Artificial intelligence-based approaches for advance care planning: a scoping review.

Ariozetal. BMC Palliative Care (2025) 24:268 https://doi.org/10.1186/s12904-025-01827-x RESEARCH Open Access © The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. BMC Palliative Care Artificial intelligence-based approaches foradvance care planning: ascoping review Umut Arioz1*, Matthew John Allsop2*, William D. Goodman2, Suzanne Timmons3, Kseniya Simbirtseva3, Izidor Mlakar1 and Grega Mocnik1 Abstract Background Advance Care Planning (ACP) empowers individuals to make informed decisions about their future healthcare. However, barriers including time constraints and a lack of clarity on professional responsibilities for ACP hinder its implementation. The application of artificial intelligence (AI) could potentially optimise elements of ACP in practice by, for example, identifying patients for whom ACP may be relevant and aiding ACP-related decision-making. However, it is unclear how applications of AI for ACP are currently being used in the delivery of palliative care. Objectives To explore the use of AI models for ACP, identifying key features that influence model performance, transparency of data used, source code availability, and generalizability. Methods A scoping review was conducted using the Arksey and O’Malley framework and the PRISMA-ScR guidelines. Electronic databases (Scopus and Web of Science (WoS)) and seven preprint servers were searched to identify published research articles and conference papers in English, German and French for the last 10 years’ records. Our search strategy was based on terms for ACP and artificial intelligence models (including machine learning). The GRADE approach was used to assess the quality of included studies. Results Included studies (N = 41) predominantly used retrospective cohort designs and real-world electronic health record data. Most studies (n = 39) focused on identifying individuals who might benefit from ACP, while fewer studies addressed initiating ACP discussions (n = 10) or documenting and sharing ACP information (n = 8). Among AI and machine learning models, logistic regression was the most frequent analytical method (n = 15). Most models (n = 28) demonstrated good to very good performance. However, concerns remain regarding data and code availability, as many studies lacked transparency and reproducibility (n = 17 and n = 36, respectively). Conclusion Most studies report models with promising results for predicting patient outcomes and supporting decision-making, but significant challenges remain, particularly regarding data and code availability. Future research should prioritize transparency and open-source code to facilitate rigorous evaluation. There is scope to explore novel AI-based approaches to ACP, including to support processes surrounding the review and updating of ACP information. Keywords Advance care planning, Digital tools, Palliative care, Artificial intelligence, Machine learning *Correspondence: Umut Arioz [email protected] Matthew John Allsop [email protected] Full list of author information is available at the end of the article Page 2 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 Introduction Advance Care Planning (ACP) “…is a voluntary process of person-centred discussion between an individual and their care providers about their preferences and priorities for their future care” [1]. This often includes multiple discussions and can include any people that a person would like to involve. The goal of ACP is to ensure that a person’s values, beliefs, and goals for care are understood and respected, particularly when they may be unable to communicate these wishes themselves [2]. This process is especially important when individuals face serious illness, empowering them to articulate their healthcare preferences and appoint a surrogate decision-maker. ACP may involve documenting these preferences through advance directives or similar legally binding instruments to ensure patient-centered care, even in moments when they cannot advocate for themselves. ACP is not limited to palliative care settings but spans community and acute care environments [3]. It plays a central role in ensuring that healthcare aligns with patient values throughout their illness trajectory. Caregivers, in their supportive roles, can help ensure that these values are continually reassessed and that the patient’s healthcare preferences evolve as needed. The regular review of ACP is essential to reflect changes in health status or personal values, further enhancing the quality of care at the end of life [4]. Moreover, early engagement in ACP, involving both the patient and their caregivers, has been shown to reduce unnecessary interventions, mitigate healthcare costs, and improve the overall quality of care by focusing on patientcentered outcomes, especially in populations such as cancer patients [5, 6]. Evidence highlights that when caregivers understand the patient’s goals, they experience reductions in both anxiety and emotional strain as they feel more prepared to navigate complex medical decisions [7]. Despite the potential benefits of ACP, its implementation remains a challenge for healthcare professionals. Insufficient communication skills, a lack of knowledge to determine when and how to have ACP conversations, and time constraints hinder its widespread adoption [8]. While time constraints are a significant barrier, an emerging approach to support ACP implementation is the integration of Artificial Intelligence/Machine Learning (AI/ML) methods which offers broader potential benefits beyond simply saving time. Artificial Intelligence (AI) encompasses various methods that enable machines to mimic human cognitive functions. Among these, Machine Learning (ML) is a dominant and widely used subset, where systems learn from data to perform tasks without being explicitly programmed. In that sence, AI can contribute to ACP by, for example, supporting the timely identification of people who could benefit from ACP through algorithms that predict mortality and disease progression [4, 7, 9–14]. Furthermore, AI can enhance predictive accuracy in identifying individuals who would benefit most from ACP discussions [15–18] personalize ACP processes by analyzing patient data to tailor information and discussions to individual needs and values [19–22] and support clinicians with decision-making tools that integrate real-world data to forecast patient outcomes and facilitate more informed end-of-life care planning [1, 20, 23–25]. While digital resources such as websites, portals, and apps have been designed to support reflection, communication, decision-making, and documentation of end-oflife care, the specific application of artificial intelligence approaches in this context remains relatively understudied [26]. AI-powered tools differ from traditional digital platforms by offering enhanced interactivity, prediction, and data-driven personalization. For instance, AI can facilitate more interactive experiences through chatbots or conversational agents that guide users through ACP discussions [27]. Predictive AI models can analyze patient data to anticipate future care needs or identify individuals who might benefit most from ACP, enabling proactive engagement [28]. Furthermore, AI algorithms can analyze individual patient characteristics to tailor information and support, adapting to specific preferences and circumstances in a way that static websites or portals cannot [29]. While AI-based approaches are beginning to show promise in identifying patients in their last year of life, few resources integrate seamlessly with existing healthcare workflows or offer interactive interfaces tailored for healthcare providers [30]. For the specific use of AI-based approaches to