Challenges in AI enabled Supportive Oncology

Supportive oncology determines how patients actually live through and beyond cancer treatment: pain and symptom control, procedural distress, toxicity and mucositis management, nutrition, psychosocial and spiritual needs, rehabilitation, survivorship and end-of-life care. Much of this work is nursing-led, delivered at the bedside, in the clinic and increasingly in the patient’s home. At population level it is also one of the least well-measured parts of cancer control: symptom burden is self-reported and inconsistently coded, supportive care utilization is poorly captured in registries, and access varies sharply between health systems, between children and adults, and between social groups. These are epidemiological and health services problems before they are technical ones.Artificial intelligence and digital technologies are now entering this space from several directions at once. Immersive and interactive tools, including virtual reality, biofeedback-driven games and adaptive distraction, are being tested against procedural pain, fear and anxiety, with children and adolescents among the most-studied groups. Electronic patient-reported outcome systems, wearables and web-based needs assessments generate continuous symptom data and route it back into care. Conversational agents and large language models are being trialled as patient and family education tools, as symptom coaches, and as drafting aids for clinicians. Prediction models built on cohort and real-world data flag treatment toxicity, symptom deterioration, unplanned admission and the need for earlier palliative care referral. Upstream, similar methods are applied to screening uptake, risk-adapted follow-up and survivorship care planning.Very little of this is routine practice, and the obstacles are consistent across settings. Trials are typically small, single-center and short, with heterogeneous outcome measures that resist synthesis. Symptom instruments validated in one language and culture are transferred to another without adequate cross-cultural adaptation, and AI tools inherit that weakness. Supportive care data are noisy, incomplete and rarely structured, so models reproduce the blind spots of the records that trained them. Nurses and other frontline staff, the people who will actually operate these tools, report uneven AI literacy, limited involvement in design, and unresolved ethical concerns — over data privacy and governance, the transparency and explainability of model outputs, the degree of human oversight retained at the bedside, and who is accountable when an AI-supported clinical decision goes wrong. Patients and families, particularly children and their caregivers, are rarely asked what they want automated. Algorithmic bias risks scaling the documented under-treatment of pain and under-referral to palliative care in minority, rural and lower-income populations. And an intervention that works in a well-resourced center may be unaffordable or unimplementable in the health system next door.This Research Topic asks what is genuinely blocking AI and digital technologies from improving supportive and palliative cancer care, and what the evidence says about getting past it. We welcome work across the lifespan, in pediatric, adult and geriatric oncology, and across health systems, including publicly funded and resource-constrained settings. Rigorous qualitative and mixed-methods studies are as welcome as quantitative ones, and we actively encourage submissions reporting negative, null or mixed results.Themes include, but are not limited to:- Randomized, feasibility and pragmatic trials of AI-enabled or digital supportive care interventions: virtual reality and immersive distraction, biofeedback, apps, conversational agents, remote monitoring- Procedural pain, fear and anxiety in children, adolescents and adults, including family- and caregiver-centered interventions- Symptom surveillance at scale: electronic patient-reported outcomes, wearables, web-based needs assessment, automated triage, alert burden and effect on care utilization- Development, digital adaptation and cross-cultural validation of symptom and needs assessment instruments, including AI-assisted scoring and short-form derivation- Large language models and conversational agents for patient, family and clinician communication, including acceptability, safety evaluation, failure modes and guardrail design- Nursing and allied health workforce perspectives: AI literacy, readiness, trust, ethical concerns, training needs, workload and role change- Prediction models for toxicity, mucositis, symptom deterioration, functional decline, unplanned admission and palliative care referral, with external or prospective validation- Integrated and shared care across hospital, primary care and home, and the role of digital tools at those interfaces- Survivorship and long-term consequences of treatment, including rehabilitation, stoma and ostomy care, late effects and return to daily life- Cohort, registry and real-world studies of supportive care need, access, utilization and outcomes, including disparities by geography, income, ethnicity, age and health system- Equity and implementation in low- and middle-income countries and publicly funded health systems, including cost, infrastructure and workforce constraints- Qualitative and mixed-methods research on patient, family and clinician experience of AI-supported care- Reporting, evaluation and governance standards in practice, including TRIPOD+AI, DECIDE-AI and SPIRIT-AI/CONSORT-AI, and the ethical, regulatory and policy frameworks governing deployment: data privacy, transparency and explainability, human oversight, and accountability for AI-supported clinical decisionsWe welcome Original Research, Systematic Reviews, Reviews and Mini Reviews, Methods, Clinical Trials, Study Protocols, Brief Research Reports, Perspectives and Opinions.Manuscripts consisting solely of bioinformatics or computational analysis of public genomic, transcriptomic, or metabolomic databases, without robust and relevant validation (an independent clinical or patient cohort, or biological validation in vitro or in vivo not derived from public databases), are out of scope for this Research Topic. Purely descriptive metabolite profiling without a functional or clinical link to T-cell immunity is also outside the scop Topic editors: Veronica Nanton, Remziye Semerci, Raquel de Souza Ramos Source: https://www.frontiersin.org/research-topics/85360/challenges-in-ai-enabled-supportive-oncology