QoL instruments used were : Functional Assessment of Cancer Therapy General (Kamdar 2024) [ 22 ] Generalized Anxiety Disorder (Kamdar 2024) [ 22 ] Brief Pain Inventory (Kamdar 2024) [ 22 ] HRQoL (Haleem 2025) [ 30 ] EORTC Quality of Life Questionnaire Core-30 (EORTC HRQoL C30) (Haleem 2025) [ 30 ] Hospital Anxiety and Depression Scale (Haleem 2025) [ 30 ] IPOS (Dong 2024, Haleem 2025) [ 27 , 30 ]. KPS (Yang 2021) [ 28 ] Pain NRS (Kawashima 2024) [ 26 ] QoC All 15 studies (100%) reported QoC-related outcomes [ 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ]. Of these, 14 measured it quantitatively [ 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ] and one was primarily a case-study-style evaluation without quantitative clinical effect endpoints [ 19 ]. In intervention-oriented studies, QoC outcomes were frequently process and utilisation endpoints, such as serious-illness conversation rates [ 21 ], Advance-care planning (ACP) documentation rates [ 20 ], palliative care consultation rates [ 23 ] and service utilisation outcomes (e.g. readmissions, Intensive care unit (ICU) transfer and length of stay) [ 23 ]. In contrast, algorithm development and validation studies have emphasised discrimination performance metrics (e.g. area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, accuracy and confusion metrics) [ 26 , 27 , 28 , 29 , 31 ], and technology validation studies have reported agreement metrics (e.g. Bland–Altman, mean error, inter-rater reliability and validity coefficients) [ 25 , 32 ]. Finally, studies evaluating clinician-facing systems used usability and expert judgment endpoints (e.g. System Usability Scale (SUS), Short User Experience Questionnaire – Short (UEQ-S) and expert ratings) [ 24 , 33 ]. Additionally, some technology-focused studies have explicitly framed QoC benefits in terms of care processes. Grießhammer (2024) evaluated contactless monitoring to enhance routine symptom management, whereas Dong (2025) suggested that automated voice-PROM extraction could reduce the clinician documentation burden. Quality of Care instruments identified include : Palliative Prognostic Index (Yang 2021) [ 28 ] Epic Systems electronic medical record ACP form entries (Epic, Verona, WI); National Quality Forum end-of-life quality measures (Gensheimer 2022) [ 20 ] Palliative care consultation; service utilisation outcomes (e.g. 30-, 60-, and 90-day readmission, ICU transfer and inpatient length of stay) (Wilson 2023) [ 23 ] Voice recognition rate (Dong 2024) [ 27 ] Predictive performance metrics (area under the curve, F1, precision, recall, specificity, sensitivity, accuracy and confusion metrics) (Kawashima 2024, Dong 2024, Yang 2021, Liu 2023, Nejatifar 2025) [ 26 , 27 , 28 , 29 , 31 ] Expert Likert scale ratings of overall treatment proposal, evidence/guideline conformity and applicability of recommendations (Braun 2024) [ 24 ] SUS (Blanes-Selva 2023) [ 33 ] Short UEQ-S (Blanes-Selva 2023) [ 33 ] Rate of documented serious-illness conversations (Manz 2023) [ 21 ] Rate of ACP documentation (Gensheimer 2022) [ 20 ] Rate of palliative-care referral (Wilson 2023) [ 23 ] Pain intensity score (0–10) (Kamdar 2024) [ 22 ] Inter-rater reliability/validity coefficients (correlation coefficient, ICC, Cohen's kappa, and Gwet's AC) (Sampson 2025) [ 32 ] Agreement Analysis (Bland-Altman plot, mean error) (Grießhammer 2024) [ 25 ] Minimal Documentation System for routine daily symptom assessment (Grießhammer 2024) [ 25 ] Health related outcomes Across the 15 studies, AI applications demonstrated predominantly positive or neutral-to-positive effects on the measured outcomes with no clear negative effects reported (see Table 4 ). The strength of this inference varied according to the study type. Interventional studies tended to report clinically interpretable processes or utilisation endpoints [ 20 , 21 , 22 , 23 ], whereas algorithm development and validation studies largely report proxy technical metrics (e.g. discrimination, agreement and reliability) rather than downstream patient benefit [ 25 , 26 , 27 , 28 , 29 , 31 , 32 ]. When statistical testing was reported, results were often significant (p < .05), especially in intervention studies and some validations [ 20 , 21 , 22 , 23 , 25 , 30 , 31 , 32 ]. Several technical studies have described performance descriptively with limited or no hypothesis testing [ 24 , 26 , 28 , 29 ]. Overall, the findings are promising, but at an early stage. Many studies have shown that this model can work rather than improving care. Table 4 Effects of AI and direction of effect from the included studies Study (first author, year) Domain Outcomes measured or discussed Instrument/method Direction of effect Effect size Statistical technique & result Statistical significance Comparison to usual care Haleem, 2025 [ 30 ] Quality of Life (QoL)/Quality of Care (QoC) HRQoL Multimodal deep-learning regression for estimating HRQoL from smartwatch data and PROMs Positive MAPE = 0.242 Linear regression, prediction error, Spearman correlation No clinical effect test; Only p < .05 / .01 / .001 for Spearman correlation No comparison to usual care Kamdar, 2024 [ 22 ] QoL/QoC Pain (NRS), QoL domain scores Digital therapeutic application (ePAL) for pain management vs control Neutral–positive Mean pain scores: 2.99 (ePAL) vs 4.05 (Control) Descriptive statistics; two-sample t-tests, Chi-square, mixed linear effect models p < .05; pain improved (significant); QoL domains did not differ Patients with ePAL were associated with lower pain and fewer pain-related hospitalisations Blanes-Selva 2023 [ 33 ] QoC Usability & user experience of palliative clinical decision support systems (CDSS) SUS and UEQ-S Neutral–positive SUS = 62.7 (Round 1) / 65.0 (Round 2); UEQ-S = 1.42 (Round 1) / 1.5 (Round 2) Descriptive statistics; t-test p < .05; round-to-round differences were not significant No comparison to usual care Bowles, 2024 [ 19 ] QoC Feasibility of AI-generated summaries for supporting palliative-care needs assessment AI-generated narrative summaries from administrative claims data Neutral–positive N/A Qualitative focus group analysis from three palliative-care team members N/A AI provides sufficient assessment for palliative-care needs but is not adequate for direct clinical recommendation Yang, 2021 [ 28 ] QoC Survival outcomes via physical activity LSTM network using wearable actigraphy Neutral–positive AUC = 0.7292 Classification metrics, comparison with QoL-related indicators (KPS/Palliative Prognostic Index (PPI) N/A The model underperforms KPS and PPI; however, it does not require clinician input. Kawashima, 2024 [ 26 ] QoC Predicting specialist palliative-care needs XGBoost using EHR-derived variables, patient-reported pain, and QoL-related indicators Positive AUC = 0.89 Classification performance metrics N/A No comparison to usual care Nejatifar, 2025 [ 31 ] QoC Frailty-based prediction of palliative-care needs in chronic obstructive pulmonary disease Super Learner model using demographics and frailty conditions Positive AUC = 0.92 Classification performance metrics, chi-square and t-tests for relationships, logistic prediction for predictors p < 0.001 No comparison to usual care Manz, 2023 [ 21 ] QoC SIC rate; end-of-life (EOL) systemic therapy; EOL utilisation outcomes Machine learning (ML) mortality prediction via EHR data; behavioural nudge to prompt SICs Positive SICs: aOR 2.09; Reduced EOL systemic therapy: aOR 0.25 Descriptive statistics, generalized estimating equations, subgroup interaction analysis p < 0.001 Behavioural nudges to clinicians led to an increase in SICs and reduction in end-of-life systemic therapy Gensheimer, 2022 [ 20 ] QoC ACP documentation frequency; EOL quality measures Prognosis model to predict survival from EHR data; lay care coaches Positive ACP documentation: OR 13.7; Prognosis documentation: OR 9.9 Mixed-effects logistic regression p < 0.001 Combining a prognosis model with care coaches increased ACP documentation Wilson, 2023 [ 23 ] QoC Palliative care consultation; readmissions; ICU transfer; length of stay ML CDSS tool via EHR data integrated into clinical workflow Positive Palliative care consultations: IRR 1.44; 60-day readmission: OR 0.75; 90-day readmission: OR 0.72 Bayesian estimation accounting for the stepped design; 95% Credible Intervals N/A Increased palliative care rate consultations and reduction in hospitalisation Sampson, 2025 [ 32 ] QoC Reliability and validity of a pain assessment tool PainChek for automated pain assessment from facial analysis, voice, movement, behaviour activity and body Positive IRR 0.714 (rest), 0.817 (post-movement); Internal consistency 0.755 (rest), 0.833 (post-movement); Concurrent validity with PAINAD: 0.528 (rest), 0.787 (post-movement); Convergent validity with SM_EOLD − 0.555 (rest), -0.544 (post-movement) Inter-rater analysis, internal consistency, concurrent validity; comparison with Abbey Pain Scale (APS), PAINAD and SM-EOLD N/A PainChek compares favourably with APS. Griebhammer, 2024 [ 25 ] QoC Feasibility of radar-based heart rate monitoring in palliative care Continuous-wave radar units installed underneath the head section of the slatted frame for contactless monitoring Neutral–Positive Mean Difference: Arm I − 0.852 beats per minute (bpm), ARM II − 0.678 bpm Concordance analysis; Bland-Altman; two one-sided test (TOST) p < 0.001 Moderate agreement when compared with electrocardiogram Dong 2024 [ 27 ] QoC Accuracy of patient-reported pain symptom detection ML to extract patient-reported pain symptoms from transcribed patient interviews Neutral–positive Voice recognition rate 55.6%; F1 0.31 − 0.46 over five symptoms Voice recognition rate; classification metrics; Kruskal–Wallis test p = 0.515 for differences in recognition rate across disease groups Moderate performance compared with human-transcribed Integrated Palliative care Outcome Scale Liu, 2023 [ 29 ] QoC 7-day mortality prediction ML to predict 7-day mortality from clinical assessment and wearable data Positive AUC = 0.960 Classification metrics N/A No comparison to usual care Braun, 2024 [ 24 ] QoC Expert rating of ChatGPT therapy suggestions ChatGPT to provide therapy suggestions on 10 patient case vignettes Neutral–positive N/A Descriptive statistics N/A No comparison to usual care The evidence is clearest in studies that compared with usual care. Three RCTs showed superiority over usual care for clinically meaningful outcomes: improved serious-illness conversation documentation with reduced end-of-life systemic therapy [ 21 ], increased palliative-care consultation activity with fewer readmissions [ 23 ], and reduced pain intensity with fewer pain-related hospitalisations [ 22 ]. By contrast, algorithm development and validation studies typically report technical or procedural proxies. Prognostic and palliative care needsidentification models frequently reported good-to-excellent AUROC values and high sensitivity [ 26 , 29 , 31 ]; however, these metrics were rarely linked to demonstrated changes in care delivery. Survival prediction is feasible with moderate prognostic accuracy or agreement relative to a standard functional status tool [ 28 ]. Technology validation studies emphasise agreement, accuracy and reliability against reference standards [ 25 , 32 ], whereas clinician-facing systems emphasise usability and expert judgment endpoints rather than patient outcomes [ 24 , 33 ]. Practical, technical and ethical consequences of applying AI in palliative care This section synthesises both the explicit considerations raised by the authors of the 15 included studies and the implicit issues identified by the review team. Although promising, the application of AI in palliative care is accompanied by substantial practical, technical and ethical challenges that must be addressed before its routine clinical adoption. Practical considerations Integration into existing clinical workflows is a major barrier. Implementation-oriented work explicitly highlighted the need to avoid creating additional burdens (e.g. adding extra electronic health record (EHR) alerts) [ 19 , 21 , 22 , 23 , 33 ], with one study specifically designing an intervention to abrogate the potential for alert fatigue by avoiding a best-practice alert [ 23 ]. Clinician-facing tools have also raised practical concerns about how outputs fit into daily work [ 19 ], and usability studies have emphasised the need for systems that support professional autonomy within real clinical workflows [ 33 ]. Overall, AI systems should not disrupt routine practice or require additional interpretation time from already overburdened palliative-care teams [ 19 , 21 , 22 , 23 , 33 ]. Palliative clinicians have also reported severe staffing shortages and “no spare capacity” to learn or troubleshoot new systems [ 19 , 22 ]. Furthermore, wearable device studies highlight patient- and unknown-level burdens, including discomfort, device management, and data upload compliance [ 28 , 29 , 30 ]. In actigraphy-based survival prediction, patients were instructed to wear the device throughout admission, except during showering, because it was not water-resistant, and the data required periodic uploads [ 28 ]. In smartwatch-based mortality prediction, participants needed to operate smartphone syncing; some days’ data were excluded when uploads were absent, and user feedback reported charging problems and minor skin irritation [ 29 ]. Methodologically, many of the included studies were small, single-centre feasibility or validation projects with limited sample sizes, restricting generalisability and real-world scalability. Interdisciplinary co-design, iterative piloting, and dedicated staff training are essential prerequisites for acceptability; however, these elements are present in only a few studies [ 30 , 33 ]. Technical considerations Models rely heavily on structured EHRs or claims data of variable quality, suffer from class imbalances, and exhibit limited generalisability beyond their original institutions and condition cohorts [ 26 , 31 ]. Moreover, class imbalance is common (e.g. a low prevalence of palliative care needs or mortality events) and is typically addressed by undersampling or ensemble methods, thereby reducing effective training data [ 26 , 28 ]. Regarding generative AI, LLM-based approaches raise risks around incorrect or unsafe content generation and the need for careful human oversight [ 24 ], while narrative-summary approaches also raised pragmatic concerns about the nature of outputs (e.g., length/fit-for-workflow and confidentiality) that affect real-world utility [ 19 ]. Deep-learning systems can remain difficult to interpret; however, some studies used explainability approaches such as SHapley Additive exPlanations (SHAP) analyses for erroneous predictions [ 29 ] and others incorporated “explanatory” design features intended to preserve clinicians’ professional autonomy [ 33 ]. Finally, data requirements can be nontrivial; in wearable actigraphy survival prediction, performance was better when using 48 h of input data compared to 24 h [ 28 ], and wearable-based studies depended on sustained device use and data completeness (e.g. synchronisation and handling missingness) [ 29 , 30 ]. Validation in several studies ends with technical performance metrics rather than sustained clinical monitoring and downstream outcomes, limiting its usefulness for implementation decisions. Ethical considerations and mitigation strategies Ethical engagement specific to AI in the included studies was strikingly limited and was typically confined to baseline research governance, with bias, autonomy, and equity rarely addressed in depth. For instance, prospective wearable work reported Institutional Review Board approval and informed consent [ 28 , 29 ]; however, explicit discussions of AI-specific consent, communication with participants and equity/fairness implications were generally limited in reporting across study types. When mapped against the World Health Organization’s (WHO) six governing principles for AI in health (2021) [ 10 ] and the UKRIO’s practical guide for AI researchers (2025) [ 14 ], the evidence reveals major gaps that must be addressed before widespread adoption in high-stakes, vulnerability-intensive fields such as palliative and end-of-life care: Principle 1 – Protecting human autonomy Multiple studies have highlighted clinicians’ concerns that AI can diminish clinical judgement or the human elements of care [ 19 , 33 ]. This concern is justified, as any perception that decisions are delegated to an algorithm risks undermining trust and the therapeutic relationship between HCPs and patients. Informed consent specifically for AI processing of data is rarely described; in wearable studies, this may be implied rather than explicitly stated. Mitigation Human-in-the-loop must be mandatory with explicit clinician override and explicit tiered consent for AI use, with patients retaining the right to opt for human-only care pathways. These strategies aim to preserve the doctor-patient relationship, which is the cornerstone of palliative care and ensure that AI remains a supportive tool rather than a replacement for human compassion and care. Principle 2 – Promoting human well-being, safety and the public interest High-sensitivity predictive models risk causing unnecessary distress through false positives or premature palliative-care discussions if poorly governed [ 23 , 26 ]. Although some studies have used firewalls and internal processing [ 19 , 30 ], none have reported formal Data Protection Impact Assessments or compliance with the WHO-recommended minimum necessary data principles [ 10 ]. This is problematic because inadequate privacy safeguards risk re-identification of sensitive end-of-life data. Specifically, participants in wearable studies [ 28 , 29 , 30 ] were exposed to re-identification risks through continuous behavioural data. Mitigation Context-specific thresholds, continuous post-launch surveillance, rapid error reporting, data minimisation, pseudonymisation, and independent security audits should be implemented. These suggestions ensure that participants’ data privacy is safeguarded, and ongoing mechanisms are in place to report potential data breaches. Principle 3 – Ensuring transparency, explainability and intelligibility Despite some explainability attempts (e.g. SHAP analyses) [ 29 , 33 ], most studies deploy black-box models without patient-facing explanations or documentation of limitations. The black-box nature inherently limits HCPs’ trust in the modules and goes against patients’ right to understand how decisions affecting their care are made. Mitigation Patient-facing plain language explanations, validated explanation strategies (e.g. SHAP/ Local Interpretable Model-agnostic