Development and validation of an interpretable machine learning model for predicting chronic atrophic gastritis in...

Area of Science:

• Medical Informatics
• Oncology
• Gastroenterology

Background:

• Chronic atrophic gastritis (CAG) is a significant precancerous condition.
• Endoscopic screening for CAG in the elderly is limited by invasiveness, cost, and accessibility.
• Developing non-invasive predictive tools is crucial for early detection in geriatric populations.

Purpose of the Study:

• To develop and validate interpretable machine learning (ML) models for predicting CAG in elderly patients.
• To identify key clinical factors associated with CAG risk in this demographic.
• To create an accessible tool for stratifying CAG risk in older adults.

Main Methods:

• Retrospective cohort study with development (n=1,268) and temporal validation (n=544) cohorts.
• Collected 28 candidate variables including demographics, lifestyle, medical history, and symptoms.
• Employed multi-dimensional feature selection and constructed/optimized nine ML models, including Multilayer Perceptron (MLP).

Main Results:

• Eight key predictive variables were identified through rigorous feature selection.
• The MLP model achieved an AUC of 0.826 (internal) and 0.780 (temporal) for CAG prediction.
• Key risk factors identified: Helicobacter pylori infection, age, smoking, high-salt pickled food; protective factor: fruit/vegetable intake.

Conclusions:

• An interpretable ML model was successfully developed and validated for predicting CAG in elderly patients.
• The model utilizes eight readily available clinical variables, offering a non-invasive and accessible risk stratification tool.
• This approach can facilitate early screening and targeted interventions for CAG in primary care settings.