Predicting achievement of clinical goals using machine learning in myasthenia gravis

Abstract

Background Myasthenia Gravis (MG) is an autoimmune disease characterized by the production of autoantibodies against neuromuscular junctions, leading to varying degrees of severity and outcomes among patients. This variability makes clinical evaluation crucial for determining appropriate treatment targets. However, accurately assessing Minimal Manifestation (MM) status is challenging, requiring expertise in MG management. Therefore, this study aims to develop a diagnostic model for MM in MG patients by leveraging their clinical scores and machine learning approaches. Methods This study included 1,603 MG patients enrolled from the Japan MG Registry in the 2021 survey. We employed non-negative matrix factorization to decompose three MG clinical scores (MG composite score, MGADL scale, and MG quality of life (QOL) 15r) into four distinct modules: Diplopia, Ptosis, Systemic symptoms, and QOL. We developed a machine learning model with the four modules to predict MM or better status in MG patients. Using 414 registrants from the Japan MG Registry in the 2015 survey, we validated the model’s performance using various metrics, including area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, accuracy, F1 score, and Matthews Correlation Coefficient (MCC). Results The ensemble model achieved an AUROC of 0.94 (95% CI: 0.94–0.94), accuracy of 0.87 (95% CI: 0.86–0.88), sensitivity of 0.85 (95% CI: 0.85–0.86), specificity of 0.89 (95% CI: 0.88–0.91), precision of 0.93 (95% CI: 0.92–0.94), F1 score of 0.89 (95% CI: 0.88–0.89), and MCC of 0.74 (95% CI: 0.72–0.75) on the validation dataset. Conclusions The developed MM diagnostic model can effectively predict MM or better status in MG patients, potentially guiding clinicians in determining treatment objectives and evaluating treatment outcomes.