Related Experiment Video
Updated: Sep 17, 2025
11:02
Induction of Invasive Transitional Cell Bladder Carcinoma in Immune Intact Human MUC1 Transgenic Mice: A Model for Immunotherapy Development
Published on: October 30, 2013
21.4K
Deep Learning Model for Natural Language to Assess Effectiveness of Patients With Non-Muscle Invasive Bladder Cancer
Makito Miyake1, Naohiro Yonemoto2, Kanae Togo3
1Department of Urology, Nara Medical University, Kashihara, Nara, Japan.
JCO Clinical Cancer Informatics
|June 27, 2025
Summary
A deep learning model accurately assesses non-muscle invasive bladder cancer outcomes from electronic health records. This approach effectively compares Bacillus Calmette-Guérin therapy completion groups, demonstrating improved time to recurrence.
Area of Science:
- Oncology
- Artificial Intelligence
- Medical Informatics
Background:
- Non-muscle invasive bladder cancer (NMIBC) treatment outcome assessment is complex.
- Electronic health records (EHRs) contain vast patient data.
- Natural Language Processing (NLP) can extract clinical information from unstructured EHR text.
Purpose of the Study:
- Develop a deep learning NLP model to assess NMIBC clinical outcomes.
- Utilize EHR data for efficient and accurate outcome analysis.
- Compare treatment effectiveness based on therapy completion.
Main Methods:
- Retrospective analysis of Japanese adult NMIBC patients receiving Bacillus Calmette-Guérin (BCG) therapy.
- Trained a Bidirectional Encoder Representations from Transformers (BERT) model for outcome classification.
- Assessed model performance using precision, recall, and F1 scores; compared BCG completion vs. non-completion groups.
Main Results:
- The BERT model achieved high F1 scores for time to recurrence (TTR) and time to progression (TTP).
- Human review was needed for only 10% of documents, indicating feasibility.
- BCG completion was associated with a significantly lower hazard ratio for TTR compared to non-completion.
Conclusions:
- A deep learning NLP model can effectively assess NMIBC clinical outcomes using EHR data.
- The model demonstrated feasibility and accuracy, with minimal human support required.
- The study confirmed differences in TTR and TTP between BCG completion and non-completion groups.
