Amyloidosis secondary to familial Mediterranean fever: machine learning-based prediction models

FMF is a monogenic autoinflammatory disease caused by MEFV mutations, with amyloidosis as its most severe complication. This study aimed to compare logistic regression, random forest, and gradient-boosting models to predict amyloidosis in FMF patients.

Patients with FMF diagnosed between 1990 and 2022 at Cerrahpaşa Faculty of Medicine were retrospectively screened. Eligible patients with available clinical records were included. Clinical, genetic, and laboratory variables were extracted. Univariate analyses and pooled multivariate logistic regression were conducted. Logistic regression, random forest, and gradient-boosting machine-learning models were developed. Models were optimized by hyperparameter tuning and evaluated on a held-out test set.

Of the 615 FMF patients included, 58 (9.4%) had amyloidosis. Patients with amyloidosis had earlier symptom onset, longer diagnostic delay and disease duration, were more often male, and more frequently carried M694V homozygosity. Comorbidities, parental consanguinity, frequent infections, erysipelas-like erythema, myalgia, arthritis, and higher median CRP levels were more prevalent in the amyloidosis group. In multivariate analysis, disease duration, M694V homozygosity, comorbidity, parental consanguinity, frequent infections, and erysipelas-like erythema were independently associated with amyloidosis. Random forest achieved the best performance on the test set (area under the receiver operating characteristic = 0.784), followed by logistic regression (0.766) and gradient boosting (0.749). SHAP analysis identified M694V homozygosity, myalgia, erysipelas-like erythema, disease duration, and arthritis as the strongest contributors to amyloidosis prediction.

Amyloidosis, the most severe FMF complication, is driven by key clinical and genetic factors, with ensemble machine-learning models outperforming conventional regression for early risk prediction.