Wang, Pengkai, Zhang, Wei-Wei, Li, Yan, Tang, Min, Hou, Zhitian, Liu, Zeyu, Zhu, Guanghao, Wang, Yuanyi, Gu, Yanggan, Wang, Wenjun, Ni, Minheng, Xie, Congkai, Sang, Zhijie, Wu, Jianmin, Sun, Ying, Yang, Hongxia · Nov 5, 2026
NPCBench: A Clinical Apprenticeship Benchmark for Multimodal Care-Pathway Reasoning in Nasopharyngeal Carcinoma
Markdown · JSON · Original source
The authors introduce NPCBench, a designed to assess whether models can reason across a patient’s course of care for nasopharyngeal carcinoma, rather than answer isolated medical questions. It combines component questions and consultations with a patient cohort whose diagnostic and care tasks are scored along care pathways. Its episode-level evaluation requires answers within a pathway to be jointly correct, addressing a gap the authors identify in tests that score individual answers without showing whether models can integrate evidence or track changes in a patient’s condition. In their evaluation of 30 models, the authors report a “composition gap”: strong performance on individual tasks did not ensure jointly correct answers for the same patient. Among 21 multimodal models, strict CarePath success reached at most 23%. The benchmark also includes paired controls and physician calibration to examine evidence grounding, longitudinal revision, and consultation judging. The authors present NPCBench as a testbed for evaluating evidence-based and temporally coherent reasoning in oncology care; these benchmark results do not, on their own, establish how models would perform in clinical practice.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T02:55:14.904Z. First observed: 2026-11-06T05:00:47.731Z. Retrieved: 2026-11-09T02:54:20.477Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The supplied evidence is the abstract; it does not provide the detailed evaluation methods or results needed to explain what drives the reported composition gap. The reported results concern model evaluation and do not establish how models would perform in clinical practice.
Open question
What factors account for the gap between strong individual-task performance and strict CarePath success in the authors’ evaluation?
Source-stated
The authors introduce NPCBench to assess clinical reasoning across care pathways for nasopharyngeal carcinoma, rather than only isolated questions, images, or static case vignettes.
we introduce NPCBench, a clinical apprenticeship benchmark for nasopharyngeal carcinoma (NPC).
Source-stated
NPCBench combines component and consultation items with a patient cohort containing scored diagnostic and care tasks.
It contains 4,777 component and consultation items and a 100-patient CarePath cohort with 417 scored diagnostic and care tasks.
Source-stated
The episode-level evaluation checks whether diagnostic and care answers are jointly correct within a care pathway.
The episode-level evaluation tests whether diagnostic and care answers are jointly correct within a care pathway.
Source-stated
In the authors’ evaluation, strong performance on individual tasks did not ensure jointly correct diagnostic and care answers for the same patient.
strong performance on individual tasks does not ensure jointly correct diagnostic and care answers for the same patient.
Source-stated
Among the evaluated multimodal models, strict CarePath success reached at most 23%.
Among 21 multimodal models, strict CarePath success reaches at most 23%.
Source-stated
The abstract presents NPCBench as a testbed for evaluating evidence-based and temporally coherent reasoning in oncology care.
thereby offering a testbed for evidence-based and temporally coherent reasoning in oncology care.
Source and contributor context
Wang, Pengkai, Zhang, Wei-Wei, Li, Yan, Tang, Min, Hou, Zhitian, Liu, Zeyu, Zhu, Guanghao, Wang, Yuanyi, Gu, Yanggan, Wang, Wenjun, Ni, Minheng, Xie, Congkai, Sang, Zhijie, Wu, Jianmin, Sun, Ying, Yang, Hongxia · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Multimodal model
A model that works with more than one kind of input or output, such as text, images and audio.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 16f67473-7fd2-44a5-8839-863b8b56492e.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T02:55:14.904Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 1379a91d94bfb6bc8a756b4930f87993bf58451210755f4bc8f55d8ce11d3be3. Source text hash: 4658d7c35364aebf54a6a58c75b2f97e946a02a48a158a52cda88f73de77f420.
Revision and source-check history
Version 1 · 2026-11-09T02:55:14.904Z
NPCBench: A Clinical Apprenticeship Benchmark for Multimodal Care-Pathway Reasoning in Nasopharyngeal Carcinoma
The authors introduce NPCBench, a designed to assess whether models can reason across a patient’s course of care for nasopharyngeal carcinoma, rather than answer isolated medical questions. It combines component questions and consultations with a patient cohort whose diagnostic and care tasks are scored along care pathways. Its episode-level evaluation requires answers within a pathway to be jointly correct, addressing a gap the authors identify in tests that score individual answers without showing whether models can integrate evidence or track changes in a patient’s condition. In their evaluation of 30 models, the authors report a “composition gap”: strong performance on individual tasks did not ensure jointly correct answers for the same patient. Among 21 multimodal models, strict CarePath success reached at most 23%. The benchmark also includes paired controls and physician calibration to examine evidence grounding, longitudinal revision, and consultation judging. The authors present NPCBench as a testbed for evaluating evidence-based and temporally coherent reasoning in oncology care; these benchmark results do not, on their own, establish how models would perform in clinical practice.