Explanations where appropriate), and transparent documentation of the model’s intended use, limitations, and known failure modes. These strategies ensure that patients and their families meaningfully engage in shared decision-making regarding their care. Principle 4 – Fostering responsibility and accountability No study has reported a formal ethical review of AI components or established accountability mechanisms. It can be inferred that the relevant ethics committee approved the studies; however, AI-specific oversight could be further contested. Mitigation Palliative-specific AI governance committees with patient/family representation and public registration of health AI systems are required. These structures are necessary to ensure that accountability is shared among developers, clinicians, and institutions, thereby protecting vulnerable patients from unaddressed harm. Principle 5 – Ensuring inclusiveness and equity Fourteen studies originated from high- or upper-middle-income settings with limited ethnic, socioeconomic, or geographic diversity, potentially widening the existing palliative care inequities in underserved populations. Certain study designs such as wearable-heavy approaches intentionally exclude patients with poor digital literacy or physical limitations [ 28 , 29 , 30 ], perpetuating digital exclusion in this vulnerable and seldom-heard-of group. Mitigation Bias audits with fairness metrics, diverse multinational training data, community co-design, and low-tech alternatives with accessibility testing are required, all of which are needed to prevent AI from exacerbating inequities in palliative care. Principle 6 – Promoting AI that is responsive and sustainable None of the studies mentioned environmental concerns, and the long-term responsiveness and sustainability of wearable devices remain unaddressed. Unsustainable designs increase ecological burden and exclude patients in resource-constrained settings, contradicting the holistic ethos of palliative care. Mitigation Environmental impact assessments and designs that are viable across resource settings must be incorporated. This is appropriate, because palliative care should consider broader societal and planetary contexts. Cross-cutting issue – Absence of informed consent for AI use (Principles 1, 2, and 3) : Retrospective data studies [ 26 , 31 ] and wearable devices routinely fail to describe specific consent for AI processing. This could be implied as an ethics committee waiver for secondary analysis; however, this was not explicitly stated. Without explicit consent for AI use, participants are denied agency over the use of sensitive data, potentially violating their autonomy. Mitigation Explicit, tiered consent or valid ethics-committee waiver is mandatory. Table 5 summarises the identified ethical considerations and proposed mitigation strategies. Table 5 Identified ethical considerations and proposed mitigation strategies Ethical considerations WHO (2021) Principle Mitigation strategies Bias, discrimination, marginalisation and digital exclusion from: - Homogeneous datasets - Wearable/application barriers - Developer biases - Bias in deployment 5 – Inclusiveness and equity 6 – Promote artificial intelligence that is responsive and sustainable - Bias audits and fairness metrics; diverse multinational data; community co-design - Protecting/minimising data colonialism - Early and ongoing stakeholder engagement - Low-tech alternatives - Accessibility testing with seldom-heard-of populations Inadequate privacy/data protection 2 – Promote human well-being, human safety and the public interest - Mandatory data protection impact assessment - Data minimisation - Pseudonymisation - Independent security audit Over-reliance on artificial intelligence (AI), erosion of autonomy and potential absence of informed consent for AI use 1 – Protect autonomy 6 – Promote AI that is responsive and sustainable - Human-in-the-loop as mandatory mechanism - ‘Graceful transition’ period and training for staff - Explicit tiered consent for AI processing - Right to opt for human-only pathway Lack of transparency and explainability 3 – Ensure transparency, explainability and intelligibility - Patient-facing explanations - publish model cards - Validated explainability modules - Regular AI Audit Harm from false positives/negatives 2 – Promote human well-being, human safety and the public interest - Context-specific thresholds - Continuous AI performance surveillance - Rapid-error reporting process No formal governance or accountability 4 – Foster responsibility and accountability 6 – Promote AI that is responsive and sustainable - Palliative-specific AI governance committee; public registration of systems - Impact assessment at protocol stage Discussion This scoping review identified three primary domains of AI application in palliative care: (1) early identification of palliative care needs through prognostic modelling (n = 9), (2) symptom assessment and relief using sensor-based and multimodal tools (n = 6) and (3) clinical decision support for care conversations via nudges and generative models (n = 4), with some studies contributing to more than one domain. These align with broader trends in the field where predictive analytics have dominated early efforts, particularly from 2021–2023, reflecting the maturation of ML techniques for mortality forecasting and palliative care identification from EHRs and related structured data [ 20, 21, 23, 28, 29, 33, 34]. A bibliometric analysisreports of this shift reported that early AI research (pre-2023) focused on prognostication to address late referrals, with annual outputs rising from a few papers to a peak of 66 in 2024, driven by multimodal integration [ 34 ]. The surge in symptom assessment applications in 2024–2025 mirrored the rapid adoption of wearable and voice technologies following the COVID-19 pandemic. This is evident from a scoping review that highlighted the role of DL in real-time HRQoL estimation and pain detection in publications increasing since 2023 [ 13 ]. Nevertheless, decision support remains consistent, but niche, often leveraging LLMs such as ChatGPT for narrative generation [ 19 , 24 ], which is consistent with the wider literature on generative AI's potential for communication aids [ 11 ]. Overall, the convergence across domains, from siloed prediction to integrated systems, parallels global trends toward "explainable AI," which emphasises hybrid models for end-of-life personalisation [ 7 ]. However, with most studies originating from Global North settings, these innovations risk irrelevance in low-resource contexts, where new technologies, such as low-cost radar monitoring [ 25 ], show promise but require Southern-led validation. Regarding the effectiveness on QoL and QoC, AI demonstrated positive effects across the measured outcomes, with moderate to large effect sizes and consistent statistical significance. Three RCTs reported superiority to usual care in conversation initiation, referrals, and pain outcomes ( Table 4 ). Direct measurement of QoL is rare [ 22 , 30 ], but indirect benefits via earlier interventions have been noted, aligning with the literature showing that AI-enhanced symptom management can improve global health scores on scales such as the EORTC Quality of Life Questionnaire Core-30 [ 22 ]. Wider evidence supports these findings. Reviews of ML in palliative care found that predictive tools reduced high-intensity end-of-life treatments, correlating with better QoL [ 7 , 35 ], while bibliometric trend projects continued to grow in AI for personalised comfort [ 34 ]. Nevertheless, QoC outcomes dominated (100% of studies), with strong evidence of usability (SUS scores 62–65) (see Table 4 ) and process improvements (e.g. an 11% absolute referral increase) (see Table 4 ), echoing a 2025 systematic review that attributes such gains to routine data integration and enhanced resource allocation in understaffed settings [ 35 ]. The emphasis on QoC over QoL reflects a broader skew in the literature, which limits generalisability [ 13 ]. This QoC bias risks overlooking holistic palliative goals, where AI's true value lies in augmenting rather than just streamlining care. The identified practical considerations include workflow disruption, alert fatigue, and resource demands, mirroring the wider challenges in AI implementation. Integration issues such as time burdens from AI outputs [ 19 , 33 ] align with a recent review which reported that the implementation of palliative AI is stalled by clinician overload, particularly in underresourced contexts [ 7 ]. Furthermore, inherent barriers to wearable compliance [ 28 , 30 ], namely high dropout rates due to discomfort in frail patients and usability, underscore the need for co-design [ 36 ]. Small, single-centre studies further constrain scalability, consistent with calls for interdisciplinary piloting to bridge the feasibility of practice [ 11 , 36 ]. Regarding technical considerations, technical limitations such as data quality variability, class imbalance, and poor generalisability [ 26 , 31 ] are recurrent patterns in the palliative AI literature. That is, overreliance on EHRs tends to lead to missing values, thereby reducing model accuracy [ 34 ]. Black-box issues persist despite SHAP’s attempts [ 33 ], mirroring the wider literature in which interpretability remains challenging [ 7 ]. Moreover, infrastructure requirements [ 19 ] and input length requirements [ 28 ] highlight scalability gaps to enhance robustness across sites [ 7 , 13 ]. Second, identified ethical considerations such as bias from homogeneous data, unclear AI consent, and opacity can amplify vulnerabilities in palliative care research and delivery. Given that most of the included studies (93%) were from Global North settings, this skewed data underperforms in diverse populations [ 10 ]. These ethical concerns can be addressed by embedding relevant AI ethical frameworks, such as those from the WHO or UKRIO, to ensure that AI use by participants is appropriately safeguarded [ 10 , 14 ]. Overall, this scoping review maps AI's evolving role of AI in palliative care, but reveals an immature, northern-centric evidence base. Future work must prioritise diverse participatory trials to realise equitable benefits guided by more robust AI ethical frameworks. Strengths of the study We included a wide range of study designs to provide a comprehensive overview of real-world data on AI applications in palliative care settings. The methodology was rigorous with a prospectively registered protocol, dual-independent screening, and structured data extraction with verification at both the title/abstract and full-text stages, along with regular team discussions for robust interpretation. The review team was deliberately diverse and comprised two palliative care clinicians/researchers and two computer engineers with AI expertise. Three members are based in the Global South (Thailand) and one is in the Global North, offering a more balanced perspective on the opportunities and equity implications of this emerging technology in different resource contexts. We also identified equity, diversity, and inclusion issues, including the digital divide, limited access to technology in low-resource settings, and the risk that most models are trained predominantly on data from high-income, white-majority populations, which warrant urgent further exploration in future research to prevent data colonisation [ 10 ]. Finally, we systematically outlined practical, technical and ethical considerations against established frameworks [ 10 , 14 ], thus providing a clear roadmap for responsible AI development in palliative care. Limitations This review has several limitations: First, the searches were limited to studies published in English and in five major databases; relevant research in other languages or unpublished/grey literature may have been missed. Second, consistent with the JBI scoping review guidelines, no formal critical appraisal of methodological quality or risk of bias was performed; therefore, readers should interpret effectiveness claims cautiously, particularly given the predominance of small, single-centre studies. Moreover, no meta-analysis was conducted owing to substantial heterogeneity in study designs, AI modalities, and outcome measures, which is an expected limitation of scoping reviews and is fully in line with JBI recommendations. Third, the evidence is heavily skewed toward Global North institutions (93%), restricting insights into their applicability in low- and middle-income settings. Finally, studies with negative or null results were frequently underreported [ 37 ]. For the quantitative and mixed-methods studies included here, publication bias may have favoured more positive findings, potentially influencing our overall interpretation. However, we have remained impartial and reported the data exactly as presented by the original authors. A practical suggestion beyond this review is to encourage routine publication or registration of all trial results [ 37 ], including negative findings, as these provide invaluable insights for the safe development of AI in palliative care. Implications for future studies Therefore, additional qualitative and codesigned studies are required. None of the 15 included studies used purely qualitative methods, and patient or caregiver voices were almost entirely absent beyond providing physiological or voice data. In-depth interviews, focus groups, or experience-based co-design approaches could provide richer insights into how patients, families, and clinicians perceive and interact with AI tools in the emotionally charged context of palliative care. Such studies would help ensure that future technologies remain person-centred rather than clinician- or data-driven. Mixed-methods designs should be prioritised to capture both measurable effectiveness (e.g. referral rates and prognostic accuracy) and contextual factors, such as cultural acceptability, trust, and relational impact. This strengthens the evidence base and improves real-world applicability. Therefore, there is an urgent need to improve geographic diversity. Fourteen of the fifteen studies originated in the Global North, with only one from Iran representing the Global South. This imbalance limits our understanding of how cultural norms, varying levels of digital literacy, health-system infrastructure, and resource constraints shape AI implementation. Future research should actively involve partners and study sites from low- and middle-income countries to produce more contextually grounded, equitable AI solutions for palliative care worldwide. Conclusions AI demonstrates clear potential to enhance palliative care by enabling earlier identification of needs, supporting real-time symptom assessments, and facilitating decision-making and care conversations. The 15 studies in this review showed generally favourable findings in prognostic accuracy, referral rates, conversation initiation, and pain management compared to that of usual care, with emerging evidence of positive effects on HRQoL. This field is rapidly maturing and moving from isolated predictive models to integrated multimodal systems. However, the current evidence base remains small, predominantly from Global North settings, and is limited in its focus on patient-reported outcomes and ethical governance. To translate these promises into an equitable, patient-centred reality, future research should prioritise larger, geographically diverse studies with explicit HRQoL endpoints, qualitative and co-design approaches that centre patient and caregiver experience, active collaboration with Global South partners, and systematic application of AI ethical frameworks from the outset. When responsibly developed, AI can significantly strengthen the compassionate and holistic delivery of palliative care. Abbreviations ACP Advance-care planning AI Artificial Intelligence AUROC Area Under the Receiver Operating Characteristic curve CDSS Clinical Decision Support Systems COPD Chronic Obstructive Pulmonary Disease DL Deep Learning EHR Electronic Health Record GBM Gradient Boosting Machine HCPs Health Care Professionals HRQoL Health-Related Quality of Life ICU Intensive care unit IPOS Integrated Palliative Care Outcome Scale JBI Joanna Briggs Institute KPS Karnofsky Performance Status LLM Large Language Model LSTM Long Short-Term Memory ML Machine Learning NLP Natural Language Processing NRS Pain Numerical Rating Scale PROM Patient-Reported Outcome Measures QoC Quality of Care QoL Quality-of-Life RCT Randomized Controlled Trial SHAP SHapley Additive exPlanations SUS System Usability Scale UEQ-S User Experience Questionnaire- Short UKRIO United Kingdom Research Integrity Office WHO World Health Organization XGBoost Extreme Gradient Boosting Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Availability of data and materials All data generated or analysed during this study are included in this published article and its supplementary information files. Competing interests The authors declare that they have no competing interests. Funding No funding was secured for this study. Authors' contributions TRP conceptualised the study, developed the methodology, performed software and validation tasks, conducted formal analysis and investigation, curated resources and data, prepared the original draft, reviewed and edited the manuscript, visualised results, supervised the project, and managed administration. SC conducted formal analysis and investigation, curated data, and contributed to reviewing and editing the manuscript. KT performed validation, formal analysis, investigation, data curation, contributed to reviewing and editing the manuscript, and visualised results. TNP conducted formal analysis and investigation, curated data, and contributed to reviewing and editing the manuscript. All authors read and approved the final manuscript. Acknowledgements We would like to express our gratitude to Scott McGregor for his assistance with the search terms and processes. References World Health Organization (WHO). Palliative care. 2020. https://www.who.int/news-room/fact-sheets/detail/palliative-care. Accessed 18 Nov 2025. Malhotra C, Shafiq M, Batcagan‑Abueg APM. What is the evidence for efficacy of advance care planning in improving patient outcomes? A systematic review of randomised controlled trials. BMJ Open. 2022; doi:10.1136/bmjopen-2021-060201 Tripodoro VA, Fidalgo JFL, Pons JJ, Connor SR, Garralda E, Bastos F, et al. First‑ever global ranking of palliative care: 2025 World Map under the new WHO framework. J Pain Symptom Manage. 2025; doi:10.1016/j.jpainsymman.2025.07.026 Pitzer S, Kutschar P, Paal P, Mülleder P, Lorenzl S, Wosko P, et al. Barriers for adult patients to access palliative care in hospitals: a mixed methods systematic review. J Pain Symptom Manage. 