Short-term outcome prediction for myasthenia gravis: an explainable machine learning model

Area of Science:

• Medical Informatics
• Computational Biology
• Autoimmune Diseases

Background:

• Myasthenia gravis (MG) is a fluctuating autoimmune disease causing muscle weakness.
• Clinical management of MG is challenging due to its unpredictable course.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting short-term clinical outcomes in MG patients.
• To account for different antibody types in MG patient prognostication.

Main Methods:

• Utilized data from 890 MG patients across 11 tertiary centers in China (2015-2021).
• Employed a two-step variable screening process and 14 ML algorithms for model construction and optimization.
• Defined short-term outcome as modified post-intervention status (PIS) at 6 months.

Main Results:

• The ML model achieved high accuracy in predicting patient outcomes (improved, unchanged, worse) in both derivation (AUCs 0.89-0.91) and validation (AUCs 0.74-0.84) cohorts.
• The model demonstrated good calibration across datasets.
• The final model is interpretable, using 25 predictors and available as a web tool.

Conclusions:

• An explainable, ML-based predictive model can accurately forecast short-term outcomes for myasthenia gravis.
• This tool can assist clinicians in managing MG patients more effectively.