Development and clinical validation of inertial sensor-based gait-clustering methods in Parkinson's disease.

Patients had to be able to walk independently without a walking aid and were tested in stable ON medication, so results may not apply to people with more advanced disease, walking aids, or off-medication fluctuations. Patients were selected to balance impaired and unimpaired groups, so classification performance may not reflect a routine clinic sample. Stride segmentation was semi-automatic and required manual correction, so the method is not fully automated. The turning isolation algorithm and the definitions of constant and non-constant strides were not validated. The analysis combined UPDRS-III scores of 1 and 2 into one impaired group, so it cannot distinguish milder from more severe impairment within that group. It used only the 4x10 m test and one internal train/test split from a single outpatient cohort; the paper does not report external validation. The non-constant cluster contains few strides, so variability measures such as coefficient of variation may be sensitive to outliers. The study cannot separate different domains of balance impairment because it only used UPDRS-III gait and postural stability scores.

The study was funded by the Bavarian State Ministry for Economic Affairs, Media, Energy and Technology and by EIT Health. The supplied text does not include an author conflict-of-interest declaration.

Do not conclude that this sensor method can diagnose Parkinson's disease or guide treatment decisions. It only showed better classification of impaired versus unimpaired clinical gait and postural stability scores in one cohort, and the stride segmentation required manual correction.