2024;67(1):e16–e33. Phenwan T, Anantapong K, Sripaew S, Vinayavekhin S. Exclusive, isolated and Buddhist‑centric at the expense of non‑conformists: a scoping review and bibliometric analysis of the palliative care ecosystem and its influence on quality of life, care and death. Advance. 2025. Topol EJ. High‑performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56. Silva‑Ferreira M, Cruz S, Luís MS, Silva MC, Monteiro‑Reis S, Henrique R, et al. Mapping study on AI‑based technologies in palliative care – a scoping study. BMC Palliat Care. 2025; doi:10.1186/s12904-025-01909-w. Arioz U, Allsop MJ, Goodman WD, Timmons S, Simbirtseva K, Mlakar I, et al. Artificial intelligence‑based approaches for advance care planning: a scoping review. BMC Palliat Care. 2025; doi:10.1186/s12904-025-01827-x. Pinto A, Santos C, Aguiar R, Oliveira S, Cunha D. The use of artificial intelligence in palliative care communication: a narrative review. Cureus. 2025; doi:10.7759/cureus.80524. World Health Organization (WHO). Ethics and governance of artificial intelligence for health: WHO guidance. 2021. https://www.who.int/publications/i/item/9789240029200. Accessed 18 Nov 2025. Nikoloudi M, Mystakidou K. Artificial intelligence in palliative care: a scoping review of current applications, challenges, and future directions. Am J Hosp Palliat Care. 2025; doi:10.1177/10499091251358379. Holm S, Ploug T. Co‑reasoning and epistemic inequality in AI supported medical decision‑making. Am J Bioeth. 2024;24(9):79–80. Bozkurt S, Fereydooni S, Kar I, Chalmers CD, Leslie SL, Pathak R, et al. AI in palliative care: a scoping review of foundational gaps and future directions for responsible innovation. J Pain Symptom Manage. 2025;70(6):e394–e418. UK Research Integrity Office (UKRIO). Embracing AI with integrity: a practical guide for researchers. 2025. https://ukrio.org/ukrio-resources/embracing-ai-with-integrity/. Accessed 29 Jul 2025 Khalil H, Peters MDJ, Tricco AC, Pollock D, Alexander L, McInerney P, et al. Conducting high quality scoping reviews: challenges and solutions. J Clin Epidemiol. 2021;130:156–160. Phenwan T, Sitthichok C, Tangudomkit K, Peerawong T. The use of artificial intelligence in palliative care: a scoping review protocol. OSF. 2025. Pollock D, Peters MDJ, Khalil H, McInerney P, Alexander L, Tricco AC, et al. Recommendations for the extraction, analysis, and presentation of results in scoping reviews. JBI Evid Synth. 2023;21(3):520–532. Peters MDJ, Marnie C, Tricco AC, Pollock D, Munn Z, Alexander L, et al. Updated methodological guidance for the conduct of scoping reviews. JBI Evid Implement. 2021;19(1). Bowles KH, Brickner C, Luth EA. Using generative AI to translate administrative claims data into narrative summaries for palliative care needs assessment: a case study. Stud Health Technol Inform. 2024;315:547–551. Gensheimer MF, Gupta D, Patel MI, Fardeen T, Hildebrand R, Teuteberg W, et al. Use of machine learning and lay care coaches to increase advance care planning conversations for patients with metastatic cancer. JCO Oncol Pract. 2023;19(2):e176–e184. Manz CR, Zhang Y, Chen K, Long Q, Small DS, Evans CN, et al. Long‑term effect of machine learning‑triggered behavioural nudges on serious illness conversations and end‑of‑life outcomes among patients with cancer: a randomized clinical trial. JAMA Oncol. 2023;9(3):414–418. Kamdar M, Jethwani K, Centi AJ, Agboola S, Fischer N, Traeger L, et al. A digital therapeutic application (ePAL) to manage pain in patients with advanced cancer: a randomized controlled trial. J Pain Symptom Manage. 2024;68(3):261–271. Wilson PM, Ramar P, Philpot LM, Soleimani J, Ebbert JO, Storlie CB, et al. Effect of an artificial intelligence decision support tool on palliative care referral in hospitalized patients: a randomized clinical trial. J Pain Symptom Manage. 2023;66(1):24–32. Braun EM, Juhasz‑Böss I, Solomayer EF, Truhn D, Keller C, Heinrich V, et al. Will I soon be out of my job? Quality and guideline conformity of ChatGPT therapy suggestions to patient inquiries with gynaecologic symptoms in a palliative setting. Arch Gynecol Obstet. 2024;309(4):1543–1549. Grießhammer SG, Malessa A, Lu H, Yip J, Leuschner J, Christgau F, et al. Contactless radar‑based heart rate estimation in palliative care – a feasibility study and possible use in symptom management. BMC Palliat Care. 2024;23(1):1–14. Kawashima A, Furukawa T, Imaizumi T, Morohashi A, Hara M, Yamada S, et al. Predictive models for palliative care needs of advanced cancer patients receiving chemotherapy. J Pain Symptom Manage. 2024;67(4):306–316.e6. Dong L, Hirayama H, Zheng X, Masukawa K, Miyashita M. Using voice recognition and machine learning techniques for detecting patient‑reported outcomes from conversational voice in palliative care patients. Jpn J Nurs Sci. 2025;22(1):e12644. Yang TY, Kuo PY, Huang Y, Lin HW, Malwade S, Lu LS, et al. Deep‑learning approach to predict survival outcomes using wearable actigraphy device among end‑stage cancer patients. Front Public Health. 2021;9:730150. Liu JH, Shih CY, Huang HL, Peng JK, Cheng SY, Tsai JS, et al. Evaluating the potential of machine learning and wearable devices in end‑of‑life care in predicting 7‑day death events among patients with terminal cancer: cohort study. J Med Internet Res. 2023; doi:10.7759/cureus.80524. Haleem MS, Aidonis V, Georga EI, Krini M, Matsangidou M, Kassianos AP, et al. A multimodal deep learning architecture for estimating health‑related quality of life for advanced cancer patients based on wearable devices and patient‑reported outcome measures. IEEE J Biomed Health Inform. 2026; doi: 10.1109/JBHI.2025.3597054. Nejatifar Z, Alizadeh A, Amerzadeh M, Omidian S, Rafiei S. The predictive role of identifying frailty in assessing the need for palliative care in the elderly: the application of machine learning algorithm. J Health Popul Nutr. 2025;44(1):133. Sampson EL, Davies N, Vickerstaff V. Evaluation of the psychometric properties of PainChek in older general hospital patients with dementia. Age Ageing. 2025;54(2):afaf027. Blanes‑Selva V, Asensio‑Cuesta S, Doñate‑Martínez A, Mesquita FP, García‑Gómez JM. User‑centred design of a clinical decision support system for palliative care: insights from healthcare professionals. Digit Health. 2023;9:20552076231181212. Pan M, Huang R, Liu C, Xiong Y, Li N, Peng H, et al. Application of artificial intelligence in palliative care: a bibliometric analysis of research hotspots and trends. Front Med (Lausanne). 2025;12:1597195. Bressler T, Song J, Kamalumpundi V, Chae S, Song H, Tark A. Leveraging artificial intelligence/machine learning models to identify potential palliative care beneficiaries: a systematic review. J Gerontol Nurs. 2025;51(1):7–14. Sandic Spaho R, Uhrenfeldt L, Fotis T, Kymre IG. Wearable devices in palliative care for people 65 years and older: a scoping review. Digit Health. 2023;9:20552076231181212. Bradley SH, DeVito NJ, Lloyd KE, Richards GC, Rombey T, Wayant C, et al. Reducing bias and improving transparency in medical research: a critical overview of the problems, progress and suggested next steps. J R Soc Med. 2020;113(11):433–443. Additional Declarations No competing interests reported. Supplementary Files Supplementaryfile1.Searchterms.docx SupplementaryFile2.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 15 May, 2026 Reviews received at journal 10 Apr, 2026 Reviewers agreed at journal 07 Apr, 2026 Reviewers agreed at journal 26 Mar, 2026 Reviewers agreed at journal 26 Feb, 2026 Reviewers invited by journal 26 Feb, 2026 Editor invited by journal 23 Feb, 2026 Editor assigned by journal 22 Feb, 2026 Submission checks completed at journal 22 Feb, 2026 First submitted to journal 20 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8924351","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":600226127,"identity":"bef037dd-1f41-4df5-9319-e9c3fb32d028","order_by":0,"name":"Thanarpan Peerawong","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Thanarpan","middleName":"","lastName":"Peerawong","suffix":""},{"id":600226129,"identity":"7dfbc5f8-0880-4e8f-83d4-4aca33c7bcdb","order_by":1,"name":"Tharin Phenwan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/ElEQVRIiWNgGAWjYFACHgjFxszY+ADIZmyAihvg1gDVws/e3GxAmhbJnuNtEkRpsZfuPfi4sM0uz+BGYlvF27Y7sv3sDYwffjAcNsZpi8y5ZOOZbcnFIC0357Y9M57Zc4BZsofhsBlOLRI5ZtK825gTNwC13OZtOwxkJDBIMzActsGjxfw377Z6sJZikJb99x8w/yagxYyZd9vhxJk9B9uYwbZIMLCBbMHtsDtnjKV5/x1P7GdvbJacc+6w8YwziW2WPQbpOL3PPrvH8DPPmerENmb2hx/elB2W7W8/fPjGjwprwwZceiQwhUBRgzNWsGsZBaNgFIyCUYAKAOSJWu5G1dW4AAAAAElFTkSuQmCC","orcid":"","institution":"University of Dundee","correspondingAuthor":true,"prefix":"","firstName":"Tharin","middleName":"","lastName":"Phenwan","suffix":""},{"id":600226131,"identity":"2a3e6b49-85b1-4c7f-8e9c-19bf224b9038","order_by":2,"name":"Sitthichok Chaichulee","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Sitthichok","middleName":"","lastName":"Chaichulee","suffix":""},{"id":600226133,"identity":"86333d09-eaa6-44a9-9e34-a4ef212272f7","order_by":3,"name":"Kamonrat Tangudomkit","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Kamonrat","middleName":"","lastName":"Tangudomkit","suffix":""}],"badges":[],"createdAt":"2026-02-20 09:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8924351/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8924351/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104403926,"identity":"6d23e850-a277-4046-81ca-5926a9e91cad","added_by":"auto","created_at":"2026-03-11 12:19:24","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":33080,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePRISMA flow diagram\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8924351/v1/3294a50847759b843d0e3ec0.png"},{"id":104171898,"identity":"913e3c09-ae1a-4a34-8cac-90b6d8225d7d","added_by":"auto","created_at":"2026-03-08 14:57:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":102599,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap of artificial intelligence (AI) applications across palliative-care domains\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8924351/v1/e0e2f878f9f1969f1f85ccde.png"},{"id":104409312,"identity":"9bb0a4e7-c2c7-4b28-8f8a-a4d8ae3ee507","added_by":"auto","created_at":"2026-03-11 12:44:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1849389,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8924351/v1/64b4a4b4-953f-4b4c-b00b-112feb7049f2.pdf"},{"id":104171894,"identity":"dac993e1-493e-444b-a9e7-4de2972a88e5","added_by":"auto","created_at":"2026-03-08 14:57:11","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":42983,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementaryfile1.Searchterms.docx","url":"https://assets-eu.researchsquare.com/files/rs-8924351/v1/67a85fd18f0a16baa186c4c5.docx"},{"id":104171896,"identity":"85d83851-2206-4ee3-8f29-6f68d49729d2","added_by":"auto","created_at":"2026-03-08 14:57:11","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":825844,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryFile2.docx","url":"https://assets-eu.researchsquare.com/files/rs-8924351/v1/a32cf1156570d27ce3fd7bb6.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Exploring the application of Artificial Intelligence in palliative care and its practical, technical and ethical considerations: a scoping review","fulltext":[{"header":"Background","content":"\u003cp\u003ePalliative care improves the quality of life of patients with life-limiting conditions and their families by addressing their physical, psychological, social and spiritual needs [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Early delivery of palliative care has been shown to reduce symptom burden, decrease unwanted aggressive interventions at the end of life and improve both patients\u0026rsquo; and carers\u0026rsquo; outcomes [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Despite these benefits, access remains limited worldwide, particularly due to workforce shortages, late referrals and resource constraints in both high- and low-income settings [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) and related technologies have emerged as potential tools to address these gaps [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. AI systems can analyse electronic health records, wearable sensor data or conversational voice data to identify unmet palliative care needs, predict prognosis, support symptom monitoring and facilitate conversations about serious illnesses [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. However, evidence of real-world applications in palliative care remains limited, with most studies focusing on retrospective electronic health record data rather than prospective use with patients or caregivers in diverse clinical contexts.\u003c/p\u003e \u003cp\u003eFew studies have evaluated the impact of AI on core palliative care outcomes, including patient quality of life (QoL) and quality of care (QoC). Moreover, although recent reviews have highlighted practical, technical and ethical challenges, comprehensive mapping is lacking [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. This is particularly concerning, given the vulnerability of the population and the additional ethical considerations that arise when AI interfaces directly with human decision-making in end-of-life care [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWith AI gaining rapid global momentum in healthcare, there is an urgent need to systematically capture the emerging evidence of its role in palliative care delivery, its effectiveness in improving QoL and QoC, and its associated practical, technical and ethical implications. Of particular concern is the risk that AI, developed predominantly in high-resource settings, may exacerbate existing inequities when applied globally. Therefore, this scoping review aims to:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003emap existing evidence on the application of AI in palliative care in a real-world context.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eexamine its reported effectiveness and influence on the quality of life and quality of care.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eidentify implicit and explicit practical, technical and ethical considerations, together with proposed mitigation strategies.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eReview questions\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhat types of AI have been implemented to improve palliative care delivery?\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhich health-related outcomes in the QoL and QoC domains have been measured or discussed?\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eHow effective are these AI applications in relation to the measured or discussed outcomes?\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eWhat practical, technical and ethical considerations have been identified, and what mitigation strategies have been proposed?\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eFor this review, relevant definitions from the United Kingdom Research Integrity Office (UKRIO) were used [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] p2:\u003c/p\u003e \u003cp\u003eAI is an umbrella term for a range of algorithm-based technologies that solve complex tasks by carrying out functions that previously required human thinking. Machine learning (ML) is a subset of AI that learns from data without being explicitly programmed. This learning can be unsupervised or human instructed/\u0026rsquo;human-in-the-loop\u0026rsquo; (reinforcement learning)\u0026rsquo;. Natural language processing (NLP) refers to processes that enable machines to understand, generate and manipulate human language. Generative AI is an \u0026lsquo;AI that can create new text, images, audio, video, and code\u0026rsquo;. An example includes ChatGPT, which is \u0026lsquo;a large language model (LLM) designed to receive text cues or prompts (inputs) and generate outputs using NLP. Multimodal AI refers to the use of multiple AI systems to process different types of data simultaneously.\u003c/p\u003e "},{"header":"Methods","content":"\u003cp\u003eA scoping review was conducted following the Joanna Briggs Institute (JBI) methodology for scoping reviews and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews checklist [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. The review protocol was registered in the Open Science Framework platform [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eEligibility criteria\u003c/p\u003e \u003cp\u003eStudies were included based on inclusion and exclusion criteria (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The Population, Concept and Context framework was used to formulate the criteria and review questions [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eInclusion and exclusion criteria\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInclusion criteria\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExclusion criteria\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePopulation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArticles that discuss:\u003c/p\u003e \u003cp\u003e- Adults receiving palliative care\u003c/p\u003e \u003cp\u003e- Family carers of adults receiving palliative care\u003c/p\u003e \u003cp\u003e- Healthcare professionals\u0026nbsp;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Children\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConcept\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Any application of AI to facilitate or enhance palliative care delivery (e.g., prognostication, symptom management, decision support, communication)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Articles not related to palliative care provision\u003c/p\u003e \u003cp\u003e- Theoretical discussions, legal/ethical/theoretical debates without any application\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eContext\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Any context of care (acute hospitals, long-term care facilities, community)\u003c/p\u003e \u003cp\u003e- Global\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ndash;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTypes of Evidence\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Quantitative studies\u003c/p\u003e \u003cp\u003e- Qualitative studies\u003c/p\u003e \u003cp\u003e- Mixed methods\u003c/p\u003e \u003cp\u003e- Grey literature\u003c/p\u003e \u003cp\u003e- Published in English\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Reviews, editorials, protocols\u003c/p\u003e \u003cp\u003e- Articles published in other languages\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eSearch strategy\u003c/p\u003e \u003cp\u003eFour team members, supported by an academic librarian, searched the Association for Computing Machinery Digital Library, the Cumulative Index to Nursing and Allied Health Literature, Cochrane Central, PubMed and Web of Science databases between September and October 2025. A combination of keywords, Boolean operators and truncations was applied to maximise the retrieval of relevant studies (see Supplementary File 1). Grey literature was eligible; however, searches were limited to the databases listed above.