Using Machine Learning to Improve Early Detection of Celiac Disease (+Video)

Celiac.com 03/10/2025 - Celiac disease is often underdiagnosed or diagnosed only after a long delay, leading to prolonged health complications for affected individuals. Researchers aimed to develop a machine learning tool to identify individuals at risk of celiac disease before they receive an official diagnosis. By analyzing electronic medical records, the study sought to create a model that could flag at-risk patients for further screening, potentially leading to earlier diagnosis and better health outcomes.

Study Approach

The research team used anonymized medical records from Maccabi Healthcare Services, a large health organization, to train and test five different machine learning models. These models were designed to predict whether a patient was at risk for celiac disease autoimmunity based on commonly available medical data, such as routine blood test results and demographic information. The study focused on patients who had high levels of tissue transglutaminase antibodies (a key marker for celiac disease) and compared them to a large group of controls with no evidence of the disease.

Machine Learning Models Evaluated

Celiac.com Sponsor (A12):
The researchers tested five different machine learning models to determine which was most effective in identifying at-risk individuals. The models included:

• XGBoost (Gradient Boosting Model)
• Logistic Regression
• Random Forest
• Multilayer Perceptron (Neural Network Model)
• Decision Tree

Among these models, XGBoost performed the best, correctly identifying at-risk patients with an accuracy score (AUC) of 0.86. Logistic regression followed closely with a score of 0.85, while the other models had slightly lower accuracy levels.

Key Predictors of Celiac Disease

The most effective model, XGBoost, identified several key medical indicators that were commonly present in patients with undiagnosed celiac disease. These included:

• Signs of anemia, such as low hemoglobin and ferritin levels
• Elevated liver enzymes, which may indicate liver inflammation
• Low high-density lipoprotein (HDL) cholesterol, often associated with untreated celiac disease

These findings align with known medical issues related to celiac disease, where nutrient absorption problems can lead to anemia and other metabolic imbalances.

Testing the Model Over Time

The model was tested to determine its ability to predict celiac disease risk years before an official diagnosis. It remained effective at identifying at-risk patients up to four years in advance. This suggests that machine learning could be a powerful tool in prompting earlier medical evaluation for patients who might otherwise remain undiagnosed.

Potential Benefits for Patients

The study highlights the importance of early detection in celiac disease. Diagnosing the condition earlier allows patients to begin a gluten-free diet sooner, which can help prevent long-term complications such as persistent gastrointestinal problems, malnutrition, and other autoimmune disorders. Early diagnosis also improves overall quality of life and reduces healthcare costs by preventing unnecessary medical procedures and hospital visits.

Limitations and Future Research

While the study showed promising results, there were some limitations. The model was trained using data from a single healthcare system, meaning its effectiveness in other populations remains uncertain. Additionally, the model relied solely on structured medical data, such as lab results, and did not include unstructured notes from doctors, which could provide further valuable insights. Future studies will need to validate the model’s accuracy in different healthcare systems and explore ways to improve its predictive ability.

Conclusion

This study demonstrates that machine learning can be a valuable tool in identifying individuals at risk for celiac disease before they develop severe symptoms. By using commonly available blood tests and demographic data, this approach has the potential to improve early detection and prompt further medical evaluation. If validated in broader studies, this tool could help doctors screen more effectively for celiac disease, leading to earlier intervention and better long-term health outcomes for patients.

Read more at: nature.com

Watch the video version of this article: