Pith. sign in

REVIEW 4 major objections 6 minor 32 references

An Agentic Generative Large Language Model for Treatment Planning of Colorectal Cancer

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read A domain-adapted 8-billion-parameter language model generates colorectal cancer treatment plans that blind oncologists rate as statistically indistinguishable from expert-written plans.

desk verdict A serious, well-engineered clinical LLM paper whose expert-level claim is real but rests on a same-institution gold standard and overlapping evaluators; worth refereeing, not desk-rejecting. read the letter →

arxiv 2608.09142 v1 pith:HWJ3XFNV submitted 2026-08-10 cs.CL

classification cs.CL
keywords colorectalcancertreatmentplanningagenticLLMretrieval-augmentedgenerationreinforcementlearningdomainadaptationNCCNguidelinesclinicalevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

GatorOnco is an 8-billion-parameter language model adapted to oncology that generates treatment plans for colorectal cancer. The paper's core claim is that, in a blind evaluation by five oncologists, GatorOnco's plans are rated statistically on par with expert-written plans for correctness, currency, and safety, and higher for readability and completeness, while outperforming much larger open-source models. The significance would be that specialized, locally deployable models, rather than massive general-purpose ones, can meet expert-level standards in a high-stakes clinical task, provided they are grounded in current guidelines through agentic retrieval. The evidence comes from 79 real patient cases at a single health system.

What carries the argument

The central mechanism is an agentic retrieval-augmented generation loop in which the model, after continued pre-training and instruction-following fusion via model merging, is optimized with group relative policy optimization under a hierarchical clinical reward. During generation, the model emits a structured search action when it has an information gap, the environment appends top-ranked chunks from an indexed NCCN guideline database, and the model resumes its chain-of-thought before writing the plan. The hierarchical reward first requires the correct high-level therapy modality and only then rewards correct drug-entity alignment, so that a plan is never rewarded if the wrong modality is chosen. This design grounds reasoning in time-sensitive guidelines and enforces a safety-first ordering of clinical decisions.

What would settle it

A cross-institution study in which treatment plans from GatorOnco, from oncologists at a different health system, and from a nationally reviewed gold standard are rated by an independent panel unaware of the source would settle the claim: if GatorOnco's correctness or safety ratings fall significantly below the expert plans, the expert-level claim is refuted. A simpler check is to remove UF Health clinical notes from the pre-training corpus and see whether the parity advantage disappears, which would indicate dependence on the specific training data.

Watch

Extended reading notes

Core claim

On the paper's own terms, GatorOnco achieves expert-level treatment planning for colorectal cancer as judged by the same institution's oncologists. In the blinded comparison, GatorOnco scored 4.09 versus 4.11 for correctness (P = 0.921), 4.04 versus 3.98 for currency (P = 0.478), and 4.22 versus 4.22 for safety (P = 0.999); it exceeded oncologists on readability (4.46 vs 4.19) and completeness (3.91 vs 3.52). The model also outperformed Llama-3.3-70B and MedGemma-27B on treatment-option determination and narrative plan quality. The paper interprets this as evidence that domain adaptation on real clinical text, combined with safety-oriented reinforcement learning and agentic retrieval of NCCN guidelines, can close the gap between generative AI and expert performance in cancer treatment planning.

Load-bearing premise

The load-bearing premise is that the oncologist-written or oncologist-edited treatment plans from UF Health are a valid and unbiased gold standard, and that the same institution's oncologists, several of whom helped annotate the training and test cases, can rate plans blindly; if the gold-standard notes are not the true standard of care or raters recognize their own or colleagues' phrasing, then statistical parity with those notes does not prove expert-level performance.

