Optimizing the diagnostic power with gastric emptying scintigraphy at multiple time points

Area of Science:

• Gastroenterology
• Medical Imaging
• Biostatistics

Background:

• Gastric Emptying Scintigraphy (GES) is the gold standard for diagnosing gastroparesis, utilizing a standardized radio-labeled meal over 4 hours.
• Accurate diagnosis of gastroparesis is crucial for effective patient management and treatment.

Purpose of the Study:

• To determine the optimal time points and combinations for diagnosing gastroparesis using repeated GES measures.
• To compare the diagnostic performance of Fisher's Linear Discriminant Analysis (LDA), a rank-based Distribution Free (DF) approach, and the Classification And Regression Tree (CART) model.

Main Methods:

• Retrospective analysis of GES data from 320 patients at 1, 2, 3, and 4 hours post-meal.
• Model comparison using Area Under the Receiver Operating Characteristic curve (AUC) and jackknife cross-validation for false classification rates.

Main Results:

• The CART model, using 4-hourly GES data and patient age, achieved the highest diagnostic accuracy (AUC=0.88, false classification=13.8%).
• The DF method with a 4-h and 3-h combination improved AUC (0.87) and reduced false classifications (15.0%) compared to individual time points.
• A 4-h gastric retention >10% indicated a 5-fold increased likelihood of gastroparesis.

Conclusions:

• The CART model demonstrates superior robustness and accuracy for gastroparesis diagnosis compared to LDA and DF methods.
• CART's ability to incorporate covariates like age enhances its generalizability across institutions.
• While powerful, CART's stability may be limited with smaller sample sizes.