\u003c/p\u003e\n\u003ch3\u003eScreening and selection\u003c/h3\u003e\n\u003cp\u003eThe search results were uploaded to Covidence, a review management software programme for semi-automated duplicate removal. Duplicates were further removed manually.\u003c/p\u003e \u003cp\u003eInitially, four reviewers performed a 25-title pilot screening against the inclusion and exclusion criteria and discussed any ambiguities. Subsequently, a dual-screening process was applied. Four reviewers were randomly assigned to screen the articles independently. The lead reviewer was responsible for the final decision on whether to include or exclude records in cases of discrepancy. Titles with unclear or absent abstracts were included in the full-text review.\u003c/p\u003e \u003cp\u003eA similar process was then applied to the full-text review, starting with a pilot screening against the inclusion and exclusion criteria, followed by a dual-screening process of randomly assigned articles, with the lead reviewer resolving discrepancies. The reasons for exclusion were documented at the full-text stage.\u003c/p\u003e \u003cp\u003eData extraction and synthesis\u003c/p\u003e \u003cp\u003eA pilot extraction file was created using the Covidence software. The lead reviewer then extracted the relevant data from the pilot extraction file, and the process was iterative, ensuring that the review questions were fully answered. Once finalised through team discussions, two reviewers extracted the remaining data until completion. The remaining reviewers verified the completeness of the extracted data.\u003c/p\u003e \u003cp\u003eThe extracted data included:\u003c/p\u003e \u003cp\u003e \u003cem\u003eAuthor(s); Year of publication; Country of origin; Aims/objectives; Population and sample size; Concept; Methodology/methods used; Context of the study; Types of AI used; AI development and testing process, if relevant; Outcome measures/discussed; Influences/effects of AI used on quality of life and quality of care; Practical considerations; Technical considerations; Ethical considerations.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eStudy authors were not contacted for clarification or to obtain missing data.\u003c/p\u003e \u003cp\u003eNo formal critical appraisal of individual study quality was undertaken, which is consistent with the JBI scoping review guidelines [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSynthesis of results\u003c/p\u003e \u003cp\u003eDue to the anticipated heterogeneity in study designs and outcomes, findings were narratively synthesised and presented in tables and figures [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The included studies were categorised according to their shared characteristics, and the findings were related to the review questions.\u003c/p\u003e \u003cp\u003ePatient and public involvement\u003c/p\u003e \u003cp\u003eThere was no direct patient or public involvement in this review.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eOverall characteristics of the included studies\u003c/p\u003e \u003cp\u003eThe databases were searched through October 2025. In total, 814 records were identified. Of these, 204 duplicates were removed, and 610 titles and abstracts were screened. In total, 144 articles were included in the full-text assessment. Of these, 129 articles were excluded, leaving 15 studies for review (see Fig.\u0026nbsp;1).\u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;1. PRISMA flow diagram\u003c/b\u003e \u003c/p\u003e \u003cp\u003eStudy characteristics\u003c/p\u003e \u003cp\u003eThe included 15 studies were published between 2021 and 2025 and spannedmultiple countries: United states of America [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;5); Germany [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;2); Japan [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;2); Taiwan, China [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;2); Cyprus [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;1); Iran [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;1); UK [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;1), and one study conducted in six countries [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] (Brazil, Italy, Greece, Portugal, Scotland and Spain), indicating global interest in the topic.\u003c/p\u003e \u003cp\u003eOnly one study originated from the Global South (Iran) [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] and demonstrated potential epistemic inequality.\u003c/p\u003e \u003cp\u003eMost studies were conducted in acute hospitals [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;11). This was followed by multiple settings [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;3) and community settings [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] (N\u0026thinsp;=\u0026thinsp;1). None of the patients were admitted to long-term care facilities.\u003c/p\u003e \u003cp\u003eRegarding study designs, 14 studies employed quantitative designs. One study used a quasi-experimental design [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Four studies were randomised controlled trials (RCTs) [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Two studies used a prospective cohort design [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], while one study used a retrospective cohort design [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Other designs included vignette-based evaluations [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], cross-sectional studies [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], device validation studies [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], observational predictive modelling studies [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], and qualitative focus group studies [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn terms of conditions, six studies focused on advanced cancer [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] and one study focused on any stage of cancer [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. One study included patients with unspecified palliative conditions [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. One study included vignettes of patients with gynaecological cancer [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], while another focused on vignettes of patients with a poor prognosis [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Two studies focused on multiple palliative conditions [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], one on chronic obstructive pulmonary disease (COPD) [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], and one on individuals with dementia [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe sample sizes of the included studies were generally small (range 3\u0026ndash;20,506, median 112). Twelve studies primarily involved palliative care patients (with or without family carers as secondary participants) or used palliative care patient records [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Three studies focused primarily on healthcare professionals (HCPs) as the end users of AI tools [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. The inclusion of patients, families and HCPs reflects the interdisciplinary nature of palliative care. However, a true participatory design, in which patients or caregivers are co-researchers rather than passive data sources, was absent. Even in studies using voice recordings [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] or wearables [\u003cspan additionalcitationids=\"CR29\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], patients contributed physiological or conversational data but had no clearly documented role in tool design, threshold setting or interpretation of outputs.\u003c/p\u003e \u003cp\u003eTherefore, most AI interventions remain clinician-centred, prioritising HCP workflow efficiency and clinical data extraction over direct patient agency (See Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of the included studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCountry\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDesign\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eParticipants\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eContext\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eConditions\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBowles et al. [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnited States of America (USA)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQualitative study using focus group interviews\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3 palliative care team members\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHome and community health provider\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3 seriously ill patients as identified by the algorithm\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGensheimer et al. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative interventional study with a controlled before-and-after design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1,251 patients with metastatic cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital, 4 oncology clinics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMetastatic cancer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eManz et al. [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRandomised controlled trials (RCT)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20,506 cancer patients (41,021 encounters)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital, 9 tertiary and community-based oncology clinics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCancer (any stage)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKamdar et al. [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRCT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e112 patients with advanced cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital, outpatient palliative care clinic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAdvanced cancer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWilson et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRCT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2,544 patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospitals, 12 nursing units\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMultiple medical conditions\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBraun et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGermany\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative descriptive study with expert evaluation of case vignettes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 experts in gynaecologic oncology (3) and palliative care (2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eUnspecified, implied academic medical centres\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e10 patient vignettes of gynaecologic cancers with metastases receiving palliative care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGriebhammer et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGermany\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative comparative clinical trial\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34 and 19 palliative care patients in study arms I and II, respectively\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eLife-threatening illnesses requiring hospitalisation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKawashima et al. [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJapan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, retrospective cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e561 patients with advanced cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAdvanced cancer receiving chemotherapy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDong et al. [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJapan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, cross-sectional study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e100 home-based palliative care patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCommunity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMultiple medical conditions\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYang et al. [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTaiwan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, prospective observational study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e60 end-stage cancer patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eEnd-stage cancer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiu et al. [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTaiwan\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, prospective observational cohort study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e40 terminal cancer patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTerminal cancer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHaleem et al. [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCyprus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative RCT with a between-subject design\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e999 patients with advanced cancer\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCommunity-based cancer care organisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAdvanced cancer\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNejatifar et al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIran\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, cross-sectional study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e140 patients with chronic obstructive pulmonary disease (COPD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospital, respiratory clinic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCOPD with frailty indicators\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSampson et al. [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnited Kingdom\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eQuantitative, cross-sectional study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e63 people with moderate and severe dementia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAcute hospitals, 6 wards\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDementia\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlanes-Selva et al. [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eItaly, Brazil, Spain, Greece, Scotland, Portugal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMixed-methods study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21 palliative care healthcare professionals (HCPs), including 15 physicians and 6 nurses\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMultiple healthcare settings where HCPs use a web-based palliative care clinical decision support system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6 patient vignettes with a poor prognosis\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eHow AI is used\u003c/p\u003e \u003cp\u003eThree types of AI have been conceptually identified, with some studies focusing on more than one category.\u003c/p\u003e\n\u003ch3\u003eEarly identification of palliative care needs (N = 9)\u003c/h3\u003e\n\u003cp\u003eThis concept is the dominant application, present in almost every year of publication and is particularly prominent from 2021 to 2023. These studies used ML or deep learning (DL) models applied to electronic health records, claims data, frailty indices or wearable actigraphy to predict mortality, deterioration or specialist palliative care requirements, with the aim of triggering earlier referral or advance care planning [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eSymptoms assessment and management (N = 6)\u003c/h3\u003e\n\u003cp\u003eThis concept became more prominent between 2024 and 2025, reflecting the emergence of novel, multimodal and sensor-based approaches. These include AI-based facial recognition for pain assessment in people with dementia [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], contactless radar heart-rate monitoring [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], voice recognition for patient-reported outcome extraction [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], multimodal deep-learning QoL estimation from wearables [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], digital therapeutic applications with AI triage for cancer pain [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], and evaluation of ChatGPT therapy suggestions [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eClinical decision support for care conversations (N = 4)\u003c/h3\u003e\n\u003cp\u003eThis concept appeared consistently but less frequently across the period, often combining predictive models with behavioural nudges [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], generative LLMs to translate claims data into narrative summaries [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], or user-centred clinical decision support systems with explainability features [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSome studies contributed to more than one concept. For instance, Haleem (2025) combined multimodal wearable data with patient-reported outcomes to estimate deep-learning QoL. This illustrates the growing convergence of predictive and symptom monitoring capabilities in newer applications (See Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eConcepts identified and how AI is used.