KM-GPT: An Automated Pipeline for Reconstructing Individual Patient Data from Kaplan-Meier Plots

Area of Science:

• Biomedical Informatics
• Clinical Research Methodology
• Artificial Intelligence in Healthcare

Background:

• Reconstructing individual patient data (IPD) from Kaplan-Meier (KM) plots is crucial for clinical research evidence synthesis.
• Current methods for IPD reconstruction from KM plots are manual, time-consuming, error-prone, and lack scalability.
• There is a need for automated, accurate, and reproducible methods to extract IPD from KM plots.

Purpose of the Study:

• To introduce KM-GPT, the first fully automated AI-powered pipeline for reconstructing IPD from KM plots.
• To address the limitations of manual digitization in terms of accuracy, scalability, and reproducibility.
• To provide a user-friendly, web-based solution for researchers to access and utilize reconstructed IPD.

Main Methods:

• Developed KM-GPT, an AI pipeline integrating image preprocessing, GPT-5 multi-modal reasoning, and iterative reconstruction algorithms.
• Employed a hybrid reasoning architecture to automate the conversion of unstructured KM plot information into structured data.
• Validated data extraction from complex KM plots through automated processes without manual intervention.

Main Results:

• KM-GPT demonstrated high accuracy, robustness, and reproducibility in reconstructing IPD from both synthetic and real-world datasets.
• The AI pipeline successfully automated the extraction of structured data from complex Kaplan-Meier plots.
• Application to gastric cancer immunotherapy trials facilitated evidence synthesis and biomarker-based subgroup analyses.

Conclusions:

• KM-GPT offers a transformative, scalable, and accurate solution for reconstructing IPD from KM plots, automating a previously manual process.
• The tool enhances clinical research by enabling more informed downstream analyses and supporting evidence-based decision-making.
• KM-GPT's user-friendly interface and AI assistant make advanced IPD reconstruction accessible to researchers without programming expertise.