Table 3.

Best Slip Outcome (LOB or Not) Prediction Models Using Logistic Regression for Proactive and Reactive Phases

Phase	Model factors		P	OR	Sensitivity/specificity, %	Accuracy, %
Proactive	1	Shank angle at SON	<.001	0.78	78.8/69.8	74.6
	2	Knee moment in MP	<.001	0.29	76.1/66.7	71.6
		Shank angle at MS	.04	0.9
Reactive	3	Shank angle at LLOa	<.001	0.69	84.7/85.7	85.1
	4b	Knee moment in ER	<.001	0.323	80.3/74.6	77.6
		Shank angle at SON	.001	0.67

Abbreviations: ER, early reactive phase; LLO, liftoff; LOB, loss of balance; MP, mid proactive phase; MS, mid swing; no-LOB, no loss of balance; OR, odds ratio; SON, slip onset. Note: MS is the initial instant of the mid proactive phase, and SON is the initial instant of the early reactive phase. Sensitivity indicated the predictive accuracy for LOB, and specificity indicated the predictive accuracy for no-LOB. For segment angles, it indicates the probability of having LOB decreases with a decrease of 1°, while for normalized joint moments, it indicates the probability of having LOB decreases with an increase of 1 SD in joint moments.

^aThe threshold for distinguishing LOB and no-LOB is 93°, which was calculated using logistic regression.

^bThe equation for model 4 could be represented as p(LOB) = 1/[1 + exp(39 − 0.39 × shank angle − 6.14 × knee moment)].