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConcepts (N)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrimary AI used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow AI is used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eStudies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEarly identification of palliative care needs\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(9)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Gradient boosting models\u0026nbsp;(XGBoost / GBM)\u003c/p\u003e \u003cp\u003e- Machine learning (ML) risk prediction models\u003c/p\u003e\u003cp\u003e- Deep learning (DL) sequence models\u0026nbsp;(LSTM / BiLSTM)\u003c/p\u003e\u003cp\u003e- Ensemble learning\u0026nbsp;(Super Learner)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Predict mortality/survival risk (short- to long- term horizons)\u003c/p\u003e \u003cp\u003e- Identify frailty or palliative-care needs\u003c/p\u003e \u003cp\u003e- Trigger referrals/serious illness conversations\u003c/p\u003e \u003cp\u003e- Prioritise advance-care planning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSymptoms assessment and management\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(6)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Generative large language models (LLMs)\u003c/p\u003e \u003cp\u003e- Multimodal DL\u003c/p\u003e\u003cp\u003e- Facial recognition ML\u003c/p\u003e\u003cp\u003e- Radar-based ML\u003c/p\u003e\u003cp\u003e- Voice recognition ML\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Real-time pain/symptom triage\u003c/p\u003e \u003cp\u003e- Contactless vital-sign monitoring\u003c/p\u003e \u003cp\u003e- Automated patient-reported outcome measures extraction from voice\u003c/p\u003e\u003cp\u003e- Multimodal quality-of-life estimation from wearables\u003c/p\u003e \u003cp\u003e- Generate/evaluate symptom-management advice\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eClinical decision support for care conversations\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(4)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e- Generative LLMs\u003c/p\u003e \u003cp\u003e- Behavioural nudges\u003c/p\u003e\u003cp\u003e- SHapley Additive exPlanations explainability\u003c/p\u003e \u003cp\u003e- User-centred clinical decision support systems (CDSS)\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Translate claims/ electronic health record data into narrative summaries\u003c/p\u003e \u003cp\u003e- Trigger nudges (emails/lists) for conversations\u003c/p\u003e \u003cp\u003e- Present predictions vs. clinician judgement in CDSS\u003c/p\u003e \u003cp\u003e- Support shared decision-making\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFigure 2 demonstrates how AI has been applied across palliative care domains.\u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;2. Heatmap of artificial intelligence (AI) applications across palliative-care domains\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAI algorithms and modalities\u003c/p\u003e \u003cp\u003eStructure-data predictive modelling was used in seven studies [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. These approaches use structured data to train prognostic classifiers (such as gradient boosting, tree-based models and ensemble approaches) to identify patients at higher risk of mortality or requiring specialist palliative input. Most studies have focused on algorithm development and validation [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], with only a few evaluating models integrated into care delivery [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTime series and multimodal DL were employed in three studies [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], primarily to handle continuous or high-dimensional data. This includes the use of Recurrent Neural Networks and Long Short-Term Memory networks to predict survival outcomes in end-stage cancer [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and estimate health-related quality of life (HRQoL) from wearable physiology [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAI-enabled digital therapeutic applications have been reported in one study [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The AI component functions as an interactive decision-support logic to guide tailored self-management content and escalation pathways for pain control.\u003c/p\u003e \u003cp\u003eGenerative AI and LLMs were incorporated in two studies [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. These were applied to unstructured text tasks, specifically generating narrative clinical summaries to support palliative care needs assessment [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and to evaluate guideline conformity and the quality of therapy suggestions in a palliative setting [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThree studies used specialised technologies to facilitate symptom assessment [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. These include facial recognition to support pain assessment in individuals with dementia [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], voice recognition to detect patient-reported outcomes using conversational audio [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] and contactless radar-based heart rate monitoring as a low-burden physiological modality [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eReported health outcomes\u003c/p\u003e\n\u003ch3\u003eQoL\u003c/h3\u003e\n\u003cp\u003eIn this review, QoL outcomes were defined using validated QoL/HRQoL instrument scores. Two of the 15 studies (13%) directly measured QoL using dedicated instrument [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Both studies discussed AI as a potential tool for improving the QoL [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Although QoL is rarely a primary endpoint, several studies have incorporated QoL-related measures within modelling [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], but not as direct QoL outcomes: Kawashima (2024) used a patient-reported Pain Numerical Rating Scale (NRS) for modelling specialist palliative-care needs [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], Dong (2024) used Integrated Palliative Care Outcome Scale (IPOS) in their Patient-Reported Outcome Measures (PROM) framework [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], and Yang (2021) used Karnofsky Performance Status (KPS) for survival prediction [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAdditionally, many studies on QoC outcomes have addressed the potential for improving QoL. Kawashima (2024) concluded that earlier identification of palliative-care needs via ML \"has the potential to contribute to earlier palliative care\" and thereby improve QoL whereas Bowles (2024) briefly mentions in the abstract that the narratives created from generative AI have \"the potential to support clinical decision-making\" and \"improve quality of life\" by enabling timelier needs assessment.\u003c/p\u003e \u003cp\u003e \u003cem\u003eQoL instruments used were\u003c/em\u003e:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eFunctional Assessment of Cancer Therapy General (Kamdar 2024) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGeneralized Anxiety Disorder (Kamdar 2024) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eBrief Pain Inventory (Kamdar 2024) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHRQoL (Haleem 2025) [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEORTC Quality of Life Questionnaire Core-30 (EORTC HRQoL C30) (Haleem 2025) [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eHospital Anxiety and Depression Scale (Haleem 2025) [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIPOS (Dong 2024, Haleem 2025) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eKPS (Yang 2021) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePain NRS (Kawashima 2024) [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eQoC\u003c/h2\u003e \u003cp\u003eAll 15 studies (100%) reported QoC-related outcomes [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Of these, 14 measured it quantitatively [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] and one was primarily a case-study-style evaluation without quantitative clinical effect endpoints [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn intervention-oriented studies, QoC outcomes were frequently process and utilisation endpoints, such as serious-illness conversation rates [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], Advance-care planning (ACP) documentation rates [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], palliative care consultation rates [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] and service utilisation outcomes (e.g. readmissions, Intensive care unit (ICU) transfer and length of stay) [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In contrast, algorithm development and validation studies have emphasised discrimination performance metrics (e.g. area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, accuracy and confusion metrics) [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], and technology validation studies have reported agreement metrics (e.g. Bland\u0026ndash;Altman, mean error, inter-rater reliability and validity coefficients) [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Finally, studies evaluating clinician-facing systems used usability and expert judgment endpoints (e.g. System Usability Scale (SUS), Short User Experience Questionnaire \u0026ndash; Short (UEQ-S) and expert ratings) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAdditionally, some technology-focused studies have explicitly framed QoC benefits in terms of care processes. Grie\u0026szlig;hammer (2024) evaluated contactless monitoring to enhance routine symptom management, whereas Dong (2025) suggested that automated voice-PROM extraction could reduce the clinician documentation burden.\u003c/p\u003e \u003cp\u003e \u003cem\u003eQuality of Care instruments identified include\u003c/em\u003e:\u003c/p\u003e \u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003ePalliative Prognostic Index (Yang 2021) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eEpic Systems electronic medical record ACP form entries (Epic, Verona, WI); National Quality Forum end-of-life quality measures (Gensheimer 2022) [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePalliative care consultation; service utilisation outcomes (e.g. 30-, 60-, and 90-day readmission, ICU transfer and inpatient length of stay) (Wilson 2023) [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eVoice recognition rate (Dong 2024) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePredictive performance metrics (area under the curve, F1, precision, recall, specificity, sensitivity, accuracy and confusion metrics) (Kawashima 2024, Dong 2024, Yang 2021, Liu 2023, Nejatifar 2025) [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eExpert Likert scale ratings of overall treatment proposal, evidence/guideline conformity and applicability of recommendations (Braun 2024) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSUS (Blanes-Selva 2023) [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eShort UEQ-S (Blanes-Selva 2023) [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eRate of documented serious-illness conversations (Manz 2023) [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eRate of ACP documentation (Gensheimer 2022) [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eRate of palliative-care referral (Wilson 2023) [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePain intensity score (0\u0026ndash;10) (Kamdar 2024) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eInter-rater reliability/validity coefficients (correlation coefficient, ICC, Cohen's kappa, and Gwet's AC) (Sampson 2025) [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eAgreement Analysis (Bland-Altman plot, mean error) (Grie\u0026szlig;hammer 2024) [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eMinimal Documentation System for routine daily symptom assessment (Grie\u0026szlig;hammer 2024) [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e \u003cp\u003eHealth related outcomes\u003c/p\u003e \u003cp\u003eAcross the 15 studies, AI applications demonstrated predominantly positive or neutral-to-positive effects on the measured outcomes with no clear negative effects reported (see Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The strength of this inference varied according to the study type. Interventional studies tended to report clinically interpretable processes or utilisation endpoints [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], whereas algorithm development and validation studies largely report proxy technical metrics (e.g. discrimination, agreement and reliability) rather than downstream patient benefit [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. When statistical testing was reported, results were often significant (p \u0026lt; .05), especially in intervention studies and some validations [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Several technical studies have described performance descriptively with limited or no hypothesis testing [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Overall, the findings are promising, but at an early stage. Many studies have shown that this model can work rather than improving care.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEffects of AI and direction of effect from the included studies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStudy (first author, year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDomain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOutcomes measured or discussed\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eInstrument/method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDirection of effect\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eEffect size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eStatistical technique \u0026amp; result\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eStatistical significance\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eComparison to usual care\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHaleem, 2025\u003c/b\u003e [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQuality of Life (QoL)/Quality of Care (QoC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHRQoL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMultimodal deep-learning regression for estimating HRQoL from smartwatch data and PROMs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMAPE\u0026thinsp;=\u0026thinsp;0.242\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLinear regression, prediction error, Spearman correlation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNo clinical effect test; Only p \u0026lt; .05 / .01 / .001 for Spearman correlation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKamdar, 2024\u003c/b\u003e [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoL/QoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePain (NRS), QoL domain scores\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDigital therapeutic application (ePAL) for pain management vs control\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMean pain scores: 2.99 (ePAL) vs 4.05 (Control)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDescriptive statistics; two-sample t-tests, Chi-square, mixed linear effect models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep \u0026lt; .05; pain improved (significant); QoL domains did not differ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePatients with ePAL were associated with lower pain and fewer pain-related hospitalisations\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBlanes-Selva 2023\u003c/b\u003e [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUsability \u0026amp; user experience of palliative clinical decision support systems (CDSS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSUS and UEQ-S\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSUS\u0026thinsp;=\u0026thinsp;62.7 (Round 1) / 65.0 (Round 2); UEQ-S\u0026thinsp;=\u0026thinsp;1.42 (Round 1) / 1.5 (Round 2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDescriptive statistics; t-test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep \u0026lt; .05; round-to-round differences were not significant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBowles, 2024\u003c/b\u003e [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeasibility of AI-generated summaries for supporting palliative-care needs assessment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAI-generated narrative summaries from administrative claims data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eQualitative focus group analysis from three palliative-care team members\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAI provides sufficient assessment for palliative-care needs but is not adequate for direct clinical recommendation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYang, 2021\u003c/b\u003e [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSurvival outcomes via physical activity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLSTM network using wearable actigraphy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.7292\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClassification metrics, comparison with QoL-related indicators (KPS/Palliative Prognostic Index (PPI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eThe model underperforms KPS and PPI; however, it does not require clinician input.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKawashima, 2024\u003c/b\u003e [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePredicting specialist palliative-care needs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost using EHR-derived variables, patient-reported pain, and QoL-related indicators\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClassification performance metrics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eNejatifar, 2025\u003c/b\u003e [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFrailty-based prediction of palliative-care needs in chronic obstructive pulmonary disease\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSuper Learner model using demographics and frailty conditions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClassification performance metrics, chi-square and t-tests for relationships, logistic prediction for predictors\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eManz, 2023\u003c/b\u003e [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSIC rate; end-of-life (EOL) systemic therapy; EOL utilisation outcomes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMachine learning (ML) mortality prediction via EHR data; behavioural nudge to prompt SICs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSICs: aOR 2.09; Reduced EOL systemic therapy: aOR 0.