support ACP, to date, there has been no compilation and outline of the current evidence base. This scoping review aims to comprehensively examine the current state of AI approaches within digital tools for ACP. Specifically, it explores how AI approaches can be applied to support ACP as part of palliative care delivery. Methods Overview This scoping review was undertaken as part of the AI4HOPE project (https:// www. ai4ho pe. eu). The review was undertaken as part of early co-design activities to guide the development of a digital ACP platform to support information provision for people living with mild to moderate dementia, alongside a focus on the documentation of information that they would like to be part of Page 3 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 ACP discussions,. As the topic of AI supporting ACP is still emerging, this review focuses on any health conditions. The scoping review questions were formed through applying the Population-Concept-Context (PCC) framework [31]. ‘Population’ included people with serious or life-limiting illnesses, ‘Concept’ included the application of AI approaches as part of the ACP process, and ‘Context’ remained broad and included any country setting internationally, alongside including any care setting supporting the population of interest. Three broad questions derived from the PCC framework elements guided the approach to the scoping review, and included: In which domains of ACP can AI approaches be effectively applied? What are the primary and secondary objectives of utilizing AI approaches in digital tools for ACP? What are the most critical features to consider in AI approaches for ACP, and how can feature selection be optimized to enhance model performance and generalizability? This scoping review adhered to the methodological framework outlined by Arksey and O’Malley [32] and complies with the PRISMA-ScR checklist [33]. Following this guidance, we conducted six sequential steps: 1) defining research questions, 2) developing a comprehensive search strategy, 3) selecting studies based on predefined criteria, 4) extracting and analyzing data, 5) summarizing and reporting findings, and 6) consulting study authors for validation. The PRISMA-ScR checklist ensured the systematic, transparent, and comprehensive nature of our review. Eligibility criteria The search strategy focused on people with serious or life-limiting illnesses, irrespective of setting. The emphasis on AI approaches supporting ACP highlights the importance of the palliative care setting but the review retains a broader focus across any care setting supporting the population of interest. Database searching included published research articles and conference papers between 2015 and 2024. Studies in three languages were included (English, German and French). Information sources Scopus and Web of Science (WoS) databases were used to identify relevant literature. Also, the following preprints servers were searched: arXiv, bioRxiv, JMIR Preprints, medRxiv, Preprints.org, Research Square and SSRN Journal. Search Our search strategy was based on the ‘AND’ operation of two main categories; keywords related to ACP, and keywords related to AI approaches. The box below provides examples of key terms used in the search strategy. ("advance care planning"OR"advance care directives"OR"advance directives"OR"advance statement"OR"advance decision"OR"advanced care planning"OR"advanced care directives"OR"advanced directives"OR"advanced statement"OR"advanced decision"OR"palliative care"OR"Palliative Medicine") AND ("Artificial intelligence"OR"machine learning"OR"multivariate approach"OR"AI-enabled"OR"MLgenerated"OR"Supervised learning"OR"unsupervised learning"OR"Deep learning"OR"classification"OR"Neural networks") Selection ofsources ofevidence A comprehensive literature search yielded a substantial number of potential studies. Prior to this independent screening, we conducted a calibration exercise. All authors participated in discussions to refine the eligibility criteria and assessed collectively a sample dataset comprising more than 10% of the total papers. We discussed our selections and the rationale behind them, which led to further refinement of the eligibility criteria. Following this calibration, the remaining papers were divided uniformly among the authors for independent title and abstract screening. In the second screening phase, all authors conducted a detailed review of the full texts of the remaining studies. The same exclusion criteria as in the previous stage were applied. Discrepancies were resolved through consensus among the authors. Data collection andcharting process Relevant data were extracted from the final list of studies and organized into a spreadsheet by all authors. Data items The data extraction characteristics chosen for the scoping review provide an overview of key characteristics of included studies. Data categories extracted included Study Identification and Design (the target disease(s) or condition(s) being investigated; study design; and number of participants); Data and Methodology (origin of the data, such as EHRs or publicly available databases, and number of records in the dataset); Application and Evaluation of AI approaches (the specific algorithms employed in the study, with categories derived from existing literature [34–36]); primary and secondary aim(s); data types included in models (specific characteristics or variables extracted from the data for analysis); data availability (whether the study dataset is publicly accessible and, if so, from where, including the source code to support replication and further exploration); model performance (evaluation metrics used to assess the accuracy and effectiveness of the AI approach, such as sensitivity, specificity, or accuracy); and clinical application within ACP processes. For the clinical application of processes, Page 4 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 included studies were charted against different elements that may be involved in the identification, documentation, revision and sharing of ACP documentation. The phases were informed by an earlier conceptual model of digital approaches to ACP [37]. Studies were classified against any of the six elements of ACP (Appendix D), outlined in Fig.Â1. Critical appraisal ofindividual sources ofevidence To assess the quality of evidence for each included study, the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) approach [24] was employed, which is a widely recognized framework for assessing the quality of evidence from research. However, GRADE ratings were not assigned to preprints due to their preliminary nature and lack of peer review, a standard process for journal articles. The GRADE ratings can be found in Appendix A. Synthesis ofresults A narrative synthesis was conducted to outline the design, sample size, participant disease types, AI methods and primary and secondary aims of the included studies. The main features of ACP were then extracted and categorised into thematically similar groups. The performance of the AI approach was evaluated by two authors (UM, GM) and given a rating on a 5-point scale: Excellent (high accuracy and effectiveness in artificial intelligence approach), Very Good (suggesting solid model performance with some room for improvement), Good (demonstrating reasonable accuracy and effectiveness), Fair (moderate performance that may require further refinement), or Poor (significant room for improvement in the model’s accuracy and effectiveness) (see Appendix C for further detail). The process of calculating performance levels for the models involved assessing each model’s performance based on several metrics, including area under the curve – receiver operating characteristic (AUC-ROC), F1 Score (a score of