LLM-assisted evidence audit of late-stage cancer incidence as a screening trial endpoint

Abstract

BackgroundLate-stage cancer incidence is being considered as an earlier endpoint in cancer-screening trials, but its trial-level association with cancer-specific mortality may depend on evidence selection and endpoint harmonization. We evaluated the robustness of this association to source-verified additions.

MethodsWe reconstructed the PubMed corpus underlying a 41-comparison review. Gemini 3.1 Pro Preview was used only to prioritize reports for blinded human reassessment. Reviewers determined eligibility, linked reports from the same trial, harmonized endpoints, and verified comparison-level data. We recalculated unweighted Pearson correlations overall and by cancer type after adding earliest-compatible trial comparisons.

ResultsAmong 1209 candidate records, 996 PDFs were assessed. Thirty-three reports absent from the source review were prioritized; 26 were eligible, representing 18 trials, and 8 provided compatible comparisons. Adding these comparisons increased the dataset from 41 to 49 and attenuated the overall correlation from 0.73 (95% confidence interval [CI] = 0.55 to 0.85) to 0.59 (95% CI = 0.37 to 0.75). Updated correlations were 0.49 (95% CI = -0.26 to 0.87) for breast, - 0.23 (95% CI = -0.71 to 0.40) for colorectal, and 0.83 (95% CI = 0.54 to 0.95) for lung cancer. One sparse-event comparison influenced the colorectal estimate.

ConclusionsThe overall association was sensitive to evidence composition, and cancer-specific stability varied. Late-stage incidence should be evaluated by cancer type and with prespecified sensitivity analyses for evidence selection and endpoint definitions. Model-assisted prioritization cannot replace human eligibility review, trial reconciliation, and source verification.

A retrospective epidemiological study of screening-derived solid cancer incidence proxy rates in South Korea before and after COVID-19 using Korean National Health Insurance Service data
Hyeran Jung, Minsun Jung
PLOS One16 Oct 2026Peer-reviewedCC BY 4.0Full textOncologyPLOS
AI-assisted nurse-led skin cancer screening in a teledermoscopy framework: a multi-site evaluation with one million lesions
N. E. K. Shaik, N. Desai, K. Wang +5
medRxiv1 Oct 2026PreprintCC BY-NC-NDFull textOncologymedRxiv
Rapid Psychosocial Screening Instruments for Cancer Patients: A Scoping Review of Measurement Tools and Psychometric Evidence
Ida Maryati, Raini Diah Susanti, Taty Hernawaty +5
Asian Pacific Journal of Cancer Prevention : APJCP22 Jun 2026Peer-reviewedCC BY-NC 4.0Full text1 citedOncologyPMC
Prospective Clinical Evidence on Artificial Intelligence-Enabled Augmented and Mixed Reality in Robot-Assisted Prostate and Kidney Cancer Surgery: A Systematic Review.
Gregory Harrison
Cureus23 Sep 2026Peer-reviewedCC BYFull textGeneral medicineEurope PMC
Silent burden of gynecological morbidities among rural women in Southern India: Evidence from a cancer screening center
Zari Anjum, Vibha Swaroop, Neha Singh +1
Journal of family medicine and primary care1 Jul 2026Peer-reviewedCC BY-NC-NDFull textEpidemiology & public healthEurope PMCPMC

Cite

BibTeX RIS

BibTeX

@misc{li2026llmassisted,
  title = {{LLM-assisted evidence audit of late-stage cancer incidence as a screening trial endpoint}},
  author = {S. Li and W. Zhang and X. Xing and Z. Shen and Y. Wang and Z. Chen and O. Neto and Y. Yu and C. Wu and L. Lin},
  howpublished = {medRxiv},
  year = {2026},
  month = aug,
  doi = {10.64898/2026.08.29.26361733},
  url = {https://www.medrxiv.org/content/10.64898/2026.08.29.26361733v1},
  note = {CC BY-NC-ND},
}

RIS

TY  - UNPB
TI  - LLM-assisted evidence audit of late-stage cancer incidence as a screening trial endpoint
AU  - S. Li
AU  - W. Zhang
AU  - X. Xing
AU  - Z. Shen
AU  - Y. Wang
AU  - Z. Chen
AU  - O. Neto
AU  - Y. Yu
AU  - C. Wu
AU  - L. Lin
PY  - 2026
DA  - 2026/08/29
PB  - medRxiv
DO  - 10.64898/2026.08.29.26361733
UR  - https://www.medrxiv.org/content/10.64898/2026.08.29.26361733v1
AB  - BackgroundLate-stage cancer incidence is being considered as an earlier endpoint in cancer-screening trials, but its trial-level association with cancer-specific mortality may depend on evidence selection and endpoint harmonization. We evaluated the robustness of this association to source-verified additions.  MethodsWe reconstructed the PubMed corpus underlying a 41-comparison review. Gemini 3.1 Pro Preview was used only to prioritize reports for blinded human reassessment. Reviewers determined eligibility, linked reports from the same trial, harmonized endpoints, and verified comparison-level data. We recalculated unweighted Pearson correlations overall and by cancer type after adding earliest-compatible trial comparisons.  ResultsAmong 1209 candidate records, 996 PDFs were assessed. Thirty-three reports absent from the source review were prioritized; 26 were eligible, representing 18 trials, and 8 provided compatible comparisons. Adding these comparisons increased the dataset from 41 to 49 and attenuated the overall correlation from 0.73 (95% confidence interval [CI] = 0.55 to 0.85) to 0.59 (95% CI = 0.37 to 0.75). Updated correlations were 0.49 (95% CI = -0.26 to 0.87) for breast, - 0.23 (95% CI = -0.71 to 0.40) for colorectal, and 0.83 (95% CI = 0.54 to 0.95) for lung cancer. One sparse-event comparison influenced the colorectal estimate.  ConclusionsThe overall association was sensitive to evidence composition, and cancer-specific stability varied. Late-stage incidence should be evaluated by cancer type and with prespecified sensitivity analyses for evidence selection and endpoint definitions. Model-assisted prioritization cannot replace human eligibility review, trial reconciliation, and source verification.
ER  -

Abstract

Related

Cite