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDescriptive statistics, generalized estimating equations, subgroup interaction analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eBehavioural nudges to clinicians led to an increase in SICs and reduction in end-of-life systemic therapy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGensheimer, 2022\u003c/b\u003e [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eACP documentation frequency; EOL quality measures\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrognosis model to predict survival from EHR data; lay care coaches\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eACP documentation: OR 13.7; Prognosis documentation: OR 9.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMixed-effects logistic regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCombining a prognosis model with care coaches increased ACP documentation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eWilson, 2023\u003c/b\u003e [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePalliative care consultation; readmissions; ICU transfer; length of stay\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eML CDSS tool via EHR data integrated into clinical workflow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePalliative care consultations: IRR 1.44; 60-day readmission: OR 0.75; 90-day readmission: OR 0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eBayesian estimation accounting for the stepped design; 95% Credible Intervals\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eIncreased palliative care rate consultations and reduction in hospitalisation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSampson, 2025\u003c/b\u003e [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eReliability and validity of a pain assessment tool\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePainChek for automated pain assessment from facial analysis, voice, movement, behaviour activity and body\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eIRR 0.714 (rest), 0.817 (post-movement); Internal consistency 0.755 (rest), 0.833 (post-movement); Concurrent validity with PAINAD: 0.528 (rest), 0.787 (post-movement); Convergent validity with SM_EOLD\u0026thinsp;\u0026minus;\u0026thinsp;0.555 (rest), -0.544 (post-movement)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eInter-rater analysis, internal consistency, concurrent validity; comparison with Abbey Pain Scale (APS), PAINAD and SM-EOLD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePainChek compares favourably with APS.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGriebhammer, 2024\u003c/b\u003e [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFeasibility of radar-based heart rate monitoring in palliative care\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eContinuous-wave radar units installed underneath the head section of the slatted frame for contactless monitoring\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;Positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMean Difference: Arm I\u0026thinsp;\u0026minus;\u0026thinsp;0.852 beats per minute (bpm), ARM II\u0026thinsp;\u0026minus;\u0026thinsp;0.678 bpm\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eConcordance analysis; Bland-Altman; two one-sided test (TOST)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eModerate agreement when compared with electrocardiogram\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDong 2024\u003c/b\u003e [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAccuracy of patient-reported pain symptom detection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eML to extract patient-reported pain symptoms from transcribed patient interviews\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eVoice recognition rate 55.6%;\u003c/p\u003e \u003cp\u003eF1 0.31 \u0026minus;\u0026thinsp;0.46 over five symptoms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eVoice recognition rate; classification metrics; Kruskal\u0026ndash;Wallis test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003ep\u0026thinsp;=\u0026thinsp;0.515 for differences in recognition rate across disease groups\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eModerate performance compared with human-transcribed Integrated Palliative care Outcome Scale\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLiu, 2023\u003c/b\u003e [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7-day mortality prediction\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eML to predict 7-day mortality from clinical assessment and wearable data\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.960\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eClassification metrics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBraun, 2024\u003c/b\u003e [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eQoC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExpert rating of ChatGPT therapy suggestions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eChatGPT to provide therapy suggestions on 10 patient case vignettes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeutral\u0026ndash;positive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDescriptive statistics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eN/A\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNo comparison to usual care\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe evidence is clearest in studies that compared with usual care. Three RCTs showed superiority over usual care for clinically meaningful outcomes: improved serious-illness conversation documentation with reduced end-of-life systemic therapy [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], increased palliative-care consultation activity with fewer readmissions [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], and reduced pain intensity with fewer pain-related hospitalisations [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBy contrast, algorithm development and validation studies typically report technical or procedural proxies. Prognostic and palliative care needsidentification models frequently reported good-to-excellent AUROC values and high sensitivity [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]; however, these metrics were rarely linked to demonstrated changes in care delivery. Survival prediction is feasible with moderate prognostic accuracy or agreement relative to a standard functional status tool [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Technology validation studies emphasise agreement, accuracy and reliability against reference standards [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], whereas clinician-facing systems emphasise usability and expert judgment endpoints rather than patient outcomes [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePractical, technical and ethical consequences of applying AI in palliative care\u003c/p\u003e \u003cp\u003eThis section synthesises both the explicit considerations raised by the authors of the 15 included studies and the implicit issues identified by the review team. Although promising, the application of AI in palliative care is accompanied by substantial practical, technical and ethical challenges that must be addressed before its routine clinical adoption.\u003c/p\u003e \u003cp\u003ePractical considerations\u003c/p\u003e \u003cp\u003eIntegration into existing clinical workflows is a major barrier. Implementation-oriented work explicitly highlighted the need to avoid creating additional burdens (e.g. adding extra electronic health record (EHR) alerts) [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], with one study specifically designing an intervention to abrogate the potential for alert fatigue by avoiding a best-practice alert [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Clinician-facing tools have also raised practical concerns about how outputs fit into daily work [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], and usability studies have emphasised the need for systems that support professional autonomy within real clinical workflows [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Overall, AI systems should not disrupt routine practice or require additional interpretation time from already overburdened palliative-care teams [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Palliative clinicians have also reported severe staffing shortages and \u0026ldquo;no spare capacity\u0026rdquo; to learn or troubleshoot new systems [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFurthermore, wearable device studies highlight patient- and unknown-level burdens, including discomfort, device management, and data upload compliance [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. In actigraphy-based survival prediction, patients were instructed to wear the device throughout admission, except during showering, because it was not water-resistant, and the data required periodic uploads [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. In smartwatch-based mortality prediction, participants needed to operate smartphone syncing; some days\u0026rsquo; data were excluded when uploads were absent, and user feedback reported charging problems and minor skin irritation [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMethodologically, many of the included studies were small, single-centre feasibility or validation projects with limited sample sizes, restricting generalisability and real-world scalability. Interdisciplinary co-design, iterative piloting, and dedicated staff training are essential prerequisites for acceptability; however, these elements are present in only a few studies [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTechnical considerations\u003c/p\u003e \u003cp\u003eModels rely heavily on structured EHRs or claims data of variable quality, suffer from class imbalances, and exhibit limited generalisability beyond their original institutions and condition cohorts [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Moreover, class imbalance is common (e.g. a low prevalence of palliative care needs or mortality events) and is typically addressed by undersampling or ensemble methods, thereby reducing effective training data [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRegarding generative AI, LLM-based approaches raise risks around incorrect or unsafe content generation and the need for careful human oversight [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], while narrative-summary approaches also raised pragmatic concerns about the nature of outputs (e.g., length/fit-for-workflow and confidentiality) that affect real-world utility [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Deep-learning systems can remain difficult to interpret; however, some studies used explainability approaches such as SHapley Additive exPlanations (SHAP) analyses for erroneous predictions [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and others incorporated \u0026ldquo;explanatory\u0026rdquo; design features intended to preserve clinicians\u0026rsquo; professional autonomy [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFinally, data requirements can be nontrivial; in wearable actigraphy survival prediction, performance was better when using 48 h of input data compared to 24 h [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], and wearable-based studies depended on sustained device use and data completeness (e.g. synchronisation and handling missingness) [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Validation in several studies ends with technical performance metrics rather than sustained clinical monitoring and downstream outcomes, limiting its usefulness for implementation decisions.\u003c/p\u003e \u003cp\u003eEthical considerations and mitigation strategies\u003c/p\u003e \u003cp\u003e Ethical engagement specific to AI in the included studies was strikingly limited and was typically confined to baseline research governance, with bias, autonomy, and equity rarely addressed in depth. For instance, prospective wearable work reported Institutional Review Board approval and informed consent [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]; however, explicit discussions of \u003cem\u003eAI-specific\u003c/em\u003e consent, communication with participants and equity/fairness implications were generally limited in reporting across study types.\u003c/p\u003e \u003cp\u003eWhen mapped against the World Health Organization\u0026rsquo;s (WHO) six governing principles for AI in health (2021) [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e] and the UKRIO\u0026rsquo;s practical guide for AI researchers (2025) [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], the evidence reveals major gaps that must be addressed before widespread adoption in high-stakes, vulnerability-intensive fields such as palliative and end-of-life care:\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePrinciple 1 – Protecting human autonomy\u003c/h3\u003e\n\u003cp\u003eMultiple studies have highlighted clinicians\u0026rsquo; concerns that AI can diminish clinical judgement or the human elements of care [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. This concern is justified, as any perception that decisions are delegated to an algorithm risks undermining trust and the therapeutic relationship between HCPs and patients. Informed consent specifically for AI processing of data is rarely described; in wearable studies, this may be implied rather than explicitly stated.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003eHuman-in-the-loop must be mandatory with explicit clinician override and explicit tiered consent for AI use, with patients retaining the right to opt for human-only care pathways. These strategies aim to preserve the doctor-patient relationship, which is the cornerstone of palliative care and ensure that AI remains a supportive tool rather than a replacement for human compassion and care.\u003c/p\u003e \u003c/p\u003e\n\u003ch3\u003ePrinciple 2 – Promoting human well-being, safety and the public interest\u003c/h3\u003e\n\u003cp\u003eHigh-sensitivity predictive models risk causing unnecessary distress through false positives or premature palliative-care discussions if poorly governed [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Although some studies have used firewalls and internal processing [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], none have reported formal Data Protection Impact Assessments or compliance with the WHO-recommended minimum necessary data principles [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. This is problematic because inadequate privacy safeguards risk re-identification of sensitive end-of-life data. Specifically, participants in wearable studies [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] were exposed to re-identification risks through continuous behavioural data.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003eContext-specific thresholds, continuous post-launch surveillance, rapid error reporting, data minimisation, pseudonymisation, and independent security audits should be implemented. These suggestions ensure that participants\u0026rsquo; data privacy is safeguarded, and ongoing mechanisms are in place to report potential data breaches.\u003c/p\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePrinciple 3 \u0026ndash; Ensuring transparency, explainability and intelligibility\u003c/h2\u003e \u003cp\u003eDespite some explainability attempts (e.g. SHAP analyses) [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], most studies deploy black-box models without patient-facing explanations or documentation of limitations. The black-box nature inherently limits HCPs\u0026rsquo; trust in the modules and goes against patients\u0026rsquo; right to understand how decisions affecting their care are made.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003ePatient-facing plain language explanations, validated explanation strategies (e.g. SHAP/ Local Interpretable Model-agnostic Explanations where appropriate), and transparent documentation of the model\u0026rsquo;s intended use, limitations, and known failure modes.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eThese strategies ensure that patients and their families meaningfully engage in shared decision-making regarding their care.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePrinciple 4 \u0026ndash; Fostering responsibility and accountability\u003c/h2\u003e \u003cp\u003eNo study has reported a formal ethical review of AI components or established accountability mechanisms. It can be inferred that the relevant ethics committee approved the studies; however, AI-specific oversight could be further contested.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003ePalliative-specific AI governance committees with patient/family representation and public registration of health AI systems are required. These structures are necessary to ensure that accountability is shared among developers, clinicians, and institutions, thereby protecting vulnerable patients from unaddressed harm.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePrinciple 5 \u0026ndash; Ensuring inclusiveness and equity\u003c/h2\u003e \u003cp\u003eFourteen studies originated from high- or upper-middle-income settings with limited ethnic, socioeconomic, or geographic diversity, potentially widening the existing palliative care inequities in underserved populations. Certain study designs such as wearable-heavy approaches intentionally exclude patients with poor digital literacy or physical limitations [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], perpetuating digital exclusion in this vulnerable and seldom-heard-of group.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003eBias audits with fairness metrics, diverse multinational training data, community co-design, and low-tech alternatives with accessibility testing are required, all of which are needed to prevent AI from exacerbating inequities in palliative care.\u003c/p\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003ePrinciple 6 \u0026ndash; Promoting AI that is responsive and sustainable\u003c/h2\u003e \u003cp\u003eNone of the studies mentioned environmental concerns, and the long-term responsiveness and sustainability of wearable devices remain unaddressed. Unsustainable designs increase ecological burden and exclude patients in resource-constrained settings, contradicting the holistic ethos of palliative care.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003eEnvironmental impact assessments and designs that are viable across resource settings must be incorporated. This is appropriate, because palliative care should consider broader societal and planetary contexts.