model accuracy), Accuracy, Precision and Recall, and Brier Score. The process of calculating performance levels for the models involves assessing each model’s performance based on several metrics, including AUCROC, F1 Score, Accuracy, Precision and Recall, and Brier Score. However, a key challenge arises when certain metrics are missing or unavailable, making it difficult to assign an accurate performance level based solely on incomplete data. To address this issue, we have developed a method where each individual metric is assessed separately for its performance level. This allows us to assign a level based on the available metrics. In cases where not all metrics are provided, we still assign a performance level for the ones available, ensuring that the evaluation is based on the best available data. Once individual performance levels for each metric are determined, a total performance level is calculated by considering the minimum value of the individual levels. This ensures that the final performance level reflects the model’s overall performance, even if some metrics are missing or unavailable. This paper presents a novel approach to evaluating AI/ ML models in scenarios where complete performance metric data is unavailable. Traditional methods often rely on a comprehensive set of metrics to assess model performance. However, when data is incomplete, accurate evaluation becomes challenging. Our proposed method addresses this limitation by evaluating each metric individually and assigning a performance level based on available data. The final performance level is determined by the minimum individual metric level, ensuring a conservative yet informative assessment of the model’s overall performance. The data and source code availability were summarised. Finally, all studies were assessed for the clinical application of the AI approach to ACP and these findings were discussed in a narrative synthesis. Results Selection ofsources ofevidence The database search yielded 394 studies, which were reduced to 385 unique records after duplicates were removed. Of these, 321 were excluded during abstract screening because they did not include AI approaches or were irrelevant to ACP. Subsequent screening of the full text led to the exclusion of 26 further studies for similar reasons (see Fig.Â2), resulting in 38 studies being Fig.Â1 Different aspects of ACP Page 5 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 included in the scoping review. Alongside databases of published research, seven preprint servers were searched, yielding a further 500 studies. Of these, 25 were selected for full-text review, from which 3 studies were included. A total of 41 studies were including in the review (Fig.Â2). A comprehensive list of included studies and extracted data is reported in Appendix B. Critical appraisal withinsources ofevidence According to the GRADE assessment, the majority of studies (n = 28; 68.42%) were rated as being of moderate quality. Two studies [12, 38] were rated as high quality and five studies [19, 25, 39–41] as low quality. In keeping with scoping review methodologies, these were not excluded. Results ofindividual sources ofevidence Study designs Most studies (n = 24; 58.55%) utilized retrospective cohort designs (Table1). Real-world EHR data was the most common data source (n = 37; 90.24%) for developing and training models, with varying sample sizes. The Fig.Â2 PRISMA flowchart of scoping review Table 1 Number of study design types used in included studies Study design nDisease n Retrospective cohort study 24 All diseases 24 Prospective cohort study 6 Cancer 13 Mixed Methods Study 4 Alzheimer’s disease 2 Prospective observational study 3 End-stage liver disease 1 Randomized controlled trial 3 Geriatric fragility fracture 1 Cross-sectional study 1 Page 6 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 focus of studies varied, with’All Diseases’being the most frequent category (n = 24; 58.53%) (Table 1). Other common disease areas included cancer, Alzheimer’s disease, and geriatric fragility fractures. Characterising artificial intelligence approaches forACP The 41 studies included in this scoping review employed a diverse array of AI approaches to address aspects of ACP (see Appendix B). Within these, logistic regression (LR) [13–15, 20, 25, 39–46] and Random Forest models (RF) [12, 15, 19, 20, 25, 38, 40–42, 45–48]were the most widely used methods (see TableÂ2). Artificial neural networks (ANN) [9, 14, 15, 40, 48–50], extreme gradient boosting (XGBoost) [13, 15, 39, 43, 51–53], support vector machines (SVM) [9, 13–15, 39, 44, 54], and decision trees (DT) [14, 15, 41, 43, 54–56] were also frequently employed. A range of other AI methods were also reported, including gradient boosting machines (GBM), deep neural networks (DNN), Gaussian naive Bayes (GNB), k-nearest neighbors (KNN), and natural language processing (NLP), where these methods were tailored to specific ACP contexts or tasks. The majority of studies (78%) detailed models that focused on forecasting or predicting future patient outcomes, such as survival, hospital length of stay, or the likelihood of specific events (TableÂ 2). These studies focussed on categorizing data into discrete classes or groups. Four studies [19, 48, 57, 58] involved classifying patients into high-risk or low-risk groups, predicting the presence or absence of a disease, or determining the optimal course of treatment. Studies in this category aimed to describe or summarize data without making specific predictions or classifications. This involved identifying patterns, trends, or relationships within the data. Two studies [21, 47] were categorized as descriptive, and focussed on finding the best possible solution to a given problem, such as optimizing treatment plans or resource allocation. One study [54] combined elements of both prediction and classification, aiming to both forecast outcomes and categorize patients. The included studies were further classified according to their secondary aims (TableÂ2). The most common secondary aim (46.34%) was decision support, alongside studies with a focus on selection of individuals who may benefit from specific interventions or treatments (41.17%). Additional secondary aims across studies included the discovery of new knowledge or insights from patient data (e.g., identifying patterns, trends, or relationships that were previously unknown; 7.31%) and optimising treatment plans or resource allocation (4.87%). A focus on improving processes within the healthcare system, such as streamlining workflows or optimising resource allocation, was also identified (7.31%). Important features forACP Below, we outlined a comprehensive overview of the data elements utilised across AI models for ACP (detailed further in Appendix B). These features can be categorised into several key dimensions: Patient‑centric features Demographics and Socioeco‑ nomic Factors: Age [9, 27, 29, 42, 43, 51], gender [13, 39, 51, 55, 56], and location [20, 39]. Clinical Data: A broad spectrum of clinical information, including diagnoses [10, 14, 53], procedures [10, 16], medications [10, 16, 57, 59], laboratory results [10, 41, 45, 52–54, 59], vital signs [10], medical history [16, 60], and symptoms (frailty [59], activity [22, 39, 43], pain [51, 57, 61], nausea [57], and delirium [13, 61]), was employed. Palliative and End‑of‑Life Indices: Factors related to end-of-life care, including palliative care consultations [12], and do-not-resuscitate orders [62], largely used for predicting patient illness trajectories. Behavioral Data: Studies incorporated patient activity data [22, 39, 43], sleep patterns [45], and body movements [19, 63]. Healthcare system features Resource Utilization: Information about healthcare costs [14], service utilization (e.g., medical care, nursing) [50, 60], and system Table 2 Methods applied within AI approaches and aims of included studies AI Method Frequency (> 5) Primary Aim n Secondary Aim n Logistic regression 13 Prediction 32 Decision Support 19 Random forest 13 Classification 4 Patient Selection 14 Artificial neural networks 7 Description 2 Knowledge Discovery 3 Decision trees 7 Optimization 2 Process Improvement 3 Support vector machines 7 Prediction and Classification 1 Optimization 2 Gradient boosting machines 7 k-nearest neighbors 6 Page 7 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 efficiency (e.g., length of stay, readmissions) [12] to optimize care delivery. Provider‑Related Factors: Information about healthcare providers, such as their experience [20], opinions [20], and work habits [38]. Model performance oftheAI approaches Within the studies that provided performance measure data, most of the studies (36.6%) demonstrated excellent performance. Some studies reported very good performance (30%) and good performance (16.6%). None of the studies were classified as fair, with some studies (n = 15; 36.6%) lacking sufficient information to assess their performance level. Particularly, traditional AI/ML models, like RF [9, 20, 43, 44, 55, 62], LR [9, 20, 40, 43, 44], SVM [9, 43, 44] and DT [43, 55] consistently demonstrated strong performance in the included studies, achieving ‘excellent’ results (level 1). Data availability The availability of data for the 41 studies varies significantly (see Appendix B). One study [64] made the source code publicly available, facilitating replication and further analysis. Some studies (n=15; 36.6%) indicated that data would be available upon reasonable request. However, many (n=17; 41.5%) did not provide any information regarding data availability. A small number of studies (n =5; 12.2%) explicitly stated that the data is not publicly available, potentially due to privacy or other concerns. Only one study provided an open dataset [16], making it accessible to other researchers for replication and further analysis. Three studies [57–59] Âprovided a link to the dataset, facilitating access for interested parties. However, a significant majority of studies (n =32; 78.04%) did not provide any information regarding source code availability. Clinical application ofartificial intelligence approaches forACP The detailed classification of the different elements of ACP in the included studies is reported in Appendix D. Almost all studies focused on identifying individuals who might benefit from ACP (n = 39; 95.1%), while fewer studies addressed initiating ACP discussions (n = 10; 24.3%) or documenting and sharing ACP preferences and decisions (n = 8; 19.5%). Few studies explored where ACP discussions are recorded (n = 2; 4.87%). The next most frequently studied aspect was accessing and using ACP to inform decision-making (n = 37; 90.2%). No studies directly addressed reviewing and updating ACP information. The tools reported across the included studies primarily focused on supporting the identification of patients who might benefit from ACP and informing decision-making. No studies were identified where models support the reviewing and updating of ACP information (Fig.3). Discussion Summary ofevidence This review provides an overview of the current literature detailing AI approaches for ACP and provides an overview of the current state, objectives, and critical features of these models. Overall, the scoping review demonstrates a strong focus on prediction and classification within the context of AI approaches in ACP. Included studies commonly focused on exploring how to embed AI approaches to assist healthcare professionals in making informed decisions about patient care and identifying patients who would benefit most from specific interventions or treatments. Wider secondary aims also highlighted AI approaches beyond prediction and classification, focusing on discovering new knowledge and improving healthcare processes. Most included studies were assessed as achieving an excellent performance, indicating the exceptional development of AI approaches for supporting ACP. However, while most studies performed very well, there is limited data availability and source code to ensure the reproducibility and transparency of emerging AI approaches for ACP research. As we will reinforce in the methods section, the findings of this review regarding AI approaches in ACP have applicability to broader contexts beyond palliative care, encompassing all individuals facing serious or life-limiting illnesses. Fig.Â3 Heatmap for AI support across ACP phases Page 8 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 Most tools reported across the included studies are primarily focused on supporting the identification of patients who might benefit from ACP and informing decision-making, suggesting AI-based tools may be well-suited for identifying patients for whom discussion of end-of-life care preferences may be appropriate. This includes, for example, models to predict mortality and frailty for older patients to guide data-driven decisionmaking about the initiation of the ACP process. This may provide a prompt for earlier identification of opportunities to initiate ACP discussions and documentation, but there is recognition that such screening approaches may need to be developed to also anticipate palliative care needs and predict the rate and course of functional decline [65]. There was, however, gaps in the coverage of AI approaches to support different phases of ACP, including during ACP discussions and ACP documentation and sharing. In busy clinical practice, clinicians may struggle to find time to support ACP at the patient’s preferred pace. A 2019 systematic review described patientlevel barriers to ACP, such as perceived irrelevance and poor awareness, and also time issues for clinicians [66]. Approaches such as using AI-powered chatbots are being considered for offering resources for spiritual exploration and facilitating religious practices for people living with head and neck cancer [67]. In the context of ACP, similar AI-powered chatbots may have potential utility for supporting exploration and documentation of ACP preferences, assisting patients across different elements of ACP, which could potentially reduce clinical workload while empowering patients. The review highlighted a notable gap in literature detailing the role of AI-based approaches to support the reviewing and updating of ACP documentation. AI models have the ability to pick up clinical changes in real time [61, 68] and may have the potential to facilitate timely review within this aspect of ACP. For example, patient-directed chatbots or conversational agents could promote reflection and revision between clinical visits [12, 67]. The use of scenario modelling tools can aid patients to anticipate future care needs and adjust preferences where necessary [69]. Further, AI added to workflows can stimulate clinicians to re-initiate ACP conversations after key health events [42, 70]. Addressing this gap in AI-based ACP research should be a priority for future research. An empowered patient who has accessed and processed a certain level of information about their disease and its likely future progression, and who has already reflected on their values, wishes and fears for the future, would likely derive more benefit from a time-limited interaction with a clinician than one who is uninformed and unprepared. AI approaches could support targeted, interactive information provision and future scenario modelling to the patient, supporting the patient to document their questions, decisions and uncertainties in advance of an ACP discussion with a clinician. A patientdriven approach also circumvents the known difficulties that clinicians