Instrumented timed up and go test and machine learning-based levodopa response evaluation: a pilot study.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• The acute levodopa challenge test (ALCT) is standard for assessing levodopa response (LR) in Parkinson's disease.
• Current ALCT relies on the Movement Disorder Society's Unified Parkinson's Disease Rating Scale part III (MDS-UPDRS III), which can be subjective and cumbersome.

Purpose of the Study:

• To develop and validate a machine learning (ML) method using the instrumented Timed Up and Go (iTUG) test to objectively evaluate LR.
• To compare the ML-based iTUG method with the conventional ALCT for LR assessment.

Main Methods:

• Forty-two parkinsonism patients underwent both ALCT and iTUG testing in OFF- and ON-medication states.
• Kinematic, time, and frequency domain features were extracted from iTUG sensor data.
• Two XGBoost models (LRR and MSE) were trained to predict LR using leave-one-subject-out cross-validation.

Main Results:

• The ML-based levodopa response regression (LRR) model showed high agreement with ALCT (ICC = 0.95).
• The LRR model achieved a positive predictive value of 0.94 for detecting positive levodopa response.

Conclusions:

• ML applied to iTUG wearable sensor data provides an effective and comprehensive approach for evaluating levodopa response.
• This method shows promise for predicting the efficacy of dopaminergic therapy in Parkinson's disease.