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eCross-cutting issue \u0026ndash; Absence of informed consent for AI use (Principles 1, 2, and 3)\u003c/b\u003e: Retrospective data studies [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] and wearable devices routinely fail to describe specific consent for AI processing. This could be implied as an ethics committee waiver for secondary analysis; however, this was not explicitly stated. Without explicit consent for AI use, participants are denied agency over the use of sensitive data, potentially violating their autonomy.\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eMitigation\u003c/strong\u003e \u003cp\u003e Explicit, tiered consent or valid ethics-committee waiver is mandatory.\u003c/p\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e summarises the identified ethical considerations and proposed mitigation strategies.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIdentified ethical considerations and proposed mitigation strategies\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEthical considerations\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWHO (2021) Principle\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMitigation strategies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBias, discrimination, marginalisation and digital exclusion from:\u003c/p\u003e \u003cp\u003e- Homogeneous datasets\u003c/p\u003e \u003cp\u003e- Wearable/application barriers\u003c/p\u003e\u003cp\u003e- Developer biases\u003c/p\u003e\u003cp\u003e- Bias in deployment\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5 \u0026ndash; Inclusiveness and equity\u003c/p\u003e \u003cp\u003e6 \u0026ndash; Promote artificial intelligence that is responsive and sustainable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Bias audits and fairness metrics; diverse multinational data; community co-design\u003c/p\u003e \u003cp\u003e- Protecting/minimising data colonialism\u003c/p\u003e \u003cp\u003e- Early and ongoing stakeholder engagement\u003c/p\u003e \u003cp\u003e- Low-tech alternatives\u003c/p\u003e \u003cp\u003e- Accessibility testing with seldom-heard-of populations\u003c/p\u003e\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInadequate privacy/data protection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 \u0026ndash; Promote human well-being, human safety and the public interest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Mandatory data protection impact assessment\u003c/p\u003e \u003cp\u003e- Data minimisation\u003c/p\u003e \u003cp\u003e- Pseudonymisation\u003c/p\u003e\u003cp\u003e- Independent security audit\u003c/p\u003e\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOver-reliance on artificial intelligence (AI), erosion of autonomy and potential absence of informed consent for AI use\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 \u0026ndash; Protect autonomy\u003c/p\u003e \u003cp\u003e6 \u0026ndash; Promote AI that is responsive and sustainable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Human-in-the-loop as mandatory mechanism\u003c/p\u003e \u003cp\u003e- \u0026lsquo;Graceful transition\u0026rsquo; period and training for staff\u003c/p\u003e \u003cp\u003e- Explicit tiered consent for AI processing\u003c/p\u003e \u003cp\u003e- Right to opt for human-only pathway\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLack of transparency and explainability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 \u0026ndash; Ensure transparency, explainability and intelligibility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Patient-facing explanations\u003c/p\u003e \u003cp\u003e- publish model cards\u003c/p\u003e\u003cp\u003e- Validated explainability modules\u003c/p\u003e\u003cp\u003e- Regular AI Audit\u003c/p\u003e\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHarm from false positives/negatives\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 \u0026ndash; Promote human well-being, human safety and the public interest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Context-specific thresholds\u003c/p\u003e \u003cp\u003e- Continuous AI performance surveillance\u003c/p\u003e \u003cp\u003e- Rapid-error reporting process\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo formal governance or accountability\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 \u0026ndash; Foster responsibility and accountability\u003c/p\u003e \u003cp\u003e6 \u0026ndash; Promote AI that is responsive and sustainable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e- Palliative-specific AI governance committee; public registration of systems\u003c/p\u003e \u003cp\u003e- Impact assessment at protocol stage\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis scoping review identified three primary domains of AI application in palliative care: (1) early identification of palliative care needs through prognostic modelling (n\u0026thinsp;=\u0026thinsp;9), (2) symptom assessment and relief using sensor-based and multimodal tools (n\u0026thinsp;=\u0026thinsp;6) and (3) clinical decision support for care conversations via nudges and generative models (n\u0026thinsp;=\u0026thinsp;4), with some studies contributing to more than one domain. These align with broader trends in the field where predictive analytics have dominated early efforts, particularly from 2021\u0026ndash;2023, reflecting the maturation of ML techniques for mortality forecasting and palliative care identification from EHRs and related structured data [ 20, 21, 23, 28, 29, 33, 34]. A bibliometric analysisreports of this shift reported that early AI research (pre-2023) focused on prognostication to address late referrals, with annual outputs rising from a few papers to a peak of 66 in 2024, driven by multimodal integration [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe surge in symptom assessment applications in 2024\u0026ndash;2025 mirrored the rapid adoption of wearable and voice technologies following the COVID-19 pandemic. This is evident from a scoping review that highlighted the role of DL in real-time HRQoL estimation and pain detection in publications increasing since 2023 [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Nevertheless, decision support remains consistent, but niche, often leveraging LLMs such as ChatGPT for narrative generation [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], which is consistent with the wider literature on generative AI's potential for communication aids [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOverall, the convergence across domains, from siloed prediction to integrated systems, parallels global trends toward \"explainable AI,\" which emphasises hybrid models for end-of-life personalisation [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. However, with most studies originating from Global North settings, these innovations risk irrelevance in low-resource contexts, where new technologies, such as low-cost radar monitoring [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], show promise but require Southern-led validation.\u003c/p\u003e \u003cp\u003eRegarding the effectiveness on QoL and QoC, AI demonstrated positive effects across the measured outcomes, with moderate to large effect sizes and consistent statistical significance. Three RCTs reported superiority to usual care in conversation initiation, referrals, and pain outcomes ( Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDirect measurement of QoL is rare [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], but indirect benefits via earlier interventions have been noted, aligning with the literature showing that AI-enhanced symptom management can improve global health scores on scales such as the EORTC Quality of Life Questionnaire Core-30 [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Wider evidence supports these findings. Reviews of ML in palliative care found that predictive tools reduced high-intensity end-of-life treatments, correlating with better QoL [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], while bibliometric trend projects continued to grow in AI for personalised comfort [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNevertheless, QoC outcomes dominated (100% of studies), with strong evidence of usability (SUS scores 62\u0026ndash;65) (see Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) and process improvements (e.g. an 11% absolute referral increase) (see Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), echoing a 2025 systematic review that attributes such gains to routine data integration and enhanced resource allocation in understaffed settings [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. The emphasis on QoC over QoL reflects a broader skew in the literature, which limits generalisability [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. This QoC bias risks overlooking holistic palliative goals, where AI's true value lies in augmenting rather than just streamlining care.\u003c/p\u003e \u003cp\u003eThe identified practical considerations include workflow disruption, alert fatigue, and resource demands, mirroring the wider challenges in AI implementation. Integration issues such as time burdens from AI outputs [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] align with a recent review which reported that the implementation of palliative AI is stalled by clinician overload, particularly in underresourced contexts [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Furthermore, inherent barriers to wearable compliance [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], namely high dropout rates due to discomfort in frail patients and usability, underscore the need for co-design [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Small, single-centre studies further constrain scalability, consistent with calls for interdisciplinary piloting to bridge the feasibility of practice [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRegarding technical considerations, technical limitations such as data quality variability, class imbalance, and poor generalisability [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] are recurrent patterns in the palliative AI literature. That is, overreliance on EHRs tends to lead to missing values, thereby reducing model accuracy [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Black-box issues persist despite SHAP\u0026rsquo;s attempts [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], mirroring the wider literature in which interpretability remains challenging [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Moreover, infrastructure requirements [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and input length requirements [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] highlight scalability gaps to enhance robustness across sites [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSecond, identified ethical considerations such as bias from homogeneous data, unclear AI consent, and opacity can amplify vulnerabilities in palliative care research and delivery. Given that most of the included studies (93%) were from Global North settings, this skewed data underperforms in diverse populations [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. These ethical concerns can be addressed by embedding relevant AI ethical frameworks, such as those from the WHO or UKRIO, to ensure that AI use by participants is appropriately safeguarded [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOverall, this scoping review maps AI's evolving role of AI in palliative care, but reveals an immature, northern-centric evidence base. Future work must prioritise diverse participatory trials to realise equitable benefits guided by more robust AI ethical frameworks.\u003c/p\u003e \u003cp\u003eStrengths of the study\u003c/p\u003e \u003cp\u003e We included a wide range of study designs to provide a comprehensive overview of real-world data on AI applications in palliative care settings. The methodology was rigorous with a prospectively registered protocol, dual-independent screening, and structured data extraction with verification at both the title/abstract and full-text stages, along with regular team discussions for robust interpretation.\u003c/p\u003e \u003cp\u003e The review team was deliberately diverse and comprised two palliative care clinicians/researchers and two computer engineers with AI expertise. Three members are based in the Global South (Thailand) and one is in the Global North, offering a more balanced perspective on the opportunities and equity implications of this emerging technology in different resource contexts.\u003c/p\u003e \u003cp\u003eWe also identified equity, diversity, and inclusion issues, including the digital divide, limited access to technology in low-resource settings, and the risk that most models are trained predominantly on data from high-income, white-majority populations, which warrant urgent further exploration in future research to prevent data colonisation [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFinally, we systematically outlined practical, technical and ethical considerations against established frameworks [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], thus providing a clear roadmap for responsible AI development in palliative care.\u003c/p\u003e \u003cp\u003eLimitations\u003c/p\u003e \u003cp\u003eThis review has several limitations:\u003c/p\u003e \u003cp\u003eFirst, the searches were limited to studies published in English and in five major databases; relevant research in other languages or unpublished/grey literature may have been missed.\u003c/p\u003e \u003cp\u003e Second, consistent with the JBI scoping review guidelines, no formal critical appraisal of methodological quality or risk of bias was performed; therefore, readers should interpret effectiveness claims cautiously, particularly given the predominance of small, single-centre studies. Moreover, no meta-analysis was conducted owing to substantial heterogeneity in study designs, AI modalities, and outcome measures, which is an expected limitation of scoping reviews and is fully in line with JBI recommendations.\u003c/p\u003e \u003cp\u003eThird, the evidence is heavily skewed toward Global North institutions (93%), restricting insights into their applicability in low- and middle-income settings.\u003c/p\u003e \u003cp\u003eFinally, studies with negative or null results were frequently underreported [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. For the quantitative and mixed-methods studies included here, publication bias may have favoured more positive findings, potentially influencing our overall interpretation. However, we have remained impartial and reported the data exactly as presented by the original authors. A practical suggestion beyond this review is to encourage routine publication or registration of all trial results [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], including negative findings, as these provide invaluable insights for the safe development of AI in palliative care.\u003c/p\u003e \u003cp\u003eImplications for future studies\u003c/p\u003e \u003cp\u003eTherefore, additional qualitative and codesigned studies are required. None of the 15 included studies used purely qualitative methods, and patient or caregiver voices were almost entirely absent beyond providing physiological or voice data. In-depth interviews, focus groups, or experience-based co-design approaches could provide richer insights into how patients, families, and clinicians perceive and interact with AI tools in the emotionally charged context of palliative care. Such studies would help ensure that future technologies remain person-centred rather than clinician- or data-driven.\u003c/p\u003e \u003cp\u003eMixed-methods designs should be prioritised to capture both measurable effectiveness (e.g. referral rates and prognostic accuracy) and contextual factors, such as cultural acceptability, trust, and relational impact. This strengthens the evidence base and improves real-world applicability.\u003c/p\u003e \u003cp\u003eTherefore, there is an urgent need to improve geographic diversity. Fourteen of the fifteen studies originated in the Global North, with only one from Iran representing the Global South. This imbalance limits our understanding of how cultural norms, varying levels of digital literacy, health-system infrastructure, and resource constraints shape AI implementation. Future research should actively involve partners and study sites from low- and middle-income countries to produce more contextually grounded, equitable AI solutions for palliative care worldwide.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eAI demonstrates clear potential to enhance palliative care by enabling earlier identification of needs, supporting real-time symptom assessments, and facilitating decision-making and care conversations. The 15 studies in this review showed generally favourable findings in prognostic accuracy, referral rates, conversation initiation, and pain management compared to that of usual care, with emerging evidence of positive effects on HRQoL. This field is rapidly maturing and moving from isolated predictive models to integrated multimodal systems. However, the current evidence base remains small, predominantly from Global North settings, and is limited in its focus on patient-reported outcomes and ethical governance.\u003c/p\u003e \u003cp\u003eTo translate these promises into an equitable, patient-centred reality, future research should prioritise larger, geographically diverse studies with explicit HRQoL endpoints, qualitative and co-design approaches that centre patient and caregiver experience, active collaboration with Global South partners, and systematic application of AI ethical frameworks from the outset. When responsibly developed, AI can significantly strengthen the compassionate and holistic delivery of palliative care.