JMIR Publications

ABSTRACT

TPurpose: The aim of the herein presented study, was to compare the predictive value of insole data, which was collected during the Timed-Up-and-Go Test (TUG-Test), to the SARC-F and TUG-Test, to assess physical frailty (PF), defined by the Short Physical Perfor-mance Battery (SPPB), using machine learning algorithms.

Methods:

This prospective diagnostic study included patients aged >60 years, with in-dependent ambulation and no mental or neurological impairment. A comprehensive set of parameters associated to PF was assessed, including body composition, question-naires (EQ-5D, SARC-F), and physical performance tests (SPPB, TUG) including digital sensor insoles gait parameters during the TUG Test. PF was defined as a SPPB score ≤ 8 points. Advanced statistics, including random forest feature selection and machine learning algorithms (K nearest neighbors and random forest) were used to compare the diagnostic value of these parameters to identify PF patients.

Results:

Classified by the SPPB, 23 of the 57 eligible patients were defined as PF. Sev-eral gait parameters were significantly different between the two groups. The receiver operator curve of the TU Test was superior to that of the SARC-F (0.862 vs. 0.639). The recursive feature elimination algorithm identified 9 parameters, 8 of which were digital insole gait parameters. Both the K-Nearest Neighbors (KNN) algorithm and the Random Forest trained with these parameters resulted in excellent results (RF = 0.919 and KNN = 0.801). Conclusion: A gait analysis based on machine learning algorithms using sensor soles is superior to the SARC-F and the TUG Test to identify PF patients in orthogeriatric pa-tients.