Clinical Report: Creation and assessment of an interpretable machine learning model for forecasting chronic atrophic gastritis in older adults

Overview

This study developed and validated interpretable machine learning models to predict chronic atrophic gastritis (CAG) in elderly patients, identifying key predictive factors. The Multilayer Perceptron model showed optimal performance with an AUC of 0.826 in internal validation and 0.780 in external validation.

Background

Chronic atrophic gastritis is a significant precancerous condition, particularly prevalent in the elderly, where early detection is crucial to prevent progression to gastric cancer. Current screening methods, primarily endoscopy, are limited by invasiveness and accessibility, necessitating alternative non-invasive approaches.

Data Highlights

Key Findings

• Eight key variables were identified as predictors of CAG through multi-dimensional feature selection.
• The Multilayer Perceptron model demonstrated the highest performance metrics among the nine models evaluated.
• Helicobacter pylori infection, age, smoking status, and high-salt pickled food intake were identified as major risk factors.
• Increased intake of fruits and vegetables was found to be a protective factor against CAG.
• The study utilized SHAP analysis to interpret the contributions of each variable in the predictive model.

Clinical Implications

The developed machine learning model provides a non-invasive tool for predicting CAG.

Conclusion

This study presents a validated machine learning model for predicting chronic atrophic gastritis in older adults.

Related Resources & Content

1. ACG Clinical Guideline: Diagnosis and Management of Gastric Premalignant Conditions - PubMed
2. The potential value of serum pepsinogen and gastrin-17 for the diagnosis of chronic atrophic gastritis at different stages of severity: a clinical diagnostic study | BMC Gastroenterology | Springer Nature Link
3. Risk prediction for chronic atrophic gastritis using a random forest model: A multicenter study - PMC
4. aace endocrine ai — AI model identifies CKD risk in older adults
5. Frontiers in Medicine — Machine learning-based prediction of gout flares during hospitalization in patients with upper gastrointestinal bleeding: a retrospective cohort study
6. Frontiers in Endocrinology — A Transparent Machine Learning Approach Utilizing Standard Metabolic Lab Indices for Detecting Advanced Chronic Kidney Disease
7. BMC Psychiatry (Springer) — A Transparent Machine Learning Approach for Forecasting Depressive Symptoms in Elderly Chinese Individuals with Chronic Illnesses
8. AI model identifies CKD risk in older adults
9. Machine learning-based prediction of gout flares during hospitalization in patients with upper gastrointestinal bleeding: a retrospective cohort study
10. A Transparent Machine Learning Approach Utilizing Standard Metabolic Lab Indices for Detecting Advanced Chronic Kidney Disease
11. ACG Clinical Guideline: Diagnosis and Management of Gastric Premalignant Conditions - PubMed
12. 25053 504..554
13. The potential value of serum pepsinogen and gastrin-17 for the diagnosis of chronic atrophic gastritis at different stages of severity: a clinical diagnostic study | BMC Gastroenterology | Springer Nature Link
14. First-of-kind guideline outlines surveillance of premalignant gastric conditions | ACP Gastroenterology Monthly
15. Serum pepsinogen Ⅰ and the pepsinogen Ⅰ/Ⅱ ratio as a non-invasive predictive marker for autoimmune metaplastic atrophic gastritis - ScienceDirect
16. Risk prediction for chronic atrophic gastritis using a random forest model: A multicenter study - PMC
17. Frontiers | Efficacy and validation of a clinical predictive model for chronic atrophic gastritis in patients: a multi-center retrospective analysis
18. Constructing and validating vision transformer-based assisted detection models for atrophic gastritis: A retrospective study - PubMed