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T02:55:14.904Z. First observed: 2026-11-06T05:00:47.731Z. Retrieved: 2026-11-09T02:54:20.477Z.
Limits
The supplied evidence is the abstract; it does not provide the detailed evaluation methods or results needed to explain what drives the reported composition gap. The reported results concern model evaluation and do not establish how models would perform in clinical practice.
Open question
What factors account for the gap between strong individual-task performance and strict CarePath success in the authors’ evaluation?
Source-stated
The authors introduce NPCBench to assess clinical reasoning across care pathways for nasopharyngeal carcinoma, rather than only isolated questions, images, or static case vignettes.
we introduce NPCBench, a clinical apprenticeship benchmark for nasopharyngeal carcinoma (NPC).
Source-stated
NPCBench combines component and consultation items with a patient cohort containing scored diagnostic and care tasks.
It contains 4,777 component and consultation items and a 100-patient CarePath cohort with 417 scored diagnostic and care tasks.
Source-stated
The episode-level evaluation checks whether diagnostic and care answers are jointly correct within a care pathway.
The episode-level evaluation tests whether diagnostic and care answers are jointly correct within a care pathway.
Source-stated
In the authors’ evaluation, strong performance on individual tasks did not ensure jointly correct diagnostic and care answers for the same patient.
strong performance on individual tasks does not ensure jointly correct diagnostic and care answers for the same patient.
Source-stated
Among the evaluated multimodal models, strict CarePath success reached at most 23%.
Among 21 multimodal models, strict CarePath success reaches at most 23%.
Source-stated
The abstract presents NPCBench as a testbed for evaluating evidence-based and temporally coherent reasoning in oncology care.
thereby offering a testbed for evidence-based and temporally coherent reasoning in oncology care.
Source and contributor context
Wang, Pengkai, Zhang, Wei-Wei, Li, Yan, Tang, Min, Hou, Zhitian, Liu, Zeyu, Zhu, Guanghao, Wang, Yuanyi, Gu, Yanggan, Wang, Wenjun, Ni, Minheng, Xie, Congkai, Sang, Zhijie, Wu, Jianmin, Sun, Ying, Yang, Hongxia · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Multimodal model
A model that works with more than one kind of input or output, such as text, images and audio.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 16f67473-7fd2-44a5-8839-863b8b56492e.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T02:55:14.904Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 1379a91d94bfb6bc8a756b4930f87993bf58451210755f4bc8f55d8ce11d3be3. Source text hash: 4658d7c35364aebf54a6a58c75b2f97e946a02a48a158a52cda88f73de77f420.
uncertain · 2026-11-09T02:55:14.904Z · anthropic/claude-opus-5.5
Publication: 16f67473-7fd2-44a5-8839-863b8b56492e
attribution — pass: The reading consistently attributes the benchmark, the composition-gap finding and the testbed framing to the paper's authors, as presented in the arXiv abstract. It does not present the authors' claims as established fact.
Across 30 evaluated models, we observe a composition gap
summary — uncertain: The summary is faithful, but key terms are not explained in plain language. It never says nasopharyngeal carcinoma is a cancer of the upper throat behind the nose; 'oncology' appears only at the end. It also leaves 'multimodal' (images plus text) and 'CarePath' undefined, so the headline '23%' figure is hard to interpret. Smallest repair: add a brief plain-language explanation of NPC and 'multimodal', and say 'strict CarePath success' means getting every diagnostic and care answer right for one patient. Including the scale (4,777 items, 100 patients, 417 tasks) would also help readers.
Among 21 multimodal models, strict CarePath success reaches at most 23%.
date — pass: The originalPublished date of 2026-11-05 matches the v1 submission date on arXiv. The feed date of Nov 6 is later and is correctly not used.
[Submitted on 5 Nov 2026]
support — pass: Each claim is supported by a verbatim excerpt from the abstract. The caveat that benchmark results do not establish clinical performance is appropriately framed. The question about what drives the gap is a genuine open issue, and the reading's limitations state that the abstract does not answer it. No unsupported benefits or costs are implied.
strong performance on individual tasks does not ensure jointly correct diagnostic and care answers for the same patient.
media — pass: The selected media is the verified arXiv abstract page, which is the direct original source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA model that works with more than one kind of input or output, such as text, images and audio.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA model that works with more than one kind of input or output, such as text, images and audio.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition reference