Comparison of different maximal isometric strength of lower limb muscle groups in predicting fall-risk among older persons.

Abstract

BackgroundMany independent studies have investigated the role of normalized maximal voluntary isometric strength (MVIS) of lower limb muscle groups (MVISLLMG) by body weight and summed knee and ankle muscle strength in predicting the risk of falling among older persons. However, it is unknown which MVISLLMG is better at predicting the fall risk.ObjectiveThis study aimed to compare different MVISLLMG in predicting fall-risk among older persons against the reference standard (history of falls).MethodsThis study had a cross-sectional retrospective diagnostic research design. 47 fallers and 93 non-fallers were recruited from Anand district, Gujarat, India, using sequential sampling. The MVISLLMG was measured with a microFET^®2 hand-held dynamometer. Following feature selection, four machine learning (ML) models (Random Forest (RF), k-Nearest Neighbors (KNN), Navie Bayes (NB), and Kernel Support Vector Machines (SVM)), were utilized to assess the diagnostic characteristics of every measured MVISLLMG in comparison to the reference standard. The best ML model was chosen based on the highest diagnostic performance in predicting fall-risk.ResultsAmong the ML models, the NB revealed that the non-normalized summed MVIS of knee and ankle muscle (Sensitivity (Se)$=$ 87%, Specificity (Sp)$=$ 91%, Accuracy (Ac)$=$ 90%, Precision (Pr)$=$ 84%) has the best diagnostic characteristics in fall-risk prediction against the fall history, followed by non-normalized MVIS of hip abductor, knee extensor, plantar flexor, and dorsiflexor, normalized summed MVIS of hip sagittal and knee muscle, and normalized MVIS of hip sagittal and frontal, knee, and plantar flexor.ConclusionThese results suggest that non-normalized summed MVIS of knee and ankle muscles is the better fall predictor in older persons compared to other index measures. This finding may assist clinicians in playing a better role in selecting suitable MVISLLMG data for fall risk assessment and predicting falls.