Deep Learning Model for Predicting Operative Mortality After Total Gastrectomy: Analysis of the Japanese National...

Area of Science:

• Surgical Oncology
• Medical Informatics
• Machine Learning

Background:

• Radical gastrectomy is the standard treatment for gastric cancer, but carries a significant complication rate (10%-20%) and mortality (2.3%).
• Accurate preoperative risk stratification is crucial for optimizing patient outcomes and surgical planning.
• Existing methods for risk assessment may not fully capture the complexity of operative mortality prediction.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for predicting operative mortality following total gastrectomy for gastric cancer.
• To leverage a large-scale national clinical database for robust model development.
• To enhance preoperative risk assessment for patients undergoing gastric cancer surgery.

Main Methods:

• Utilized data from the National Clinical Database (NCD) for patients aged 18+ undergoing total gastrectomy (Jan 2018-Dec 2019).
• Included 62 preoperative variables (demographics, medical history, lab results, tumor characteristics) to predict operative mortality.
• Developed DL models using Python, TensorFlow, and Keras, with hyperparameter tuning via k-fold cross-validation.

Main Results:

• A four-layer model with 5217 variables was developed using 11,980 training cases and validated on 3,000 cases.
• The model achieved a C-statistic of 0.79 on training data and 0.74 on validation data.
• The overall operative mortality rate in the study cohort was 1.2%.

Conclusions:

• A deep learning model effectively predicts operative mortality after total gastrectomy using big data.
• The model demonstrates potential for improving preoperative risk stratification in gastric cancer surgery.
• Future improvements may involve incorporating additional variables related to postoperative complications or novel analytical factors.