Agreement Between Large Language Model-Generated Treatment Recommendations With Guideline-Based and Tumor Board Decisions in Gastrointestinal Cancer (KITuKo)

Concordance of Large Language Model-Generated Treatment Recommendations With Multidisciplinary Tumor Board and Guideline-Based Decisions in Gastrointestinal Cancer: A Retrospective Cohort Study

The goal of this observational study is to learn whether a computer program can suggest cancer treatments that match expert recommendations for people with gastrointestinal cancer (cancer of the pancreas, stomach, or colon and rectum).

The main questions it aims to answer are:

• Do the treatment suggestions from the computer program match current medical guidelines?
• Do these suggestions match decisions made by a multidisciplinary tumor board (a team of cancer specialists)?

Researchers will review existing medical records from people who have already been treated for these cancers. They will enter key clinical information into a computer program that uses artificial intelligence (AI). The program will generate treatment suggestions for each case.

Researchers will then compare these suggestions with:

• guideline-based treatment recommendations
• decisions made by the tumor board

This study will help researchers understand whether AI tools could support doctors in making cancer treatment decisions in the future.

Study Overview

• Status: Completed
• Conditions: Colorectal Cancer; Pancreatic Cancer; Gastric Cancer (GC)
• Intervention / Treatment: Other: Treatment recommendation according to official German cancer guideline; Other: Treatment recommendation of a LLM; Other: Treatment recommendation of a multidisciplinary tumor board

Detailed Description

Gastrointestinal cancers require complex treatment planning that often involves surgery, systemic therapy, and multidisciplinary coordination. Clinical decision-making is typically guided by evidence-based recommendations and discussed in multidisciplinary tumor boards. However, the increasing complexity of treatment strategies and guideline frameworks can make consistent and reproducible decision-making challenging in routine clinical practice.

Recent advances in artificial intelligence have enabled the development of large language models (LLMs) that can process structured clinical information and generate text-based recommendations. These systems may offer a scalable approach to support clinical workflows, but their ability to produce reliable and clinically appropriate treatment suggestions in oncology remains uncertain.

This study evaluates the performance of an LLM-based system in the context of gastrointestinal oncology using retrospectively collected clinical case data. Structured case summaries derived from routine clinical documentation are used as standardized input. The model generates treatment recommendations under controlled conditions, allowing systematic comparison with established clinical reference standards.

The analysis focuses on the level of agreement between model-generated recommendations and established decision-making frameworks. In addition, the study explores how model performance varies across different clinical scenarios, including varying levels of disease complexity. Particular attention is given to situations in which recommendations differ, in order to better understand potential limitations of the model and identify patterns that may be clinically relevant.

Furthermore, the study examines the consistency of model outputs when the same clinical information is processed multiple times. This provides insight into the stability and reproducibility of the system, which are important considerations for potential real-world use.

The findings of this study are intended to inform the potential role of LLM-based tools as supportive systems in clinical decision-making. The study does not evaluate clinical outcomes or patient benefit, but instead focuses on agreement with established standards and expert-driven decisions as an initial step in assessing feasibility and safety.

• Study Type: Observational
• Enrollment (Actual): 30

Contacts and Locations

Study Locations

• Germany
  • Brandenburg
    • Brandenburg an der Havel, Brandenburg, Germany, 14770
      • University Hospital Brandenburg

Participation Criteria

Eligibility Criteria

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

• Sampling Method: Non-Probability Sample

Study Population

The study population consists of adult patients with gastrointestinal adenocarcinoma treated at a tertiary care academic center in the Federal State of Brandenburg, Germany. The population is derived from routine clinical practice and includes patients whose cases were evaluated in a multidisciplinary tumor board.

Description

Inclusion Criteria:

• Histologically confirmed pancreatic, gastric, or colorectal adenocarcinoma
• Treatment discussed in a multidisciplinary tumor board

Exclusion Criteria:

• Non-adenocarcinoma histology

Study Plan

How is the study designed?

