The FLATCAN Model: A Novel Score for Predicting Mortality Risk in Anti‐Melanoma Differentiation‐Associated Gene 5‐Positive Dermatomyositis

BACKGROUND AND OBJECTIVE: Anti-melanoma differentiation-associated gene 5-positive dermatomyositis (MDA5 + DM) exhibits the worst prognosis among all subtypes of idiopathic inflammatory myopathies, with substantial heterogeneity in patient outcomes. This study aimed to investigate prognostic factors for MDA5 + DM and develop a scoring system to determine mortality risk. METHODS: This retrospective study included 621 patients with MDA5 + DM. Variables were selected using univariable Cox regression and LASSO regression. Predictive models for mortality risks were constructed using machine learning-based algorithms. A simplified scoring system was established based on the optimal model with thorough validation to ensure predictive accuracy. RESULTS: Seven variables emerged as key factors associated with mortality in MDA5 + DM and incorporated into the mortality risk prediction model: ferritin, lactate dehydrogenase, age at onset, CD8+ T-cell count, C-reactive protein, albumin, and lung computed tomography pattern of NSIP + OP. Among six models, the Cox proportional hazards model demonstrated superior discriminative ability and clinical utility and was translated into a simplified scoring system 'FLATCAN'. This model achieved a concordance index of 0.815 and time-dependent area under the receiver operating characteristic curves for predicting 3-, 6-, and 12-month mortality of 0.895, 0.855, and 0.850, respectively. Patients were effectively stratified into low-, intermediate-, and high-risk groups using the FLATCAN score. Further internal cross-validation, time-point splitting, and rapidly progressive interstitial lung disease-based splitting confirmed the FLATCAN score's robust predictive ability. CONCLUSION: The FLATCAN score provides an easy-to-use tool for predicting mortality risk in patients with MDA5 + DM and may facilitate improved risk stratification-based patient management.