Detecting Fatigue From Voice in Generalised Myasthenia Gravis

This study addresses a significant gap in understanding and measuring central fatigue in generalised myasthenia gravis (gMG), a debilitating symptom that differs from the characteristic muscle weakness fluctuations of the condition. Central fatigue encompasses mental and physical exhaustion originating in the central nervous system and remains poorly characterised with limited validated assessment tools.

Study Rationale and Innovation:

Recent developments in artificial intelligence and digital biomarkers have demonstrated potential for detecting fatigue-related changes in voice characteristics. This approach offers advantages over traditional assessment methods by providing objective, standardised measurements that can be collected remotely with minimal participant burden. Voice-based biomarkers may capture subtle physiological changes associated with central fatigue that are not readily apparent through conventional questionnaire-based assessments.

Study Design and Methodology:

This single-cohort observational study employs an intensive longitudinal monitoring design to capture the dynamic nature of fatigue fluctuations characteristic of gMG. The twice-daily assessment schedule (morning and evening sessions two days a week) over four weeks is designed to account for diurnal variation in fatigue symptoms commonly reported by MG patients.

Each assessment session lasts approximately 10-15 minutes and includes standardised voice recording tasks alongside validated fatigue questionnaires. Voice recording activities consist of structured reading tasks and answering questions out loud, designed to elicit natural speech patterns while maintaining consistency across sessions and participants.

Technical Approach:

Voice data will be analysed using machine learning algorithms to identify acoustic features potentially associated with central fatigue states. \[Note: Specific algorithmic approaches and feature extraction methods are proprietary and not detailed here\]. The study uses triangulated participant self-reported fatigue assessments as ground truth labels for model training and validation.

Data Collection and Management:

All data collection occurs remotely through a secure web-based platform accessible via standard internet browsers. Participants use their personal devices (computers, tablets, or smartphones) equipped with microphone capabilities. The platform captures voice recordings, questionnaire responses, and relevant metadata including device specifications and environmental conditions that may affect recording quality.

Sample Size Considerations:

The target enrolment of 240 participants is designed to generate sufficient data points for robust machine learning model development while accounting for expected attrition and technical issues.

• Study Completion Rate Among Enrolled Participants [Time frame: From enrolment through completion of final assessment session at 4 weeks]
• Individual Session Completion Rate Across All Participants [Time frame: From enrolment through completion of final assessment session at 4 weeks]
• Adherence to Specified Assessment Time Windows [Time frame: From enrolment through completion of final assessment session at 4 weeks]
• Participant Acceptability of Voice-Based Monitoring System [Time frame: At completion of final assessment session at 4 weeks]
• Participant Withdrawal Patterns and Reasons [Time frame: From enrolment through 4 weeks or until participant withdrawal]