The PALLI-AI Framework: An Evidence-Informed Checklist for Designing and Reporting Trustworthy Artificial...

Background/Objectives: Artificial intelligence (AI) is being applied across palliative and end-of-life care, including prognostication, symptom detection, clinical documentation, communication support, decision support and care coordination. General AI reporting guidelines exist, but none captures the value-sensitive requirements specific to palliative care, which is integrated from the diagnosis of a serious illness and spans the whole trajectory, alongside disease-directed treatment, to the end of life and bereavement. We aimed to develop an evidence-informed checklist to guide the design and the transparent reporting of AI tools in this setting. Methods: We conducted a narrative review of the literature on AI in palliative care, prognostic communication, surrogate decision-making, digital-health co-design, algorithmic fairness, explainability, AI governance and existing reporting guidelines (TRIPOD+AI, CONSORT-AI, SPIRIT-AI, DECIDE-AI). During revision, the search was formalised and re-executed in PubMed (25 August 2026), each item was graded for certainty of evidence, and the checklist was piloted on four published studies. Rather than reusing generic requirements, we derived each item from a tension specific to serious illness, named the failure modes those tensions produce, and mapped where the framework diverges from generic guidance. Results: PALLI-AI comprises seven domains-Purpose set by goals of care; Alignment with total suffering and dignity; Lived experience, family and bereavement; Learning data and the equity of access; Interpretability and prognostic communication; Accountability in serious illness; and Implementation and palliative outcomes-operationalised as 22 items, each with a researcher and a developer formulation. We name six failure modes specific to this setting, including the self-fulfilling prognosis, the algorithmic surrogate and abandonment by automation, and we set out, domain by domain, what the framework adds beyond existing guidelines. In the pilot, all items were ratable; safeguards against the self-fulfilling prognosis, decedent data governance and answerability were unreported across studies. Conclusions: PALLI-AI is a pragmatic, non-prescriptive checklist intended to raise the trustworthiness of AI tools for people approaching the end of life. It is a starting point for formal consensus development and empirical validation, not a finished standard.