report in finding ‘the right time’ to have ACP conversations [71], as the patient can interact with information relevant to them, in their own time, and be empowered to seek the ACP interaction, rather than waiting for a clinician to initiate it. AI could also enable real-time monitoring of patients’ health conditions and provide alerts to consider adjustments to ACP documentation. This use of AI could embed ACP as a continual process, rather than a one-time event [72]. There are points that should be considered when implementing AI approaches in a healthcare system. While AI approaches could offer data-driven insights that may assist clinicians in making informed decisions and improve the quality of ACP discussions, they should augment clinical skills rather than supplant clinical decision-making [73]. Furthermore, patients’ end-of-life care choices can be shaped by cultural factors [74], therefore, AI systems should be intentionally designed to accommodate a wide range of cultural backgrounds, as there may be variability when it comes to how and who should be involved in ACP discussions. For example, the development of AI-based approaches could utilise diverse datasets, ACP content may be personalised in response to known or disclosed cultural or religious affiliation (e.g., for a Muslim patient presenting content addressing fasting during illness), prioritise the ability of systems to interpret and respond when people use religious and culturally nuanced terms, and ensuring the co-design and development of AI-based approaches with input and involvement of patient and communities from diverse backgrounds. There is also scope to draw on emerging frameworks such as a focus on adopting AI approaches in culturally diverse healthcare settings to avoid culturally inappropriate discussions [75]. The use of AI in healthcare raises ethical concerns, such as data privacy, informed consent, and algorithmic bias. It is essential that AI systems are transparent and that patients’ data are safeguarded to ensure patients can have continued trust in their healthcare [76]. The findings of this scoping review highlight the potential of AI approaches to support ACP. However, despite the model performance of some of the artificial intelligence approaches being categorised as good or excellent, this does not hold across all the models. Limitations in the performance of the models can be attributed to the Page 9 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 quantity and diversity of the training sample available for the AI/ML models to use. This scoping review identified only one study providing an open dataset. Promoting open-source practices is essential for improving transparency, collaboration, and the overall quality of ACP research with AI-based approaches [77] and is an increasing requirement of research funders. While several studies acknowledge the use of proprietary datasets or models without sharing source code, this limited transparency poses significant challenges. Specifically, it hampers reproducibility, as other researchers cannot verify or replicate the reported findings. Furthermore, the inability to inspect training data raises concerns about potential bias, which may undermine the fairness and generalizability of AI systems. Finally, without open access to code or validation data, independent assessment of model performance is not possible, limiting the credibility of these approaches in clinical settings. Furthermore, it is suggested that a range of features used within these models can help improve their performance including patient-level variables (functional status and symptoms) [78, 79], healthcare utilisation variables (resource utilisation, patient satisfaction) and healthcare professional factors (experience and opinions) [13]. Therefore, the diversity of the sample and inputs is key to improving model performance. This review identified that most studies focussed on all disease types when building the models for AI use in ACP. While this approach can make the models generalisable to multiple patient populations, it also requires the model to be more complex as it requires an understanding of disease-specific factors and variations which can be a challenge to finding samples of sufficient size and quality to train the models. Future research focusing on disease-specific models of ACP may help in developing more effective and tailored tools to support decisionmaking that addresses the specific needs of different disease populations. For example, research has highlighted the different barriers in ACP for certain disease populations, including a lack of understanding of the disease trajectory by the healthcare professional, treatment option availability, and reduced mental capacity in the patient [80]. The latter is particularly important in dementia, a common disease where ACP is needed as the person will lose communication and decision-making abilities as the disease progresses [81]. AI approaches could also support ACP as a more continuous process, as is known best practice [82], rather than a discrete, rushed event. AI approaches could be particularly beneficial in atypical dementia sub-types, where a person may be showing signs of impaired speech, understanding or concentration early in the disease, and AI could flag up the need for more urgent ACP conversations [82], supporting simple information being presented in the means that best suited the person’s information processing needs at the time. Therefore, by accurately predicting future outcomes and classifying patients into appropriate categories, AI may help healthcare providers initiate timely ACP, alongside providing tailored information to guide ACP discussions and documentation. Our review identified a diverse range of applications of AI models in the context of ACP. To provide a more nuanced understanding of these applications, we categorized the included studies based on the directness of their support for ACP: Category 1) Direct ACP Support: AI models designed to directly facilitate or perform core ACP activities (e.g., eliciting preferences, supporting shared decision-making, generating ACP documents). Category 2) Indirect ACP Support: AI models that provide information or perform tasks that support ACP but do not directly carry out the core activities (e.g., predicting events that might trigger ACP discussions, identifying patients who might benefit from ACP) and Category 3) Tangential Relevance: AI models with a more tangential or distant relationship to ACP, focusing on related areas (e.g., palliative care more broadly) but potentially having implications for ACP. Our analysis suggests that many included studies fall into Categories 2 and 3. For example, Study #4 used AI to predict patient mortality as a trigger for ACP discussions, while other studies focused on predicting patient outcomes or managing symptoms in palliative care settings. While these applications can contribute to the context of ACP, they do not always directly engage with the core elements of ACP itself. The limited number of studies that explicitly focused on Category 1 (direct ACP support) highlights a potential gap in the current literature. This observation has several implications: * The current evidence base may be stronger for AI’s ability to support ACP-related tasks than for its ability to directly facilitate core ACP processes. * The potential of AI to transform ACP by directly engaging with patients in preference elicitation or shared decision-making remains largely unexplored. * Future research should prioritize the development and rigorous evaluation of AI models that actively support the core elements of ACP. Limitations While the search strategy was comprehensive, there are certain limitations. The search was conducted in Scopus