Editorial extensions

If this is right

  • If the central claim is correct, an 8-billion-parameter model with agentic retrieval can match expert oncologists on treatment planning, implying that scale is not the main bottleneck for high-stakes clinical text generation.
  • The full pipeline (continued pre-training, model merging, two-stage post-training, and agentic RL with hierarchical rewards) becomes a reusable recipe for other disease domains.
  • Because the agent retrieves from current NCCN guideline versions, treatment plans can be audited by linking each recommendation to specific guideline chunks and patient-specific evidence.
  • The reported readability and completeness advantages over expert notes suggest AI-generated plans may serve as drafting aids that reduce documentation burden.
  • The authors state that GatorOnco remains a prototype; practical deployment would require fail-safes, uncertainty quantification, and deferral policies.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same pipeline applied to other cancer types and other health systems would test whether the expert-level parity is due to the method or to the single-institution training and rating context.
  • Because raters and ground-truth writers belong to the same institution, the parity scores may partly reflect shared documentation style; a cross-institutional panel could separate house style from clinical quality.
  • The authors' call for machine-readable guidelines (e.g., FHIR) points to a concrete engineering bottleneck: manual parsing of PDF guidelines is substantial, so any shift to structured guideline releases could accelerate deployment of such agents.
  • The hierarchical reward design could transfer to other safety-critical generation tasks, where fine-grained rewards are gated behind coarse categorical correctness.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. GatorOnco is an 8B-parameter Llama-3.1-based LLM adapted to colorectal cancer treatment planning through continued pretraining on a 282B-token biomedical corpus (166B tokens from UF Health EHRs), model merging, supervised fine-tuning, and agentic reinforcement learning with interleaved retrieval of NCCN guidelines. The model is evaluated on 79 held-out UF Health CRC cases for treatment-category determination (F1 up to 0.924) and narrative plan generation using automatic lexical, semantic, and faithfulness metrics, and in a blinded human evaluation by five UF Health oncologists that compared GatorOnco, Llama-3.1-8B-Instruct, and expert gold-standard plans across five Likert dimensions. The authors report that GatorOnco matches expert oncologists on correctness, currency, and safety and exceeds them on readability and completeness.

Significance. If the results held as stated, GatorOnco would be an important demonstration that a compact domain-adapted model with agentic retrieval can approach expert-level treatment planning. The work combines an unusually large clinical corpus, a structured reward hierarchy that enforces modality-to-regimen reasoning, and a stratified held-out test design with appropriate mixed-effects and aligned-rank-transform analyses. The authors also commit to releasing code and a de-identified test set. However, the headline 'expert-level' claim is currently supported by a same-institution reference standard rated in part by oncologists who also contributed to the training annotations, and the parity conclusion is drawn from non-significant p-values without an equivalence margin. These issues do not undermine the engineering contributions or the automatic-evaluation results, but they do constrain how the central claim can be stated.

