Voice analyses using smartphone-based data in patients with bipolar disorder, unaffected relatives and healthy control individuals, and during different affective states

International Journal of Bipolar Disorders

Table 2 Discrimination between patients with bipolar disorder (BD) (n = 121), unaffected relatives (UR) (n = 21) and healthy control individuals (HC) (n = 38) based on voice features collected from smartphones, n = 180

Binary classifier (n = number of observations)	Model type	Accuracy (SD)	F1 score (SD)	Sensitivity (SD)	Specificity (SD)	AUC (SD)
BD (n = 78,731) compared with HC (n = 20,296)	Random Forest model	0.72 (0.09)	0.81 (0.07)	0.79 (0.11)	0.54 (0.20)	0.76 (0.11)
BD (n = 78,731) compared with HC (n = 20,296)	Majority vote	0.67 (0.00)	0.88 (0.06)	1.0 (0.00)	0.00 (0.00)	0.50 (0.00)
BD (n = 78,731) compared with UR (n = 8004)	Random Forest model	0.68 (0.049)	0.81 (0.03)	0.73 (0.07)	0.28 (0.11)	0.52 (0.09)
BD (n = 78,731) compared with UR (n = 8004)	Majority vote	0.95 (0.00)	0.95 (0.02)	1.00 (0.00)	0.00 (0.00)	0.50 (0.00)
UR (n = 8004) compared with HC (n = 20,296)	Random Forest model	0.59 (0.13)	0.38 (0.15)	0.53 (0.21)	0.67 (0.24)	0.72 (0.12)
UR (n = 8004) compared with HC (n = 20,296)	Majority vote	0.80 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)