Inteligência artificial como ferramenta para inclusão de pacientes oncológicos em cuidados paliativos: uma revisão de escopo

Abstract

Palliative Care (PC) is fundamental to improving the quality of life for patients with life-threatening illnesses, such as cancer, and their families. This approach, which advocates for symptom management and psychosocial and spiritual support, has been reinforced in Brazil with the establishment of the National Policy on Palliative Care (2024), consolidating its relevance within the Brazilian Unified Health System (SUS). In the oncological context, PC is essential for aligning therapeutic decisions with patient values. Concurrently, Artificial Intelligence (AI) is emerging as a promising tool in healthcare, with the potential to optimize diagnostics, predictive analyses, and personalized care. However, a gap is noted in the scientific literature regarding how AI is specifically being used for the inclusion of oncology patients in PC, as well as its actual benefits and outcomes. Objective: To map the existing literature on AI technologies applied to the inclusion of cancer patients in PC and to analyze the subsequent clinical, operational, and economic outcomes. Methodology: A scoping review was conducted following the Joanna Briggs Institute (JBI) and PRISMA-ScR guidelines, with a protocol registered on the Open Science Framework. A systematic search was performed across prominent databases such as PubMed, Scopus, Web of Science, and IEEE Xplore, as well as in the grey literature, without date or language restrictions, to ensure comprehensive coverage of the topic. Results: The analysis resulted in the inclusion of five studies, published between 2021 and 2025, predominantly conducted in high-income countries, such as the United States and Canada, and based on retrospective cohort designs. The identified AI technologies were primarily machine learning models, augmented intelligence, and natural language processing, applied to predict mortality in the last year of life and thereby identify early eligibility for PC. These models utilized extensive datasets, including clinical, demographic, socioeconomic, and textual data from electronic health records, with a focus on adult populations with advanced cancer. The outcomes assessed ranged from time to referral and mortality to symptom control and economic impacts. Barriers to implementation included difficulty in accessing unstructured data and low physician confidence in the algorithms. Conclusion: AI demonstrates significant potential to support the timely inclusion of oncology patients in PC, aiding clinical decision-making, especially for patients in their last year of life. However, the concentration of studies in high-income, hospital-based settings, a scarcity of research in public health systems, and a greater emphasis on technical aspects over practical implementation reveal critical gaps. The safe and equitable advancement of AI in this field requires interdisciplinary approaches, clear regulation, and strategies that prioritize dignity and equity at the end of life. This study stands as the first scoping review on the topic, offering essential insights for future research and the development of informed public policies.