major comments (4)
  1. [Methods, 'Expert annotation and quality control' and 'Blind evaluation by UF Health Oncologists'; Author contributions] The central expert-level claim rests on a reference standard that is not independent of the model's training signal. The gold-standard plans were written or edited by UF Health oncologists (Methods, 'Expert annotation and quality control'), the model was trained with SFT and RL to reproduce those same annotations, and two of the five blind raters (TJ, TL) are listed both as case annotators and as human evaluators in the Author Contributions. Consequently, parity with the gold standard could reflect raters' familiarity with their own or colleagues' notes and with UF-specific documentation conventions rather than independent evidence of expert-level care. The blind protocol prevents raters from knowing which system produced a plan, but no check is reported for whether raters recognized the source of ground-truth notes. Please report such a recognition check, re-analyze the primary comparisons excluding raters who participated in annotation, or obtain independent external oncologist ratings; at minimum, the 'expert-level' claim should be rephrased to reflect that the comparison is to institutional expert annotations.
  2. [Results, 'GatorOnco achieved expert-level treatment planning for CRC...' and Methods, 'Blind evaluation by UF Health…] The conclusion that GatorOnco is 'statistically comparable' or 'on par' with experts for correctness (4.09 vs. 4.11, P = 0.921), currency (4.04 vs. 3.98, P = 0.478), and safety (4.22 vs. 4.22, P = 0.999) is inferred entirely from non-significant p-values. This does not establish equivalence: a large confidence interval or a low-powered test can also yield P > 0.05. The manuscript provides neither a pre-specified non-inferiority margin nor two one-sided tests (TOST) or confidence intervals for the differences. In addition, the composite mixed-model contrast shows GatorOnco significantly higher than experts (β = 0.136, P < 0.01), which is inconsistent with a pure parity claim. Please add equivalence testing with an explicit margin and report confidence intervals for the dimension-specific differences.
  3. [Methods, 'Blind evaluation by UF Health Oncologists'; Abstract] The human evaluation compares only GatorOnco, Llama-3.1-8B-Instruct, and the expert ground truth; MedGemma-27B and Llama-3.3-70B-Instruct were evaluated only with automatic metrics. The abstract's claim that GatorOnco 'significantly outperformed open-source LLMs' is therefore supported for only a single human-evaluated baseline, not for the class of open-source LLMs. Either include the additional baselines in the blind human evaluation or restrict the claim to the parameter-matched Llama-3.1-8B model.
  4. [Discussion (Limitations) and Abstract] The abstract and title claim expert-level treatment planning without institutional or temporal qualification, yet the evaluation is a single-institution, single-disease, 79-case study with a reference standard that was edited by the same group to conform to contemporaneous NCCN guidelines. The Limitations section acknowledges the single-institution scope, but the abstract should carry the same qualification, for example by stating 'expert-level on this institutional test set' or 'comparable to UF Health oncologists' rather than the unqualified 'expert-level treatment planning'.
minor comments (6)
  1. [Results, 'GatorOnco achieved expert-level treatment planning for CRC...'] The phrase 'batter ratings' should be 'better ratings'.
  2. [Methods, 'Optimization Objective and Hierarchical Reward'] The GRPO loss equation does not render in the text, and the reward formula is garbled (e.g., the definition of R and the Dice coefficient term are not readable); please provide a clean mathematical formulation, as the reward design is central to the method's reproducibility.
  3. [Table 2 and Methods, 'Generating narrative treatment plan sections'] The 'Overall' score is described as the arithmetic mean of the metrics, but it is not stated whether all seven metrics are weighted equally; please specify the aggregation rule explicitly.
  4. [Author Contributions and Methods] The overlap between annotators (TJ, TL, TG, LE) and raters (TJ, TL, LG, CS, OM) is not disclosed in the Methods or Limitations; please add an explicit disclosure and discuss its implications for the blind evaluation.
  5. [Code availability] The code availability section refers to 'GatorTronGPT' training code, while the model is called GatorTronLlama; please align the model names to avoid confusion.
  6. [Abstract] The phrase 'statistically comparable' should be replaced with 'not significantly different' unless equivalence tests with confidence intervals are added, to avoid a common statistical misinterpretation.

Circularity Check

1 steps flagged · score 4.0 of 10

Expert-level claim is validated against ground truth produced by the same UF oncologists whose annotations trained and rewarded the model; partial circularity in the evaluation reference.

  1. self definitional [Results, 'GatorOnco achieved expert-level treatment planning for CRC...'; Methods, 'Expert annotation and quality control' and 'Blind evaluation by UF Health Oncologists'; Author contributions]
    "Five UF Health oncologists manually reviewed treatment plans generated by GatorOnco, Llama3.1-8B-Instruct, and the ground-truth treatment plans reviewed by oncologists. ... When the treatment plan is incomplete or contains clear errors, we ask oncologists to complete or modify it as needed. ... TJ, LE, TL, and TG are oncologists who were involved in patient case review and annotation. TJ, TL, LG, CS, and OM are the oncologists who performed the human evaluation."

    The claimed outcome 'expert-level treatment planning' is operationally defined as statistical parity with ground-truth plans written or edited by UF Health oncologists. GatorOnco was trained to reproduce exactly this signal: SFT on 276 UF Health CRC cases annotated by the same oncologist group, RL with a hierarchical reward co-designed with UF oncologists, and post-training trajectories reviewed and corrected by oncologists. Two of the five raters (TJ, TL) also participated in case annotation, and all raters are from the same institution whose note style the model was optimized to emulate. Thus 'on par with expert oncologists' largely restates the model's fit to the same expert judgments used to build it.

