Figure 3.

Faller and recurrent faller optimal predictors: results of the LASSO logistic regression. The variables simultaneously tested in the LASSO logistic regression models were: age, gender (M: male), condition (Chr: chronic disease), cognitive impairment (Cimp), urinary incontinence (UI), the TUG test total duration (TUG), the Mini-BESTest (MB) and the turn duration of the TUG test (Turn). White bars: fall risk factors from the medical history; blue bars: mobility measures from clinical tests; green bars: ITUG measures. The bars mark the predictors selected by the LASSO procedure, while variables without bars are those whose coefficients were shrunk to zero by the LASSO. Upward bars indicate positive predictors’ coefficients (i.e., variables positively associated with the faller or recurrent faller risk). Downward bars indicate otherwise. For example, M decreased the risk of being a faller, while Chr increased this risk. According to the LASSO logistic regression, the predictors of the optimal model for faller risk assessment (A) were: male gender, chronicity, urinary incontinence, the Mini-BESTest logit measure, and the turn duration. Optimal predictors for recurrent faller (B) were chronicity, cognitive impairment, and the Mini-BESTest.