Number of groups / cohorts

3

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Colorectal cancer; Patients with colorectal cancer	Other: Treatment recommendation according to official German cancer guideline; Detailed treatment recommendation according to the official guideline of the Association of the Scientific Medical Societies in Germany (AWMF; Arbeitsgemeinschaft der Wissenschaftlichen Medizinischen Fachgesellschaften),; Other: Treatment recommendation of a LLM; Structured clinical case summaries were analyzed by a GPT-4-class large language model to generate treatment recommendations.; Other: Treatment recommendation of a multidisciplinary tumor board; Detailed treatment recommendation according to the case-specific postoperative tumor board review.
Pancreatic cancer; Patients with pancreatic cancer	Other: Treatment recommendation according to official German cancer guideline; Detailed treatment recommendation according to the official guideline of the Association of the Scientific Medical Societies in Germany (AWMF; Arbeitsgemeinschaft der Wissenschaftlichen Medizinischen Fachgesellschaften),; Other: Treatment recommendation of a LLM; Structured clinical case summaries were analyzed by a GPT-4-class large language model to generate treatment recommendations.; Other: Treatment recommendation of a multidisciplinary tumor board; Detailed treatment recommendation according to the case-specific postoperative tumor board review.
Gastric cancer; Patients with gastric cancer	Other: Treatment recommendation according to official German cancer guideline; Detailed treatment recommendation according to the official guideline of the Association of the Scientific Medical Societies in Germany (AWMF; Arbeitsgemeinschaft der Wissenschaftlichen Medizinischen Fachgesellschaften),; Other: Treatment recommendation of a LLM; Structured clinical case summaries were analyzed by a GPT-4-class large language model to generate treatment recommendations.; Other: Treatment recommendation of a multidisciplinary tumor board; Detailed treatment recommendation according to the case-specific postoperative tumor board review.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Concordance with guideline-based management	Agreement between LLM-generated recommendations and AWMF guideline-supported treatment strategies	At the time of multidisciplinary tumor board evaluation up to 4 weeks after surgery

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Concordance with multidisciplinary tumor board decisions	Agreement between LLM-generated recommendations and tumor board treatment strategies	At the time of multidisciplinary tumor board evaluation up to 4 weeks after surgery
Reproducibility of LLM recommendations across repeated runs	Structured clinical case vignettes were entered into ChatGPT using a standardized prompt template. To assess within-model reproducibility, each clinical vignette was analyzed in 3 independent model sessions performed on different days using identical clinical input.	At the time of multidisciplinary tumor board evaluation up to 4 weeks after surgery
Characterization of discordant recommendations (e.g., overtreatment, undertreatment)	Overtreatment was defined as an LLM-generated recommendation exceeding the intensity of the reference recommendation. Undertreatment was defined as omission of a recommended treatment or recommendation of a less intensive strategy.	At the time of multidisciplinary tumor board evaluation up to 4 weeks after surgery

Collaborators and Investigators

• Sponsor: Medizinische Hochschule Brandenburg Theodor Fontane

Study record dates

Study Major Dates

• Study Start (Actual): January 1, 2025
• Primary Completion (Actual): January 1, 2026
• Study Completion (Actual): February 25, 2026

Study Registration Dates

• First Submitted: May 4, 2026
• First Submitted That Met QC Criteria: May 14, 2026
• First Posted (Actual): May 18, 2026

Study Record Updates

• Last Update Posted (Actual): May 18, 2026
• Last Update Submitted That Met QC Criteria: May 14, 2026
• Last Verified: May 1, 2026

More Information

Terms related to this study

• Keywords: Artificial Intelligence; Colorectal Neoplasms; Pancreatic Neoplasms; Stomach Neoplasms; Clinical Decision Support Systems
• Additional Relevant MeSH Terms: Endocrine System Diseases; Neoplasms by Site; Neoplasms; Intestinal Diseases; Gastrointestinal Neoplasms; Digestive System Neoplasms; Digestive System Diseases; Gastrointestinal Diseases; Stomach Diseases; Intestinal Neoplasms; Rectal Diseases; Endocrine Gland Neoplasms; Pancreatic Diseases; Colonic Diseases; Stomach Neoplasms; Colorectal Neoplasms; Pancreatic Neoplasms
• Other Study ID Numbers: KITuKo

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Individual participant data will not be shared. The dataset consists of retrospective, pseudonymized clinical data from a single institution, and sharing is restricted due to data protection regulations and institutional policies.

Drug and device information, study documents

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