full rationale

The automatic F1 treatment-option benchmarks, the text-quality metrics, and the comparisons against three external baselines are self-contained: the held-out test labels were fixed before model development, the stratified split is balanced, and no numerical result reduces by construction to a fitted parameter. The genuinely circular element is the load-bearing 'expert-level' claim. The reference standard is the UF Health oncologist-written and oncologist-edited ground-truth plans; the model's SFT data, RL reward, and post-training corrections all originate from the same UF Health oncologist group, and two of the five blind raters also served as annotators. Consequently, parity with that reference largely measures reproduction of the same institutional expert signal rather than an independent gold standard. The blind, randomized protocol and mixed-effects modeling are sound practices and prevent direct source recognition from biasing the comparison, but they do not remove the dependence of the evaluation criterion on the training signal. The paper's 'statistically comparable' inference from non-significant P-values is a statistical risk (equivalence is not established without a pre-specified non-inferiority margin or confidence intervals), but this is not circularity. Self-citations such as the GatorTron reference are not load-bearing for the central claim. Overall score 4: partial circularity in the evaluation reference, with substantial independent benchmark content.

Assumptions & free parameters 3 free parameters · 7 assumptions · 0 invented entities

The central evaluation rests on the expert-annotated UF Health dataset and the NCCN guidelines as the standard of care. The model introduces no new biological or physical entities, but the benchmark and raters are institution-internal, so the evaluation is only partially independent of the training signal.

free parameters (3)
  • omega1 (modality match reward weight) = 1.0
    Chosen by hand in the hierarchical reward function described in Methods, 'Hierarchical oncology-oriented reward'; it sets the priority of correct high-level modality over granular regimen details.
  • omega2 (entity-level Dice overlap reward weight) = 0.5
    Chosen by hand in the same reward function; it controls partial credit for identifying some but not all drugs in a regimen, modeled on clinical board examination logic.
  • Clinical note token cutoff for cohort inclusion = 1,000 tokens
    Cohort curation in Methods, 'UF Health CRC cohort' filters out notes with fewer than 1,000 tokens; this arbitrary threshold affects which patient encounters enter the training and test sets.
assumptions (7)
  • domain assumption NCCN colon v4.2025 and rectal v3.2025 are the correct standards for evaluating currency and safety.
    The agentic RAG and the reward function ground treatment plans in these specific guideline versions, and the currency evaluation is defined as adherence to them.
  • domain assumption The expert-annotated ground-truth treatment plans are correct, complete, and representative of expert-level care.
    All automatic and human evaluations compare model output against these annotations; there is no external audit of the annotations themselves.
  • domain assumption Five-point Likert ratings by five UF Health oncologists are a valid measure of clinical quality across the five dimensions.
    The central 'expert-level' claim is operationalized as these ratings, and inter-rater agreement is only moderate for correctness, currency, and safety (alpha 0.626 to 0.679).
  • domain assumption The UF Health test set is representative of CRC treatment planning beyond UF Health.
    The paper itself concedes a single-institution limitation; portability across Epic implementations, documentation styles, and patient populations is assumed for broader claims.
  • domain assumption The de-identification and exclusion of CRC cohort notes from pretraining prevented data leakage.
    The paper states that CRC cohort notes were excluded, but no formal leakage audit is provided to verify that test-case information did not influence pretraining.
  • domain assumption Automatic metrics such as AlignScore and UniEval measure factual faithfulness in oncology text.
    These metrics are used as primary evidence for narrative plan quality, but they have not been clinically validated for this task.
  • standard math GRPO and linear mixed-effects model assumptions hold for the reported comparisons.
    The paper applies standard statistical machinery without formal verification of distributional assumptions; this is a normal but unproven background assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Agentic Generative Large Language Model for Treatment Planning of Colorectal Cancer." pith.science (2026). https://pith.science/paper/HWJ3XFNV

@misc{pith2026260809142,
  author       = {Pith},
  title        = {Pith review of: An Agentic Generative Large Language Model for Treatment Planning of Colorectal Cancer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HWJ3XFNV}},
  note         = {Machine review of arXiv:2608.09142}
}
read the original abstract

