Do kinematic parameters of gait discriminate fallers and non-fallers in Parkinson's disease?

Introduction Many clinical factors have been associated with falling in PD but the utility of gait analysis remains uncertain. To determine if additional use of clinical characteristics of the disease combined with gait parameters distinguish PD fallers from non-fallers Material and methods One hundred and seventy-four subjects were recorded during stabilized with a videomotion system. Clinical characteristics included UPDRS 3, disease duration, MMSE, occurrence of falls and freezing of gait. Kinematic parameters of gait first include gait speed, stride length and time, gait asymmetry, gait variability, foot clearance. A classification approach was investigated to build models allowing distinguishing fallers from non-fallers. Two different models were built, one by considering only the clinical data: age, disease duration, UPDRS III and MMSE scores and the second by adding to these data, gait analysis data. Support vectors machines (SVM) algorithm was used to build the models. The models were learned to tune their parameters using 4/5 of the patients’ data, randomly selected and validated blindly of the remaining 1/5 data. Results Clinical factors have an accuracy of 0.91 to distinguish fallers from non-fallers. Adding some relevant parameters such as stride length, foot clearance, variability improves this accuracy of 0.95. This was confirmed with a test sample with an accuracy of 0.87. Discussion/conclusion Fallers distinguished from non-fallers by disease duration and motor impairment but adding gait measurement improves the accuracy of the clinical data. Further research should concentrate to validate these models in prospective cohorts.