Predicting Response to Neuromodulators or Prokinetics in Patients With Suspected Gastroparesis Using Machine...

Area of Science:

• Gastroenterology
• Medical Informatics
• Pharmacology

Background:

• Pharmacologic therapies for gastroparesis (GP) exhibit limited efficacy.
• Predicting patient response to current GP treatments is challenging.
• Novel approaches are needed to personalize GP management.

Purpose of the Study:

• To implement and evaluate a machine learning model for predicting treatment response in patients with GP-like symptoms.
• To identify key predictors of response to prokinetics and/or neuromodulators.

Main Methods:

• A cohort of 150 patients with suspected GP underwent simultaneous gastric emptying scintigraphy (GES) and wireless motility capsule.
• Machine learning models (lasso, ridge, random forest) were trained using five-fold cross-validation.
• Treatment response was defined as a ≥1 decrease in the GP Cardinal Symptom Index at 6 months.

Main Results:

• A ridge regression model achieved an AUC-ROC of 0.72, identifying predictors like body mass index, infectious prodrome, delayed GES, and absence of diabetes.
• The model showed higher accuracy for patients on prokinetics alone (AUC-ROC 0.83) compared to neuromodulators alone.
• A separate model incorporating gastric emptying time and duodenal motility index yielded an AUC-ROC of 0.75.

Conclusions:

• Machine learning models demonstrate acceptable accuracy in predicting response to prokinetics and/or neuromodulators for GP.
• Validated models can provide valuable insights for predicting treatment outcomes in GP patients.
• This approach may facilitate personalized treatment strategies for gastroparesis.