Treatment planning in precision oncology requires synthesizing heterogeneous patient information with rapidly evolving clinical guidelines to ensure guideline-concordant care. While large language models (LLMs) show promise in many diagnostic tasks, their adoption for high-stakes treatment planning is hindered by complex reasoning, adherence to timely clinical guidelines, and safety concerns. In this study, we present GatorOnco, an agentic LLM for colorectal cancer (CRC) treatment planning. GatorOnco is developed using a total of 282 billion tokens of biomedical text, including healthcare system-scale clinical text comprising 166 billion tokens from UF Health. We implemented a domain-adaptation method that integrates pre-training, model merging, a two-stage post-training approach, and agent-based reinforcement learning. An agentic retrieval-augmented generation (RAG) approach dynamically integrates time-sensitive clinical guidelines into the reasoning process. In a blind, randomized clinical evaluation conducted by five UF Health oncologists, GatorOnco significantly outperformed open-source LLMs (P < 0.01) and achieved expert-level performance comparable to UF Health oncologists. Compared with expert oncologists, GatorOnco received significantly higher ratings for readability (4.46 vs. 4.19, P < 0.01) and completeness (3.91 vs. 3.52, P < 0.01), while showing statistically comparable performance in correctness (4.09 vs. 4.11, P = 0.921), currency (4.04 vs. 3.98, P = 0.478), and safety (4.22 vs. 4.22, P = 0.999). These findings demonstrate that integrating agentic reasoning with large-scale domain adaptation can help bridge the gap for generative AI in high-stakes cancer treatment planning.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 15 canonical work pages

  1. [1]

    reasoning

    1 An agentic generative large language model for treatment planning of colorectal cancer Authors: Mengxian Lyu1†, Cheng Peng1†, Tim Jang7,12, Ang Li1, Mengyuan Zhang1, Ziyi Chen1, Leighton Elliott7,12, Tianshi Liu7,12, Lidice Galindo7,12, Chiranjeevi Sainatham7,12, Oscar F. Borja-Montes7,12, Kaleb E. Smith3, Ying Zhang4, Lichao Sun9, Jiang Bian10, 11, Glo...

  2. [2]

    forgetting issue

    Comparison of GatorOnco with general-purpose Llama and medical LLM MedGemma. Metrics are categorized into lexical (BLEU, Meteor, ROUGE-L), semantic (BERTScore, BLEURT), and faithfulness (AlignScore, UniEval) dimensions. GatorOnco achieved expert-level treatment planning for CRC, per the evaluation of five UF Health oncologists We conducted a blinded stati...

  3. [4]

    The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding institutions

    Guideline-derived QA pairs will be further released on Huggingface Datasets Computer code The computer codes for GatorOnco are available from: https://github.com/uf-hobi-informatics-lab/GatorOnco The computer codes to train GatorTronGPT models are available from: https://github.com/NVIDIA-NeMo/NeMo/blob/main/tests/collections/llm/llama3_pretraining.py The...

  4. [5]

    22 Sellergren A, Kazemzadeh S, Jaroensri T, et al

    http://arxiv.org/abs/2402.03300. 22 Sellergren A, Kazemzadeh S, Jaroensri T, et al. MedGemma Technical Report. arXiv [cs.AI]. 2025; published online July

  5. [7]

    42 Douze M, Guzhva A, Deng C, et al

    http://arxiv.org/abs/2212.03533. 42 Douze M, Guzhva A, Deng C, et al. The Faiss library. arXiv [cs.LG]. 2025; published online Oct

  6. [8]

    44 Jin B, Zeng H, Yue Z, et al

    http://arxiv.org/abs/2508.10925. 44 Jin B, Zeng H, Yue Z, et al. Search-R1: Training LLMs to reason and leverage search engines with reinforcement learning. arXiv [cs.CL]. 2025; published online Aug

  7. [9]

    14 Chen S, Kann BH, Foote MB, et al

    DOI:10.1038/s41586-025-08869-4. 14 Chen S, Kann BH, Foote MB, et al. Use of artificial intelligence chatbots for cancer treatment information. JAMA Oncol 2023; 9: 1459–62. 31 15 Froicu E-M, Creangă-Murariu I, Afrăsânie V-A, et al. Artificial Intelligence and decision-making in oncology: A review of ethical, legal, and informed consent challenges. Curr Onc...

