Palumbo, Peter

Artificial intelligence (AI) is now applied across oncology, including imaging, pathology, therapy selection, toxicity assessment, and survival modeling, and large language models (LLMs) have been adopted particularly quickly. Reported physician use of AI roughly doubled between 2023 and 2026. The frameworks used to evaluate these tools have not advanced at the same rate. Many are available, but each was designed for a single stage of the AI lifecycle, and none directly answers the question most relevant to a treating clinician: whether a given tool is appropriate for the individual patient under care. We reviewed the frameworks in current use, including TRIPOD+AI, PROBAST+AI, CLAIM, SPIRIT-AI, CONSORT-AI, DECIDE-AI, MINIMAR, and CREMLS, together with the oncology-specific ESMO EBAI and the newer LLM-specific guidance (TRIPOD-LLM, MI-CLEAR-LLM, CHART, and ESMO ELCAP). Each addresses part of the evaluation problem, but none is sufficient on its own for oncology, where the standard of care changes rapidly, assays drift, biomarker-defined subgroups are small, and a model validated in one period may perform poorly in the next. LLMs introduce additional challenges, including sensitivity to prompting, undisclosed vendor updates, behavioral drift, and hallucination rates approaching 50% on clinical quality-assurance tasks. The predictive-model frameworks were not designed to capture these failure modes. The central argument of this review is that the frameworks the field requires already exist; what is missing is their mandatory adoption. We propose that journals and regulators move from recommending these frameworks to requiring their use, and we outline a cancer-aware evaluation pathway together with the specific responsibilities of authors, reviewers, journals, and regulators in implementing it.

1617 Background: Large language models (LLMs) are increasingly evaluated for oncology decision support, yet guideline concordance and hallucinations remain variable across cancer types and tasks. Retrieval-augmented generation (RAG) may improve safety by constraining the evidence base used during generation. Prior oncology LLM studies report discordance with guideline-based care, but few describe a reproducible health-system “vetting mechanism” that couples guideline-only retrieval with clinician-scored safety rubrics across multiple oncology domains. Methods: We assembled 316 de-identified oncology cases in a tumor-board template: medical oncology (breast, GI, GU, leukemia; n=216), gynecologic oncology (n=50), and neuro-oncology/CNS metastatic cancer (n=50). Each case was run on three systems: baseline GPT-4 (O1), NCCN-guideline anchored RAG (O2; guideline-constrained retrieval), and a literature-anchored system (OpenEvidence; O3). Oncologists independently scored outputs using a modified Generative Performance Score (mGPS; −1 to +1) combining guideline concordance with hallucination penalties; readability/rationality was scored on a 1–5 Likert scale. Inter-rater reliability was assessed (κ =.63). This workflow was designed as a pre-deployment model credentialing step for health-system governance (source policy → case bank → clinician-scored acceptance criteria). Results: Across all 316 cases (case-averaged), O2 achieved the highest mean total mGPS (0.75) compared with O1 (0.43) and O3 (0.59). In the medical oncology cohort (n=216), O2 outperformed O1 and O3 (mean mGPS 0.72 vs 0.35 and 0.54), with higher guideline concordance (0.85 vs 0.68 and 0.75) and lower hallucination penalties (−0.09 vs −0.25 and −0.19); readability was also highest for O2 (4.17 vs 3.22 and 3.86). In gynecologic oncology (n=50), O2 again performed best (mGPS 0.827) relative to O3 (0.697) and O1 (0.647). In the CNS cohort (n=50), mean composite mGPS favored O2 (0.82) over O3 (0.68) and O1 (0.55), while readability remained stable across models (3.55–3.73). Severe hallucination/unsafe recommendation rate: O1 [26%], O2 [5%], O3 [17%] (pre-specified definition; e.g., major fabricated citation or guideline-discordant harmful recommendation). Conclusions: In a 316-case, multi-domain clinician-scored evaluation, limiting retrieval to oncology guideline sources improved safety-aligned performance relative to unconstrained generation or literature-only anchoring. A guideline-constrained RAG configuration paired with clinician-scored rubrics and tail-risk reporting can serve as a practical health-system vetting mechanism for governance, re-validation, and change control of LLM decision support, including in resource-limited settings where reliable guideline-consistent recommendations are essential.