Prediction of Myasthenia Gravis Worsening: A Machine Learning Algorithm Using Wearables and Patient-Reported...

Area of Science:

• Neurology
• Digital Health
• Machine Learning

Background:

• Myasthenia gravis (MG) presents with fluctuating muscle weakness and potential crises.
• Limited access to care and fragmented data can delay critical interventions for MG patients.
• Developing predictive algorithms for MG deterioration is crucial for timely management.

Purpose of the Study:

• To develop predictive algorithms for myasthenia gravis (MG) deterioration.
• To utilize multimodal telemedicine data, including patient-reported and wearable sensor data.
• To enable earlier clinical interventions for worsening MG symptoms.

Main Methods:

• A 12-week randomized controlled study involving 30 MG patients.
• Utilized patient-reported outcome measures (PROMs) and patient-performed measures via a mobile app.
• Collected data from wearables and trained a machine learning linear classifier to predict deterioration.

Main Results:

• The predictive model achieved an AUROC of 0.85 using all input signals.
• Performance remained stable across 4-10 day look-back windows.
• PROMs indicated symptom worsening preceding deterioration; wearables alone showed higher predictive power.

Conclusions:

• Multimodal self-monitoring via the MyaLink system effectively predicts MG deterioration.
• The approach demonstrates good performance with acceptable false alarm rates.
• This technology holds potential for earlier clinical intervention in myasthenia gravis.