and Web of Science databases, covering a large proportion of existing health research literature. However, the Page 16 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 Study No Study Directness of their support for ACP Identity person who may benefit from ACP Initiate ACP discussion Documentation and sharing of ACP Identifying records where ACP information has been recorded Accessing and using ACP to inform decisionmaking Reviewing and updating ACP information 14 [43] 2 X X X X 15 [47] 2 X X X X 16 [44] 2 X X 17 [16] 2 X X 18 [48] 3 X X 19 [60] 3 X X X 20 [69] 3 X X 21 [54] 2 X X 22 [38] 2 X X X 23 [57] 3 X X 24 [64] 2 X X X 25 [14] 2 X X X 26 [20] 3 X 27 [40] 3 X X 28 [51] 2 X X 29 [45] 2 X X 30 [19] 3 X X 31 [59] 2 X X 32 [41] 3 X X 33 [58] 3 X X 34 [55] 2 X X 35 [52] 2 X X Study No Study Directness of their support for ACP Identity person who may benefit from ACP Initiate ACP discussion Documentation and sharing of ACP Identifying records where ACP information has been recorded Accessing and using ACP to inform decisionmaking Reviewing and updating ACP information 36 [9] 2 X X 37 [21] 3 X X 38 [62] 3 X X 39 [53] 2 X X 40 [46] 2 X 41 [18] 1 X X X X X 39 10 8 2 37 0 Abbreviations ACP Advance Care Planning AdaBoost Adaptive boosting AI Artificial Intelligence ANN Artificial Neural Networks ARM Association rule mining AUPRC Area Under the Precision-Recall Curve AUROC Area Under the Receiver Operating Characteristic Curve BiLSTM Bidirectional Long short-term memory CFA Collaborative filtering algorithm CNN Convolutional neural network DF Decision forest DL Deep Learning DNN Deep Neural Networks DT Decision Trees EHR Electronic Health Record ESLD End-stage liver disease GBM Gradient Boosting Machines Gboost Gradient Boost classifier GNB Gaussian Naïve Bayes GRU Gated Recurrent Units ID3 Iterative Dichotomiser 3 KNN K-Nearest Neighbour LDA Linear discriminant analysis LMT Logistic model tree LR Logistic Regression LSTM Long short-term memory ML Machine Learning NLP Natural Language Processing RF Random Forest Page 17 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 SUS System Usability Scale SVM Support Vector Machine UEQ-S User Experience Questionnaire – Short Version XGBoost Extreme gradient boosting Authors’ contributions IM, UA, GM and MJA conceived and designed the study. UA, GM, WG and MJA performed the search, data extractions, and quality assessment. UA, MJA, and WG interpreted the study findings. UM, MJA and WG drafted the manuscript, and all authors (UA, MJA, WG, GM, ST, KS and IM) critically revised the manuscript. All authors approved the final version of this paper. Funding This work has been funded by the European Union Horizon Europe Research and Innovation Programme projects AI4HOPE (grant number 101136769). The content of this paper does not reflect the official position of the European Union or any other institution. The information and views expressed are the sole responsibility of the authors. Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Author details 1 Faculty of Electrical Engineering and Computer Science, University of Maribor, Maribor 2000, Slovenia. 2 School of Medicine, Faculty of Medicine and Health, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK. 3 Centre for Gerontology and Rehabilitation, School of Medicine, College of Medicine and Health, University College Cork, Cork, Ireland. Received: 9 December 2024 Accepted: 23 June 2025 References 1. ‘Universal principles for advance care planning (ACP)’, Ministerial oversight group, Mar. 2022. Available: https:// www. engla nd. nhs. uk/ wpconte nt/ uploa ds/ 2022/ 03/ unive rsalprinc iplesforadvan cecareplann ing. pdf 2. Dinescu A. Advance Care Planning. Clin Geriatr Med. Nov 2021;37(4):605– 10. https:// doi. org/ 10. 1016/j. cger. 2021. 06. 001 3. Fliedner M, Halfens RJG, King CR, Eychmueller S, Lohrmann C, Schols JMGA. Roles and Responsibilities of Nurses in Advance Care Planning in Palliative Care in the Acute Care Setting: A Scoping Review. J Hosp Palliat Nurs. Feb 2021;23(1):59–68. https:// doi. org/ 10. 1097/ NJH. 00000 00000 000715. 4. Wilkin K, Fang ML, Sixsmith J. Implementing advance care planning in palliative and end of life care: a scoping review of community nursing perspectives. BMC Geriatr. Mar 2024;24(1):294. https:// doi. org/ 10. 1186/ s1287702404888-4. 5. Ma J, et al. Early Palliative Care Consultation in the Medical ICU: A Cluster Randomized Crossover Trial. Crit Care Med. Dec 2019;47(12):1707–15. https:// doi. org/ 10. 1097/ CCM. 00000 00000 004016. 6. Smith S, Brick A, O’Hara S, Normand C. Evidence on the cost and cost-effectiveness of palliative care: A literature review. Palliat Med. Feb 2014;28(2):130–50. https:// doi. org/ 10. 1177/ 02692 16313 493466. 7. 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;12(7):e060201. https:// doi. org/ 10. 1136/ bmjop en2021060201. 8. Vanderhaeghen B, Bossuyt I, Menten J, Rober P. Helping hospital professionals to implement Advance Care Planning in daily practice: a European Delphi study from field experts. J Res Nurs. Sep 2019;24(6):433–43. https:// doi. org/ 10. 1177/ 17449 87118 772604. 9. Blom MC, Ashfaq A, Sant’Anna A, Anderson PD, Lingman M. ‘Training machine learning models to predict 30-day mortality in patients discharged from the emergency department: a retrospective, populationbased registry study’. BMJ Open, Aug 2019;9(8):e028015. https:// doi. org/ 10. 1136/ bmjop en2018028015. 10. Chi S, et al. Advanced Care Planning for Hospitalized Patients Following Clinician Notification of Patient Mortality by a Machine Learning Algorithm. JAMA Netw Open. Apr 2023;6(4): e238795. https:// doi. org/ 10. 1001/ jaman etwor kopen. 2023. 8795. 11. Berg GD, Gurley VF. Development and validation of 15-month mortality prediction models: a retrospective observational comparison of machinelearning techniques in a national sample of Medicare recipients. BMJ Open. Jul 2019;9(7):e022935. https:// doi. org/ 10. 1136/ bmjop en2018022935. 12. Wilson PM, 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. Jul 2023;66(1):24–32. https:// doi. org/ 10. 1016/j. jpain symman. 2023. 02. 317. 13. Kim E-J, Koo Y-R, Nam I-C. Patients and Healthcare Providers’ Perspectives on Patient Experience Factors and a Model of Patient-Centered Care Communication: A Systematic Review. Healthcare. May2024;12(11):1090. https:// doi. org/ 10. 3390/ healt hcare 12111 090. 14. Zhang H, Li Y, McConnell W. Predicting potential palliative care beneficiaries for health plans: A generalized machine learning pipeline. J Biomed Inform. Nov 2021;123: 103922. https:// doi. org/ 10. 1016/j. jbi. 2021. 103922. 15. Y. Shi, Z. Wu, S. Zhang, H. Xiao, and Y. Zhao, ‘Assessing palliative care needs using machine learning approaches’, in 2021 IEEE 45th annual computers, software, and applications conference (COMPSAC), Madrid, Spain: IEEE. Jul 2021;297–302. https:// doi. org/ 10. 1109/ COMPS AC517 74. 2021. 00049. 16. Avati A, Jung K, Harman S, Downing L, Ng A, Shah NH. Improving palliative care with deep learning. BMC Med Inform Decis Mak. Dec 2018;18(S4):122. https:// doi. org/ 10. 1186/ s129110180677-8. 17. Jung K, et al. A framework for making predictive models useful in practice. J Am Med Inform Assoc. Jun 2021;28(6):1149–58. https:// doi. org/ 10. 1093/ jamia/ ocaa3 18. 18. Lau S, et al. ‘Natural Language Word-Embeddings as a glimpse into healthcare at the End Of Life’, Jul. 24, 2021. https:// doi. org/ 10. 1101/ 2021. 07. 22. 21260 874. 19. Braeunig J, et al. ‘Radar-based recognition of activities of daily living in the palliative care context using deep learning’, in 2023 IEEE EMBS international conference on biomedical and health informatics (BHI), Pittsburgh, PA, USA: IEEE. Oct 2023;1–4. https:// doi. org/ 10. 1109/ BHI58 575. 2023. 10313 506. 