Related Experiment Videos
Interpretable Machine Learning Enables Preoperative Physiologic Risk Stratification for Dysphagia After Anti-Reflux
Anjani H Turaga1, Yashwanth Alakky2, Paneed Jalili1
1Weill Cornell Medicine.
Research Square
|July 29, 2026
Summary
Predicting postoperative dysphagia after anti-reflux surgery is improved by a new machine learning tool. This score integrates multiple esophageal function tests to identify high-risk patients for better surgical planning.
Area of Science:
- Gastroenterology and Hepatology
- Surgical Innovation
- Medical Informatics
Background:
- Postoperative dysphagia is a significant complication after anti-reflux surgery.
- Current risk assessment tools lack comprehensive integration of esophageal physiology.
- Esophageal motility, reflux burden, and esophagogastric junction biomechanics are key factors.
Purpose of the Study:
- To develop an interpretable machine learning framework to predict postoperative dysphagia.
- To integrate multimodal physiologic and clinical data for risk stratification.
- To create a clinically deployable, point-based risk score for patient assessment.
Main Methods:
- Screening of 878 patients, with 428 included for analysis.
- Multimodal preoperative assessment: high-resolution manometry, EndoFLIP, Bravo pH monitoring, GERD-HRQL.
- Ensemble machine learning (logistic regression, random forest, XGBoost, SVM) with interaction features.
Main Results:
- The ensemble model showed strong predictive performance (AUC=0.91 in derivation, AUC=0.72 in validation).
- Integrated features (reflux, contractility, distensibility) outperformed isolated variables.
- The developed risk score effectively stratified patients into distinct dysphagia susceptibility groups.
Conclusions:
- Integrated multimodal esophageal physiology and machine learning accurately predict postoperative dysphagia.
- The interpretable risk score facilitates personalized perioperative risk stratification.
- An open-source web calculator is available for individualized risk estimation.