Preliminary Evaluation of a Large Language Model-Based Tool for Complex Surgical Decision Support in Lung Cancer

This study is an exploratory effect-size estimation study, with the following specific objectives: ① to estimate the point estimate and 95% confidence interval of the Win Ratio for the experimental group (GAPS-Agent) versus the control group (large language model) in blinded pairwise preference judgments by thoracic surgery expert adjudicators, to serve as a sample size planning parameter for subsequent multicenter confirmatory clinical trials; ② to preliminarily evaluate the value of GAPS-Agent within clinical workflows.The hypothesis of this study is as follows: compared with a general-purpose large language model without medical enhancement (control group), a structured agentic workflow optimized on the basis of the GAPS evaluation framework (GAPS-Agent, experimental group) can help junior resident physicians generate clinical decision plans for complex lung cancer cases that are more strongly preferred by senior thoracic surgery expert adjudicators.

Study Overview

• Status: Enrolling by invitation
• Conditions: Lung Cancer (NSCLC); Large Language Models
• Intervention / Treatment: Other: GAPS-Agent; Other: LLM

• Study Type: Interventional
• Enrollment (Estimated): 12
• Phase: Not Applicable

Contacts and Locations

Study Locations

• China
  • Beijing Municipality
    • Beijing, Beijing Municipality, China, 100044
      • Peking University People's Hospital

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

1. Resident Physician Subjects:

   1. Holds a valid and legally effective Physician Practice License of the People's Republic of China;
   2. Currently holds the rank of resident physician in a thoracic surgery department at a tertiary Class A (3A) hospital;
   3. Agrees to complete all assessment tasks of the main study phase in accordance with the study protocol;
   4. Can guarantee the time and effort required to complete all assessment tasks of the main study.

2. Study Cases:

   1. The case was discussed at the Thoracic Oncology Multidisciplinary Team (MDT) conference of Peking University People's Hospital between January 2025 and May 2026;
   2. The current version of the NCCN guidelines does not provide an explicit recommendation covering the management of the case;
   3. Does not overlap with the GAPS evaluation set;
   4. The case is presented in pure text in a structured format, with all direct and indirect identifiers removed and complete de-identification performed prior to inclusion;
   5. From the pool of eligible cases, 12 cases will be randomly drawn using Python (numpy.random, with a fixed and archived seed) to serve as the main study cases. The cases will cover 6 themes (chest mass of undetermined diagnosis, early-stage lung cancer, locally advanced lung cancer, oligometastatic/oligoprogressive disease, special intraoperative situations, and tumor recurrence), with 2 cases per theme.

3. Adjudication Expert Panel:

   1. Holds a valid and legally effective Physician Practice License of the People's Republic of China;
   2. Currently holds the rank of attending physician or above in a thoracic surgery department at a tertiary Class A hospital;
   3. Chairs or regularly participates in lung cancer multidisciplinary team (MDT) work in their department.

Exclusion Criteria:

1. Resident Physician Subjects:

   1. Has previously participated in the construction of the GAPS evaluation set or the development of GAPS-Agent;
   2. Unable to complete the tasks of the study phase.

2. Study Cases:

   1. Key case information is missing, such as text-form data on pathology (including IHC/NGS), imaging, laboratory tests, prior medical history, comorbidities, or PS score;
   2. Decision-making for the case is strictly dependent on non-text information.

3. Adjudication Expert Panel:

   1. Participated in the construction of the GAPS evaluation set, the content validity verification, or the development of GAPS-Agent for this study;
   2. Has a direct conflict of interest with any specific product among the two-arm tools of this study.

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Other
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: test arm; GAPS-Agent	Other: GAPS-Agent; The research group has previously developed the GAPS evaluation framework for complex clinical decision-making in lung cancer. In this framework, G (Grounding) characterizes the cognitive depth of decision-making (ranging from knowledge retrieval to decisions that go beyond clinical guidelines), A (Authority) corresponds to the grading of evidence strength, P (Perturbation) describes the identification and management of real-world clinical confounding factors, and S (Strength) corresponds to the calibration of recommendation strength. Within this framework, the research group has completed the construction of a 100-item complex lung cancer decision-making evaluation set along with its corresponding rubrics, and has invited multiple thoracic oncology experts to complete content validity validation. Based on this, the research group developed GAPS-Agent, which uses an open-source large language model as its foundation and integrates functional modules such as guideline and evidence retri
Active Comparator: control arm; LLM	Other: LLM; Open source large language model that is not specifically enhanced in medical field.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Overall plan Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Inter-rater agreement	For the ternary preference judgment results of 10 expert judges across 192 paired comparisons and 6 evaluation domains, Fleiss' kappa was used to assess inter-rater agreement. The kappa value and its 95% confidence interval are reported for each evaluation domain.	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
Redundancy Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
Evidence-based medicine adherence Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
Actionability Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
Completeness Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
Safety Win Ratio	A total of 10 blinded expert judges made Win/Tie/Loss ternary preference judgments on 192 paired scheme comparisons in terms of overall scheme quality. The win ratio was calculated as Wins ÷ Losses, and the 95% confidence interval was estimated using a two-level (physician × case) cluster bootstrap resampling method (B = 10,000, quantile method on the log scale).	Measured at the time when experts completed their preference judgements. Calculated up to 3 weeks after the preference judgements.
GAPS automated rubric score	A third-party large language model, independent of the two study arms' base models, served as the judge model and automatically scored all 96 plans according to the GAPS rubric.	Generated up to 3 weeks after residents finished their plan generation.
Subject physician's self-confidence score	After submitting each case plan, the participating physicians self-rated their confidence in their own plan using a 1-5 point Likert scale.	Completed at the time when residents submitted their plans. Calculated up to 3 weeks after the submission.
Tool satisfaction score	After submitting each case plan, the participating physicians rated their satisfaction with the tool using a 1-5 point Likert scale.	Completed at the time when residents submitted their plans. Calculated up to 3 weeks after the submission.
Tool trustworthiness score	After submitting each case plan, the participating physicians rated the tool's credibility using a 1-5 point Likert scale.	Completed at the time when residents submitted their plans. Calculated up to 3 weeks after the submission.
Decision-making time	The time taken (in minutes) by each participating physician to complete the production of each case plan was automatically recorded by the evaluation platform. Differences between groups were analyzed using a linear mixed-effects model.	Completed at the time when residents submitted their plans. Calculated up to 3 weeks after the submission.

Collaborators and Investigators

• Sponsor: Peking University People's Hospital

Study record dates

Study Major Dates

• Study Start (Actual): June 10, 2026
• Primary Completion (Estimated): June 21, 2026
• Study Completion (Estimated): June 21, 2026

Study Registration Dates

• First Submitted: June 10, 2026
• First Submitted That Met QC Criteria: June 13, 2026
• First Posted (Actual): June 17, 2026

Study Record Updates

• Last Update Posted (Actual): June 17, 2026
• Last Update Submitted That Met QC Criteria: June 13, 2026
• Last Verified: June 1, 2026

More Information

Terms related to this study

• Keywords: Lung Cancer; Clinical Trial; Multidisciplinary Team; Large Language Models; Benchmark Test
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Respiratory Tract Diseases; Lung Diseases; Respiratory Tract Neoplasms; Thoracic Neoplasms; Carcinoma, Bronchogenic; Bronchial Neoplasms; Lung Neoplasms; Carcinoma, Non-Small-Cell Lung
• Other Study ID Numbers: 2026PHB458-001

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No