  8. [10]

    27 Asgari E, Montaña-Brown N, Dubois M, et al

    DOI:10.1093/jamia/ocaf213. 27 Asgari E, Montaña-Brown N, Dubois M, et al. A framework to assess clinical safety and hallucination rates of LLMs for medical text summarisation. NPJ Digit Med 2025; 8:

Show all 32 references
  1. [12]

    23 Likert R

    http://arxiv.org/abs/2507.05201. 23 Likert R. A technique for the measurement of attitudes. Arch Psychol (Chic)

  2. [15]

    52 Wobbrock JO, Findlater L, Gergle D, Higgins JJ

    http://arxiv.org/abs/2505.10483. 52 Wobbrock JO, Findlater L, Gergle D, Higgins JJ. The aligned rank transform for nonparametric factorial analyses using only anova procedures. In: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York, NY, USA: ACM,

  3. [17]

    Defining an evidence-based strategy for streamlining cancer multidisciplinary team meetings

    4 Soukup T, Stewart GD, Lamb BW. Defining an evidence-based strategy for streamlining cancer multidisciplinary team meetings. Lancet Oncol 2023; 24: 1061–3. 5 Robinson TP, Kaiser K, Lark M, et al. NCCN guideline concordance in colon and rectal cancer patients within a comprehe...

  4. [19]

    32 30 An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-Tuning

    http://arxiv.org/abs/1612.00796. 32 30 An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-Tuning. . 31 Hager P, Jungmann F, Holland R, et al. Evaluation and mitigation of the limitations of large language models in clinical decision-ma...

  5. [21]

    49 Sellam T, Das D, Parikh A

    http://arxiv.org/abs/1904.09675. 49 Sellam T, Das D, Parikh A. BLEURT: Learning robust metrics for text generation. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA, USA: Association for Computational Linguistics,

  6. [22]

    38 Chen Z, Cano AH, Romanou A, et al

    http://arxiv.org/abs/2101.00027. 38 Chen Z, Cano AH, Romanou A, et al. MEDITRON-70B: Scaling medical pretraining for large language models. arXiv [cs.CL]. 2023; published online Nov

  7. [23]

    43 OpenAI, Agarwal S, Ahmad L, et al

    DOI:10.48550/arXiv.2401.08281. 43 OpenAI, Agarwal S, Ahmad L, et al. gpt-oss-120b & gpt-oss-20b Model Card. arXiv [cs.CL]. 2025; published online Aug

  8. [25]

    36 Johnson AEW, Pollard TJ, Shen L, et al

    http://arxiv.org/abs/2406.17557. 36 Johnson AEW, Pollard TJ, Shen L, et al. MIMIC-III, a freely accessible critical care database. Sci Data 2016; 3: 160035. 37 Gao L, Biderman S, Black S, et al. The Pile: An 800GB dataset of diverse text for language modeling. arXiv [cs.CL]. 2...

  9. [27]

    39 health360/Healix-Shot · Datasets at Hugging Face

    http://arxiv.org/abs/2311.16079. 39 health360/Healix-Shot · Datasets at Hugging Face. https://huggingface.co/datasets/health360/Healix-Shot (accessed Dec 13, 2025). 40 Cohen J. A coefficient of agreement for nominal scales. Educ Psychol Meas 1960; 20: 37–46. 41 Wang L, Yang N,...

  10. [28]

    33 45 Papineni K, Roukos S, Ward T, Zhu W-J

    http://arxiv.org/abs/2503.09516. 33 45 Papineni K, Roukos S, Ward T, Zhu W-J. BLEU: a method for automatic evaluation of machine translation. In: Isabelle P, Charniak E, Lin D, eds. Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL ’02. ...

  11. [31]

    19 https://github.com/arcee-ai/mergekit.", (Accessed Jan 27, 2026)

    http://arxiv.org/abs/2407.21783. 19 https://github.com/arcee-ai/mergekit.", (Accessed Jan 27, 2026). 20 Akiba T, Shing M, Tang Y, Sun Q, Ha D. Evolutionary optimization of model merging recipes. Nat Mach Intell 2025; 7: 195–204. 21 Shao Z, Wang P, Zhu Q, et al. DeepSeekMath: P...