20. Chu T, Zhang H, Xu Y, Teng X, Jing L. Predicting the behavioral intentions of hospice and palliative care providers from real-world data using supervised learning: A cross-sectional survey study. Front Public Health. Sep.2022;10: 927874. https:// doi. org/ 10. 3389/ fpubh. 2022. 927874. 21. Blanes-Selva V, Asensio-Cuesta S, Doñate-Martà nez A, Pereira Mesquita F, Garcà a-Gómez JM. ‘User-centred design of a clinical decision support system for palliative care: Insights from healthcare professionals’. DIGITAL HEALTH. Jan 2023;9:205520762211507. https:// doi. org/ 10. 1177/ 20552 07622 11507 35. 22. Huang Y, et al. Deep Learning Prediction Model for Patient Survival Outcomes in Palliative Care Using Actigraphy Data and Clinical Information. Cancers. Apr 2023;15(8):2232. https:// doi. org/ 10. 3390/ cance rs150 82232. 23. Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N. Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLoS ONE. Apr 2017;12(4): e0174944. https:// doi. org/ 10. 1371/ journ al. pone. 01749 44. 24. Guyatt GH, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. Apr 2008;336(7650):924– 6. https:// doi. org/ 10. 1136/ bmj. 39489. 470347. AD. 25. Haimovich AD, Shah MN, Southerland LT, Hwang U, Patterson BW. Automating risk stratification for geriatric syndromes in the emergency department. J American Geriatrics Society. Jan 2024;72(1):258–67. https:// doi. org/ 10. 1111/ jgs. 18594. 26. Ramirez-Valdez EA, et al. Towards cataloguing and characterising advance care planning and end-of-life care resources. BMC Palliat Care. Nov 2022;21(1):211. https:// doi. org/ 10. 1186/ s1290402201102-3. 27. Casheekar A, Lahiri A, Rath K, Prabhakar KS, Srinivasan K. A contemporary review on chatbots, AI-powered virtual conversational agents, ChatGPT: Page 18 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 Applications, open challenges and future research directions. Computer Science Review. May 2024;1(52):100632. 28. Khalifa M, Albadawy M. Artificial intelligence for clinical prediction: exploring key domains and essential functions. Computer Methods and Programs in Biomedicine Update. 2024Mar;7: 100148. 29. Rani S, Kataria A, Bhambri P, Pareek PK, Puri V. Artificial intelligence in personalized health services for better patient care. inrevolutionizing healthcare: AI integration with iot for enhanced patient outcomes 2024 Sep 24 (pp. 89–108). Cham: Springer Nature Switzerland. 30. Dupont C, et al. Publicly Available, Interactive Web-Based Tools to Support Advance Care Planning: Systematic Review. J Med Internet Res. Apr 2022;24(4): e33320. https:// doi. org/ 10. 2196/ 33320. 31. Peters MD, Godfrey CM, Khalil H, McInerney P, Parker D, Soares CB. Guidance for conducting systematic scoping reviews. Int J Evid Based Healthc. Sep 2015;13(3):141–6. https:// doi. org/ 10. 1097/ XEB. 00000 00000 000050. (PMID: 26134548). 32. Arksey H, O’Malley L. Scoping studies: towards a methodological framework. Int J Soc Res Methodol. Feb 2005;8(1):19–32. https:// doi. org/ 10. 1080/ 13645 57032 00011 9616. 33. Tricco AC, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. Oct 2018;169(7):467–73. https:// doi. org/ 10. 7326/ M180850. 34. Alowais SA, et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ. Sep 2023;23(1):689. https:// doi. org/ 10. 1186/ s1290902304698-z. 35. Islam MR, Ahmed MU, Barua S, Begum S. A Systematic Review of Explainable Artificial Intelligence in Terms of Different Application Domains and Tasks. Appl Sci. Jan 2022;12(3):1353. https:// doi. org/ 10. 3390/ app12 031353. 36. Loh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011–2022). Comput Methods Programs Biomed. Nov 2022;226:107161. https:// doi. org/ 10. 1016/j. cmpb. 2022. 107161. 37. Bradshaw A, et al. ‘Exploring the contextual assumptions, interventions and outcomes of digital advance care planning systems: a theory of change approach to understand implementation and evaluation’, Palliat Med. Sep 2024;02692163241280134. https:// doi. org/ 10. 1177/ 02692 16324 12801 34. 38. Sridevi M, B.R AK. ‘A FRAMEWORK FOR PERFORMANCE EVALUATION OF MACHINE LEARNING TECHNIQUES TO PREDICT THE DECISION TO CHOOSE PALLIATIVE CARE IN ADVANCED STAGES OF ALZHEIMER’S DISEASE’. INDJCSE. Feb 2021,12(1):35–46. https:// doi. org/ 10. 21817/ indjc se/ 2021/ v12i1/ 21120 1140. 39. Guo W, Gao G, Dai J, Sun Q. Prediction of Lung Infection during Palliative Chemotherapy of Lung Cancer Based on Artificial Neural Network. Comput Math Methods Med. Jan 2022;2022:1–7. https:// doi. org/ 10. 1155/ 2022/ 43121 17. 40. Kum L, et al. Routine laboratory parameters to support decision on parenteral nutrition in palliative care. Front Nutr. Nov 2023;10:1173106. https:// doi. org/ 10. 3389/ fnut. 2023. 11731 06. 41. Morgan B, Tarbi E. Behavioral Economics: Applying Defaults, Social Norms, and Nudges to Supercharge Advance Care Planning Interventions. J Pain Symptom Manage. Oct 2019;58(4):e7–9. https:// doi. org/ 10. 1016/j. jpain symman. 2019. 06. 014. 42. Liu J-H, 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. Aug 2023;25:e47366. https:// doi. org/ 10. 2196/ 47366. 43. Heyman ET, et al. Improving Machine Learning 30-Day Mortality Prediction by Discounting Surprising Deaths. J Emerg Med. Dec 2021;61(6):763– 73. https:// doi. org/ 10. 1016/j. jemer med. 2021. 09. 004. 44. Patel SD, Davies A, Laing E, Wu H, Mendis J, Dijk D-J. Prognostication in Advanced Cancer by Combining Actigraphy-Derived Rest-Activity and Sleep Parameters with Routine Clinical Data: An Exploratory Machine Learning Study. Cancers. Jan.2023;15(2):503. https:// doi. org/ 10. 3390/ cance rs150 20503. 45. Wilson P, et al. ‘Improving time to palliative care review with predictive modeling: study protocol for a stepped-wedge, pragmatic randomized controlled trial’, Mar. 25, 2021. https:// doi. org/ 10. 21203/ rs.3. rs55430/ v1. 46. Li E, et al. Oncologist phenotypes and associations with response to a machine learning-based intervention to increase advance care planning: Secondary analysis of a randomized clinical trial. PLoS ONE. May 2022;17(5): e0267012. https:// doi. org/ 10. 1371/ journ al. pone. 02670 12. 47. Taseen R, Ethier J-F. Expected clinical utility of automatable prediction models for improving palliative and end-of-life care outcomes: Toward routine decision analysis before implementation. J Am Med Inform Assoc. Oct 2021;28(11):2366–78. https:// doi. org/ 10. 1093/ jamia/ ocab1 40. 48. Sandham MH, Hedgecock EA, Siegert RJ, Narayanan A, Hocaoglu MB, Higginson IJ. Intelligent Palliative Care Based on Patient-Reported Outcome Measures. J Pain Symptom Manage. May 2022;63(5):747–57. https:// doi. org/ 10. 1016/j. jpain symman. 2021. 11. 008. 49. Lai C-H, Mok PK-L, Chau W-W, Law S-W. Application of machine learning models on predicting the length of hospital stay in fragility fracture patients. BMC Med Inform Decis Mak. Jan 2024;24(1):26. https:// doi. org/ 10. 1186/ s1291102402417-2. 50. Lebedev G, et al. Artificial Neural Networks to Monitor Bedridden Patients. Procedia Computer Science. 2020;176:1923–9. https:// doi. org/ 10. 1016/j. procs. 2020. 09. 231. 51. Kawashima A, et al. Predictive Models for Palliative Care Needs of Advanced Cancer Patients Receiving Chemotherapy. J Pain Symptom Manage. Apr 2024;67(4):306-316.e6. https:// doi. org/ 10. 1016/j. jpain symman. 2024. 01. 009. 52. Zhuang Q, et al. Towards proactive palliative care in oncology: developing an explainable EHR-based machine learning model for mortality risk prediction. BMC Palliat Care. May 2024;23(1):124. https:// doi. org/ 10. 1186/ s1290402401457-9. 53. Roberts LM, Rossi LA, Zachariah F. ‘Design, EHR Integration and evaluation of clinical decision support workflows driven by a mortality prediction model to promote goal concordant care*’, Feb. 15, 2023. https:// doi. org/ 10. 1101/ 2023. 02. 14. 23285 926. 