Machine-learning algorithms for predicting colchicine resistance in Familial Mediterranean Fever

Area of Science:

• Genetics and immunology
• Computational biology
• Rheumatology

Background:

• Familial Mediterranean Fever (FMF) is a genetic autoinflammatory disease.
• Colchicine is the standard treatment, but 5-10% of adult patients show resistance.
• Early identification of colchicine resistance is crucial for effective management.

Purpose of the Study:

• To explore the potential of machine learning (ML) and deep learning (DL) algorithms for predicting colchicine resistance in adult FMF patients.
• To develop AI-based models for early identification of treatment resistance.
• To facilitate personalized therapeutic strategies for FMF management.

Main Methods:

• Retrospective analysis of 965 adult FMF patients with confirmed genetic diagnosis and follow-up.
• Feature selection included mutation type, MEFV mutations, arthritis, arthralgia, age at diagnosis, and attack frequency.
• Development and evaluation of logistic regression and fully connected neural network models using AUC metrics.

Main Results:

• Both logistic regression and deep learning models achieved an Area Under the Curve (AUC) of 0.79.
• Significant predictors of colchicine resistance included homozygous mutation type, recurrent arthritis, chronic arthralgia, age of diagnosis, and attack frequency.
• These factors highlight key clinical and genetic indicators associated with treatment response.

Conclusions:

• AI-based algorithms, especially deep learning, demonstrate significant potential for predicting colchicine resistance in adult FMF.
• These predictive models can aid clinicians in making timely treatment decisions.
• The findings support the development of tailored therapeutic approaches for FMF patients resistant to colchicine.