  12. [79]

    hallucinations

    sets using stratified sampling. 20 Construct semantic vectors to index NCCN guidelines to help reasoning To improve LLMs’ adherence to clinical guidelines, we integrated the NCCN Guidelines in Oncology for colon cancer (Version 4.2025) and rectal cancer (Version 3.2025) into t...

  13. [141]

    Enhancing EHR-based pancreatic cancer prediction with LLM-derived embeddings

    11 Park J, Patterson J, Acitores Cortina JM, Gu T, Hur C, Tatonetti N. Enhancing EHR-based pancreatic cancer prediction with LLM-derived embeddings. NPJ Digit Med 2025; 8:

  14. [274]

    ESMO guidance on the use of Large Language Models in Clinical Practice (ELCAP)

    28 Wong EYT, Verlingue L, Aldea M, et al. ESMO guidance on the use of Large Language Models in Clinical Practice (ELCAP). Ann Oncol 2025; 36: 1447–57. 29 Kirkpatrick J, Pascanu R, Rabinowitz N, et al. Overcoming catastrophic forgetting in neural networks. arXiv [cs.LG]. 2016; ...

  15. [329]

    Generative artificial intelligence to transform inpatient discharge summaries to patient-friendly language and format

    25 Zaretsky J, Kim JM, Baskharoun S, et al. Generative artificial intelligence to transform inpatient discharge summaries to patient-friendly language and format. JAMA Netw Open 2024; 7: e240357. 26 Hu Y, Zuo X, Zhou Y, et al. Information extraction from clinical notes: are we...

  16. [465]

    Large language model influence on diagnostic reasoning: A randomized clinical trial

    12 Goh E, Gallo R, Hom J, et al. Large language model influence on diagnostic reasoning: A randomized clinical trial. JAMA Netw Open 2024; 7: e2440969. 13 McDuff D, Schaekermann M, Tu T, et al. Towards accurate differential diagnosis with large language models. Nature 2025; pu...

  17. [561]

    Review of precision cancer medicine: Evolution of the treatment paradigm

    2 Tsimberidou AM, Fountzilas E, Nikanjam M, Kurzrock R. Review of precision cancer medicine: Evolution of the treatment paradigm. Cancer Treat Rev 2020; 86: 102019. 3 Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. An overview of clinical decision suppo...

  18. [600]

    Evaluating large language models and agents in healthcare: key challenges in clinical applications

    34 Chen X, Xiang J, Lu S, Liu Y, He M, Shi D. Evaluating large language models and agents in healthcare: key challenges in clinical applications. Intell Med 2025; 5: 151–63. 35 Penedo G, Kydlíček H, Allal LB, et al. The FineWeb datasets: Decanting the web for the finest text d...

  19. [1032]

    A large language model for electronic health records

    9 Yang X, Chen A, PourNejatian N, et al. A large language model for electronic health records. npj Digital Medicine 2022; 5: 1–9. 10 Xie Q, Chen Q, Chen A, et al. Medical foundation large language models for comprehensive text analysis and beyond. NPJ Digit Med 2025; 8:

  20. [1932]

    The quality and safety of using generative AI to produce patient-centred discharge instructions

    24 Stanceski K, Zhong S, Zhang X, et al. The quality and safety of using generative AI to produce patient-centred discharge instructions. NPJ Digit Med 2024; 7:

  21. [2011]

    DOI:10.1145/1978942.1978963

  22. [2020]

    50 Zha Y, Yang Y, Li R, Hu Z

    DOI:10.18653/v1/2020.acl-main.704. 50 Zha Y, Yang Y, Li R, Hu Z. AlignScore: Evaluating factual consistency with A unified alignment function. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Stroudsburg, PA, ...

  23. [2023]

    51 Li Y, Wang H, Zhang Q, et al

    DOI:10.18653/v1/2023.acl-long.634. 51 Li Y, Wang H, Zhang Q, et al. UniEval: Unified holistic evaluation for unified multimodal understanding and generation. arXiv [cs.CV]. 2025; published online May

  24. [9799]

    The Llama 3 herd of models

    18 Grattafiori A, Dubey A, Jauhri A, et al. The Llama 3 herd of models. arXiv [cs.AI]. 2024; published online July

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.