Isometric strength of quadriceps as a best fall predictor among older adults: analysis by machine learning approach

Abstract

Abstract This study aimed to: i) investigate which maximum voluntary isometric strength (MVIS) of lower limb muscle groups has good sensitivity (Se) and specificity (Sp) in predicting fall among older adults against the reference standard (history of fall) and their comparison with functional mobility (FM) and fear of falling (FoF); ii) identify the top three index measures in influencing fall to facilitate targeted assessment and exercise prescription by clinicians. A cross-sectional diagnostic study was conducted among one hundred and forty older adults (47 fallers) and (93 non-fallers) and recruited using consecutive sampling. The MVIS of plantar flexors, dorsiflexors, knee extensors and flexors, hip flexors, extensors, abductors, and adductors were measured using a microFET®2 hand-held dynamometer, FM using the timed up and go test, and FoF using the modified fall efficacy scale. The Se, Sp, accuracy (Ac), and precision (Pr) of all index measures against the reference standard were evaluated by four machine learning (ML) models. The top index measures in influencing fall were evaluated by mean decline in the Gini index (DGI). Among ML models, the Random Forest revealed that the MVIS of the quadriceps (Se = 81%, Sp = 90%, Ac = 87%, Pr = 80%) has the highest diagnostic features against the reference standard compared to other muscle groups, FM (Se = 48%, Sp = 75%, Ac = 66%, Pr = 50%), and FoF (Se = 46%, Sp = 79%, Ac = 68%, Pr = 53%). DGI suggests that the MVIS of quadriceps has the greatest influence on fall, followed by plantar flexors and hip flexors, while FoF has the least. These results suggest that MVIS of the quadriceps is the best fall predictor in older adults, followed by plantar flexors and hip flexors. The findings imply that clinicians can better predict and impact falls in older adults by targeting muscles with higher fall influence during intervention prescribing.