54. Lin Y-J, et al. Machine-Learning Monitoring System for Predicting Mortality Among Patients With Noncancer End-Stage Liver Disease: Retrospective Study. JMIR Med Inform. Oct 2020;8(10): e24305. https:// doi. org/ 10. 2196/ 24305. 55. Oliveira T, Silva A, Satoh K, Julian V, Leão P, Novais P. Survivability Prediction of Colorectal Cancer Patients: A System with Evolving Features for Continuous Improvement. Sensors. Sep 2018;18(9):2983. https:// doi. org/ 10. 3390/ s1809 2983. 56. Miladinia M, Zarea K, Gheibizadeh M, Jahangiri M, Karimpourian H, Rokhafroz D. A multiphase study protocol of identifying, and predicting cancer-related symptom clusters: applying a mixed-method design and machine learning algorithms. Front Digit Health. Apr 2024;6:1290689. https:// doi. org/ 10. 3389/ fdgth. 2024. 12906 89. 57. Haas O, et al. Predicting Anxiety in Routine Palliative Care Using BayesianInspired Association Rule Mining. Front Digit Health. Aug 2021;3: 724049. https:// doi. org/ 10. 3389/ fdgth. 2021. 724049. 58. Zhang Z, Jiang Z, Yin B, Chen Z, Peng X. “Song of Life”: A Comprehensive Evaluation of Biographical Music Therapy in Palliative Care by the EMWTOPSIS Method. Processes. Sep.2022;10(10):1962. https:// doi. org/ 10. 3390/ pr101 01962. 59. Blanes-Selva V, Doñate-Martà nez A, Linklater G, Garcés-Ferrer J, GarcÃa-Gómez JM. Responsive and Minimalist App Based on Explainable AI to Assess Palliative Care Needs during Bedside Consultations on Older Patients. Sustainability. Sep 2021;13(17):9844. https:// doi. org/ 10. 3390/ su131 79844. 60. Soltani M, Farahmand M, Pourghaderi AR. Machine learning-based demand forecasting in cancer palliative care home hospitalization. J Biomed Inform. Jun 2022;130: 104075. https:// doi. org/ 10. 1016/j. jbi. 2022. 104075. 61. Wang L, et al. Development and Validation of a Deep Learning Algorithm for Mortality Prediction in Selecting Patients With Dementia for Earlier Palliative Care Interventions. JAMA Netw Open. Jul 2019;2(7): e196972. https:// doi. org/ 10. 1001/ jaman etwor kopen. 2019. 6972. 62. Deschepper M, Waegeman W, Vogelaers D, Eeckloo K. Using structured pathology data to predict hospital-wide mortality at admission. PLoS ONE. Jun 2020;15(6): e0235117. https:// doi. org/ 10. 1371/ journ al. pone. 02351 17. 63. Yang TY, et al. Deep-Learning Approach to Predict Survival Outcomes Using Wearable Actigraphy Device Among End-Stage Cancer Patients. Front Public Health. Dec 2021;9: 730150. https:// doi. org/ 10. 3389/ fpubh. 2021. 730150. 64. Beeksma M, Verberne S, Van Den Bosch A, Das E, Hendrickx I, Groenewoud S. Predicting life expectancy with a long short-term memory Page 19 of 19 Ariozetal. BMC Palliative Care (2025) 24:268 recurrent neural network using electronic medical records. BMC Med Inform Decis Mak. Dec 2019;19(1):36. https:// doi. org/ 10. 1186/ s129110190775-2. 65. ElMokhallalati Y, et al. Identification of patients with potential palliative care needs: A systematic review of screening tools in primary care. Palliat Med. Sep 2020;34(8):989–1005. https:// doi. org/ 10. 1177/ 02692 16320 929552. 66. Risk J, Mohammadi L, Rhee J, Walters L, Ward PR. Barriers, enablers and initiatives for uptake of advance care planning in general practice: a systematic review and critical interpretive synthesis. BMJ Open. Sep 2019;9(9):e030275. https:// doi. org/ 10. 1136/ bmjop en2019030275. 67. Dos Reis LBM, Leles CR, Freire MDCM. Religiosity, spirituality, and the quality of life of patients with sequelae of head and neck cancer. Oral Dis. May 2020;26(4):838–42. https:// doi. org/ 10. 1111/ odi. 13284. 68. Klug M, et al. A Gradient Boosting Machine Learning Model for Predicting Early Mortality in the Emergency Department Triage: Devising a NinePoint Triage Score. J GEN INTERN MED. Jan 2020;35(1):220–7. https:// doi. org/ 10. 1007/ s1160601905512-7. 69. Kim YJ, et al. Machine learning-based model to predict delirium in patients with advanced cancer treated with palliative care: a multicenter, patient-based registry cohort. Sci Rep. May 2024;14(1):11503. https:// doi. org/ 10. 1038/ s4159802461627-w. 70. Manz CR, et al. Effect of Integrating Machine Learning Mortality Estimates With Behavioral Nudges to Clinicians on Serious Illness Conversations Among Patients With Cancer: A Stepped-Wedge Cluster Randomized Clinical Trial. JAMA Oncol. Dec 2020;6(12): e204759. https:// doi. org/ 10. 1001/ jamao ncol. 2020. 4759. 71. Robinson L, Dickinson C, Bamford C, Clark A, Hughes J, Exley C. ‘A qualitative study: Professionals’ experiences of advance care planning in dementia and palliative care, “ a good idea in theory but … ”’. Palliat Med. May 2013;27(5):401–408. https:// doi. org/ 10. 1177/ 02692 16312 465651. 72. Li RC, et al. ‘Using AI to empower collaborative team workflows: two implementations for advance care planning and care escalation’. NEJM Catalyst. Mar 2022;3(4). https:// doi. org/ 10. 1056/ CAT. 21. 0457. 73. ‘Ethical dimensions of using artificial intelligence in health care’. AMA Journal of Ethics. Feb 2019;21(2):E121–124. https:// doi. org/ 10. 1001/ amaje thics. 2019. 121. 74. Bullock K. The Influence of Culture on End-of-Life Decision Making. J Soc Work End Life Palliat Care. Feb 2011;7(1):83–98. https:// doi. org/ 10. 1080/ 15524 256. 2011. 548048. 75. Parag N, Govender R, Ally SB, ‘Promoting cultural inclusivity in healthcare artificial intelligence: a framework for ensuring diversity’. Health Management, Policy and Innovation. 2023;8(3). 76. Harishbhai M, Tilala et al. ‘Ethical considerations in the use of artificial intelligence and machine learning in health care: a comprehensive review’, Cureus. Jun 2024, https:// doi. org/ 10. 7759/ cureus. 62443. 77. Ritoré Á, et al. ‘The role of Open Access Data in democratizing healthcare AI: A pathway to research enhancement, patient well-being and treatment equity in Andalusia. Spain’, PLOS Digit Health. Sep 2024;3(9):e0000599. https:// doi. org/ 10. 1371/ journ al. pdig. 00005 99. 78. Hiriscau EI, Buzdugan E-C, Hui L-A, Bodolea C. ‘Exploring the Relationship between Frailty. Functional Status, Polypharmacy, and Quality of Life in Elderly and Middle-Aged Patients with Cardiovascular Diseases: A OneYear Follow-Up Study’, IJERPH. Feb2022;19(4):2286. https:// doi. org/ 10. 3390/ ijerp h1904 2286. 79. Inouye SK. Importance of Functional Measures in Predicting Mortality Among Older Hospitalized Patients. JAMA. Apr 1998;279(15):1187. https:// doi. org/ 10. 1001/ jama. 279. 15. 1187. 80. De Vleminck A, et al. Barriers to Advance Care Planning in Cancer, Heart Failure and Dementia Patients: A Focus Group Study on General Practitioners’ Views and Experiences. PLoS ONE. Jan 2014;9(1): e84905. https:// doi. org/ 10. 1371/ journ al. pone. 00849 05. 81. Timmons S, Fox S. ‘Palliative care for people with dementia’, in handbook of clinical neurology. Elsevier. 2023;191:81–105. https:// doi. org/ 10. 1016/ B978-0128245354. 00013-6. 82. Timmons S, Fox S, Drennan J, Guerin S, Kernohan WG. ‘Palliative care for older people with dementia—we need a paradigm shift in our approach’. Age and Ageing. Mar 2022;51(3):afac066. https:// doi. org/ 10. 1093/ ageing/ afac0 66. 83. Blanes-Selva V, Doñate-Martà nez A, Linklater G, Garcà a-Gómez JM. Complementary frailty and mortality prediction models on older patients as a tool for assessing palliative care needs. Health Informatics J. Jan.2022;28(2):146045822210925. https:// doi. org/ 10. 1177/ 14604 58222 10925 92. 84. Mandrekar JN. Receiver Operating Characteristic Curve in Diagnostic Test Assessment. J Thorac Oncol. Sep.2010;5(9):1315–6. https:// doi. org/ 10. 1097/ JTO. 0b013 e3181 ec173d. 85. Steyerberg EW, et al. Assessing the Performance of Prediction Models: A Framework for Traditional and Novel Measures. Epidemiology. Jan 2010;21(1):128–38. https:// doi. org/ 10. 1097/ EDE. 0b013 e3181 c30fb2. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.