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eACP \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Advance-care planning\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAI\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eAUROC \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Area Under the Receiver Operating Characteristic curve\u003c/p\u003e\n\u003cp\u003eCDSS \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Clinical Decision Support Systems\u003c/p\u003e\n\u003cp\u003eCOPD Chronic Obstructive Pulmonary Disease\u003c/p\u003e\n\u003cp\u003eDL\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Deep Learning\u003c/p\u003e\n\u003cp\u003eEHR \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Electronic Health Record\u003c/p\u003e\n\u003cp\u003eGBM\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Gradient Boosting Machine\u003c/p\u003e\n\u003cp\u003eHCPs \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Health Care Professionals\u003c/p\u003e\n\u003cp\u003eHRQoL \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Health-Related Quality of Life\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eICU \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Intensive care unit\u003c/p\u003e\n\u003cp\u003eIPOS \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Integrated Palliative Care Outcome Scale\u003c/p\u003e\n\u003cp\u003eJBI \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Joanna Briggs Institute\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eKPS \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Karnofsky Performance Status\u003c/p\u003e\n\u003cp\u003eLLM \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Large Language Model\u003c/p\u003e\n\u003cp\u003eLSTM\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Long Short-Term Memory\u003c/p\u003e\n\u003cp\u003eML\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Machine Learning\u003c/p\u003e\n\u003cp\u003eNLP \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Natural Language Processing\u003c/p\u003e\n\u003cp\u003eNRS \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Pain Numerical Rating Scale\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePROM \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Patient-Reported Outcome Measures\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eQoC\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Quality of Care\u003c/p\u003e\n\u003cp\u003eQoL \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Quality-of-Life\u003c/p\u003e\n\u003cp\u003eRCT \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Randomized Controlled Trial\u003c/p\u003e\n\u003cp\u003eSHAP \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;SHapley Additive exPlanations\u003c/p\u003e\n\u003cp\u003eSUS \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;System Usability Scale\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUEQ-S User Experience Questionnaire- Short\u003c/p\u003e\n\u003cp\u003eUKRIO \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;United Kingdom Research Integrity Office\u003c/p\u003e\n\u003cp\u003eWHO \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;World Health Organization\u003c/p\u003e\n\u003cp\u003eXGBoost \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Extreme Gradient Boosting\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eEthics approval and consent to participate\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eConsent for publication\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAvailability of data and materials\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated or analysed during this study are included in this published article and its supplementary information files.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eCompeting interests\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFunding\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo funding was secured for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAuthors' contributions\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTRP conceptualised the study, developed the methodology, performed software and validation tasks, conducted formal analysis and investigation, curated resources and data, prepared the original draft, reviewed and edited the manuscript, visualised results, supervised the project, and managed administration. SC conducted formal analysis and investigation, curated data, and contributed to reviewing and editing the manuscript. KT performed validation, formal analysis, investigation, data curation, contributed to reviewing and editing the manuscript, and visualised results. TNP conducted formal analysis and investigation, curated data, and contributed to reviewing and editing the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAcknowledgements\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to express our gratitude to Scott McGregor for his assistance with the search terms and processes.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWorld Health Organization (WHO). Palliative care. 2020. https://www.who.int/news-room/fact-sheets/detail/palliative-care. Accessed 18 Nov 2025.\u003c/li\u003e\n\u003cli\u003eMalhotra C, Shafiq M, Batcagan‑Abueg APM. What is the evidence for efficacy of advance care planning in improving patient outcomes? A systematic review of randomised controlled trials. BMJ Open. 2022; doi:10.1136/bmjopen-2021-060201\u003c/li\u003e\n\u003cli\u003eTripodoro VA, Fidalgo JFL, Pons JJ, Connor SR, Garralda E, Bastos F, et al. First‑ever global ranking of palliative care: 2025 World Map under the new WHO framework. J Pain Symptom Manage. 2025; doi:10.1016/j.jpainsymman.2025.07.026\u003c/li\u003e\n\u003cli\u003ePitzer S, Kutschar P, Paal P, M\u0026uuml;lleder P, Lorenzl S, Wosko P, et al. Barriers for adult patients to access palliative care in hospitals: a mixed methods systematic review. J Pain Symptom Manage. 2024;67(1):e16\u0026ndash;e33.\u003c/li\u003e\n\u003cli\u003ePhenwan T, Anantapong K, Sripaew S, Vinayavekhin S. Exclusive, isolated and Buddhist‑centric at the expense of non‑conformists: a scoping review and bibliometric analysis of the palliative care ecosystem and its influence on quality of life, care and death. Advance. 2025.\u003c/li\u003e\n\u003cli\u003eTopol EJ. High‑performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44\u0026ndash;56.\u003c/li\u003e\n\u003cli\u003eSilva‑Ferreira M, Cruz S, Lu\u0026iacute;s MS, Silva MC, Monteiro‑Reis S, Henrique R, et al. Mapping study on AI‑based technologies in palliative care \u0026ndash; a scoping study. BMC Palliat Care. 2025; doi:10.1186/s12904-025-01909-w.\u003c/li\u003e\n\u003cli\u003eArioz U, Allsop MJ, Goodman WD, Timmons S, Simbirtseva K, Mlakar I, et al. Artificial intelligence‑based approaches for advance care planning: a scoping review. BMC Palliat Care. 2025; doi:10.1186/s12904-025-01827-x.\u003c/li\u003e\n\u003cli\u003ePinto A, Santos C, Aguiar R, Oliveira S, Cunha D. The use of artificial intelligence in palliative care communication: a narrative review. Cureus. 2025; doi:10.7759/cureus.80524.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization (WHO). Ethics and governance of artificial intelligence for health: WHO guidance. 2021. https://www.who.int/publications/i/item/9789240029200. Accessed 18 Nov 2025.\u003c/li\u003e\n\u003cli\u003eNikoloudi M, Mystakidou K. Artificial intelligence in palliative care: a scoping review of current applications, challenges, and future directions. Am J Hosp Palliat Care. 2025; doi:10.1177/10499091251358379.\u003c/li\u003e\n\u003cli\u003eHolm S, Ploug T. Co‑reasoning and epistemic inequality in AI supported medical decision‑making. Am J Bioeth. 2024;24(9):79\u0026ndash;80.\u003c/li\u003e\n\u003cli\u003eBozkurt S, Fereydooni S, Kar I, Chalmers CD, Leslie SL, Pathak R, et al. AI in palliative care: a scoping review of foundational gaps and future directions for responsible innovation. J Pain Symptom Manage. 2025;70(6):e394\u0026ndash;e418.\u003c/li\u003e\n\u003cli\u003eUK Research Integrity Office (UKRIO). Embracing AI with integrity: a practical guide for researchers. 2025. https://ukrio.org/ukrio-resources/embracing-ai-with-integrity/. Accessed 29 Jul 2025\u003c/li\u003e\n\u003cli\u003eKhalil H, Peters MDJ, Tricco AC, Pollock D, Alexander L, McInerney P, et al. Conducting high quality scoping reviews: challenges and solutions. J Clin Epidemiol. 2021;130:156\u0026ndash;160.\u003c/li\u003e\n\u003cli\u003ePhenwan T, Sitthichok C, Tangudomkit K, Peerawong T. The use of artificial intelligence in palliative care: a scoping review protocol. OSF. 2025.\u003c/li\u003e\n\u003cli\u003ePollock D, Peters MDJ, Khalil H, McInerney P, Alexander L, Tricco AC, et al. Recommendations for the extraction, analysis, and presentation of results in scoping reviews. JBI Evid Synth. 2023;21(3):520\u0026ndash;532.\u003c/li\u003e\n\u003cli\u003ePeters MDJ, Marnie C, Tricco AC, Pollock D, Munn Z, Alexander L, et al. Updated methodological guidance for the conduct of scoping reviews. JBI Evid Implement. 2021;19(1).\u003c/li\u003e\n\u003cli\u003eBowles KH, Brickner C, Luth EA. Using generative AI to translate administrative claims data into narrative summaries for palliative care needs assessment: a case study. Stud Health Technol Inform. 2024;315:547\u0026ndash;551.\u003c/li\u003e\n\u003cli\u003eGensheimer MF, Gupta D, Patel MI, Fardeen T, Hildebrand R, Teuteberg W, et al. Use of machine learning and lay care coaches to increase advance care planning conversations for patients with metastatic cancer. JCO Oncol Pract. 2023;19(2):e176\u0026ndash;e184.\u003c/li\u003e\n\u003cli\u003eManz CR, Zhang Y, Chen K, Long Q, Small DS, Evans CN, et al. Long‑term effect of machine learning‑triggered behavioural nudges on serious illness conversations and end‑of‑life outcomes among patients with cancer: a randomized clinical trial. JAMA Oncol. 2023;9(3):414\u0026ndash;418.\u003c/li\u003e\n\u003cli\u003eKamdar M, Jethwani K, Centi AJ, Agboola S, Fischer N, Traeger L, et al. A digital therapeutic application (ePAL) to manage pain in patients with advanced cancer: a randomized controlled trial. J Pain Symptom Manage. 2024;68(3):261\u0026ndash;271.\u003c/li\u003e\n\u003cli\u003eWilson PM, Ramar P, Philpot LM, Soleimani J, Ebbert JO, Storlie CB, et al. Effect of an artificial intelligence decision support tool on palliative care referral in hospitalized patients: a randomized clinical trial. J Pain Symptom Manage. 2023;66(1):24\u0026ndash;32.\u003c/li\u003e\n\u003cli\u003eBraun EM, Juhasz‑B\u0026ouml;ss I, Solomayer EF, Truhn D, Keller C, Heinrich V, et al. Will I soon be out of my job? Quality and guideline conformity of ChatGPT therapy suggestions to patient inquiries with gynaecologic symptoms in a palliative setting. Arch Gynecol Obstet. 2024;309(4):1543\u0026ndash;1549.\u003c/li\u003e\n\u003cli\u003eGrie\u0026szlig;hammer SG, Malessa A, Lu H, Yip J, Leuschner J, Christgau F, et al. Contactless radar‑based heart rate estimation in palliative care \u0026ndash; a feasibility study and possible use in symptom management. BMC Palliat Care. 2024;23(1):1\u0026ndash;14.\u003c/li\u003e\n\u003cli\u003eKawashima A, Furukawa T, Imaizumi T, Morohashi A, Hara M, Yamada S, et al. Predictive models for palliative care needs of advanced cancer patients receiving chemotherapy. J Pain Symptom Manage. 2024;67(4):306\u0026ndash;316.e6.\u003c/li\u003e\n\u003cli\u003eDong L, Hirayama H, Zheng X, Masukawa K, Miyashita M. Using voice recognition and machine learning techniques for detecting patient‑reported outcomes from conversational voice in palliative care patients. Jpn J Nurs Sci. 2025;22(1):e12644.\u003c/li\u003e\n\u003cli\u003eYang TY, Kuo PY, Huang Y, Lin HW, Malwade S, Lu LS, et al. Deep‑learning approach to predict survival outcomes using wearable actigraphy device among end‑stage cancer patients. Front Public Health. 2021;9:730150.\u003c/li\u003e\n\u003cli\u003eLiu JH, Shih CY, Huang HL, Peng JK, Cheng SY, Tsai JS, et al. Evaluating the potential of machine learning and wearable devices in end‑of‑life care in predicting 7‑day death events among patients with terminal cancer: cohort study. J Med Internet Res. 2023; doi:10.7759/cureus.80524.\u003c/li\u003e\n\u003cli\u003eHaleem MS, Aidonis V, Georga EI, Krini M, Matsangidou M, Kassianos AP, et al. A multimodal deep learning architecture for estimating health‑related quality of life for advanced cancer patients based on wearable devices and patient‑reported outcome measures. IEEE J Biomed Health Inform. 2026; doi: 10.1109/JBHI.2025.3597054.\u003c/li\u003e\n\u003cli\u003eNejatifar Z, Alizadeh A, Amerzadeh M, Omidian S, Rafiei S. The predictive role of identifying frailty in assessing the need for palliative care in the elderly: the application of machine learning algorithm. J Health Popul Nutr. 2025;44(1):133.\u003c/li\u003e\n\u003cli\u003eSampson EL, Davies N, Vickerstaff V. Evaluation of the psychometric properties of PainChek in older general hospital patients with dementia. Age Ageing. 2025;54(2):afaf027.\u003c/li\u003e\n\u003cli\u003eBlanes‑Selva V, Asensio‑Cuesta S, Do\u0026ntilde;ate‑Mart\u0026iacute;nez A, Mesquita FP, Garc\u0026iacute;a‑G\u0026oacute;mez JM. User‑centred design of a clinical decision support system for palliative care: insights from healthcare professionals. Digit Health. 2023;9:20552076231181212.\u003c/li\u003e\n\u003cli\u003ePan M, Huang R, Liu C, Xiong Y, Li N, Peng H, et al. Application of artificial intelligence in palliative care: a bibliometric analysis of research hotspots and trends. Front Med (Lausanne). 2025;12:1597195.\u003c/li\u003e\n\u003cli\u003eBressler T, Song J, Kamalumpundi V, Chae S, Song H, Tark A. Leveraging artificial intelligence/machine learning models to identify potential palliative care beneficiaries: a systematic review. J Gerontol Nurs. 2025;51(1):7\u0026ndash;14.\u003c/li\u003e\n\u003cli\u003eSandic Spaho R, Uhrenfeldt L, Fotis T, Kymre IG. Wearable devices in palliative care for people 65 years and older: a scoping review. Digit Health. 2023;9:20552076231181212.\u003c/li\u003e\n\u003cli\u003eBradley SH, DeVito NJ, Lloyd KE, Richards GC, Rombey T, Wayant C, et al. Reducing bias and improving transparency in medical research: a critical overview of the problems, progress and suggested next steps. J R Soc Med. 2020;113(11):433\u0026ndash;443.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-palliative-care","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pcar","sideBox":"Learn more about [BMC Palliative Care](http://bmcpalliatcare.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pcar/default.aspx","title":"BMC Palliative Care","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Digital health technologies, Artificial Intelligence, Machine learning, Digital tools, Scoping review","lastPublishedDoi":"10.21203/rs.3.rs-8924351/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8924351/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePalliative care improves the quality of life of patients with life-limiting conditions and their families; however, global access remains constrained by workforce shortages and late referrals. Artificial Intelligence (AI) has been proposed as a scalable solution for optimising the identification of needs, supporting clinical decision-making, and enhancing care delivery. However, real-world evidence of the application of AI in palliative care remains sparse, particularly regarding its impact on quality of life, quality of care, and associated practical, technical and ethical challenges.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e A scoping review was conducted following the Joanna Briggs Institute methodology and the PRISMA-ScR checklist. Five databases (the ACM Digital Library, CINAHL, Cochrane Central, PubMed and Web of Science) were searched. Studies reporting the use of AI to facilitate or enhance palliative care delivery in adults were eligible. Four reviewers independently screened the records, and two reviewers extracted the data using Covidence software. A narrative synthesis was then performed.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFifteen studies, published between 2021 and 2025, were included. Fourteen originated from Global North settings (USA 5, Germany 2, Japan 2, Taiwan 2, UK 1, Spain 1 and Cyprus 1) and one from Iran. Conceptually, AI applications fall into three domains: (1) early identification of palliative care needs, (2) symptom assessment and management and (3) clinical decision support for care conversations. Fifteen studies (100%) reported or discussed quality of care outcomes, most commonly prognostic performance, usability and referral/conversation rates, and only two (13.3%) directly addressed quality of life. Effectiveness was consistently positive, with four randomised controlled trials demonstrating superiority over usual care in referrals, advance care planning, pain control and quality of life domains. Practical barriers were centred on workflow integration and resource demands, while technical limitations include data quality, generalisability, and interpretability. Ethical discourse is underdeveloped, with major gaps in the principles of AI governance.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eAI shows potential to improve prognostic accuracy, trigger earlier involvement of palliative care specialists and support symptom management. However, this evidence is geographically skewed, methodologically immature and ethically underdeveloped. Future research must prioritise diverse global settings, patient-reported quality of life outcomes, participatory co-design and systematic ethical governance to ensure equitable implementation.\u003c/p\u003e","manuscriptTitle":"Exploring the application of Artificial Intelligence in palliative care and its practical, technical and ethical considerations: a scoping review","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-08 14:57:06","doi":"10.21203/rs.3.rs-8924351/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-16T01:10:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-10T13:17:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"163093501240200189619027324480400094216","date":"2026-04-07T04:56:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"203378079342110223385818817585964196885","date":"2026-03-26T08:45:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"46375426969096349874332637049552916080","date":"2026-02-26T18:13:08+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-26T17:18:49+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-23T06:57:20+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-22T22:29:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-22T22:28:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Palliative Care","date":"2026-02-20T09:23:22+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-palliative-care","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pcar","sideBox":"Learn more about [BMC Palliative Care](http://bmcpalliatcare.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pcar/default.aspx","title":"BMC Palliative Care","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f2d931c3-5260-44cd-abc0-43f5d94ea51e","owner":[],"postedDate":"March 8th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-16T01:10:28+00:00","index":88,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-08T14:57:06+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-08 14:57:06","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8924351","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8924351","identity":"rs-8924351","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}