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

Faurholt-Jepsen, Maria; Rohani, Darius Adam; Busk, Jonas; Vinberg, Maj; Bardram, Jakob Eyvind; Kessing, Lars Vedel

doi:10.1186/s40345-021-00243-3

International Journal of Bipolar Disorders

Table 3 Classification within patients with bipolar disorder (n = 100) according to patient-reported smartphone-based data on mood, activity and sleep^a

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

Binary classifier n = number of observations	Model type	Accuracy (SD)	F1 score (SD)	Sensitivty (SD)	Specificity (SD)	AUC (SD)	B10 (SD)^b
Mood
Mania (n = 1205) versus euthymia (n = 38,329)	Random Forest model-user independent	0.72 (0.18)	0.05 (0.03)	0.23 (0.12)	0.73 (0.19)	0.51 (0.07)	− 2.65 (0.34)
	Random Forest model-user dependent	0.74 (0.16)	0.25 (0.22)	0.42 (0.22)	0.75 (0.16)	0.66 (0.11)	0.53 (3.69)
	Majority vote-user independent	0.94 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.18 (1.61)
	Majority vote-user dependent	0.97 (0.00)	0.08 (0.25)	0.11 (0.31)	0.89 (0.31)	0.48 (0.02)	− 3.23 (1.35)
	Random model-user independent	0.49 (0.00)	0.06 (0.03)	0.49 (0.05)	0.50 (0.01)	0.50 (0.00)	− 2.82 (0.27)
	Random model-user dependent	0.49 (0.00)	0.15 (0.09)	0.52 (0.03)	0.48 (0.03)	0.50 (0.00)	− 2.21 (0.59)
Depression (n = 5329) versus euthymia (n = 38,329)	Random Forest model-user independent	0.63 (0.06)	0.21 (0.07)	0.40 (0.07)	0.66 (0.05)	0.55 (0.05)	− 1.82 (0.22)
	Random Forest model-user dependent	0.70 (0.13)	0.40 (0.21)	0.53 (0.22)	0.70 (0.16)	0.66 (0.12)	2.78 (5.49)
	Majority vote-user independent	0.89 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.78 (1.31)
	Majority vote-user dependent	0.71 (0.00)	0.11 (0.28)	0.13 (0.33)	0.88 (0.33)	0.49 (0.02)	− 3.29 (0.00)
	Random model-user independent	0.50 (0.00)	0.24 (0.06)	0.51 (0.01)	0.49 (0.00)	0.50 (0.00)	− 2.07 (0.16)
	Random model-user dependent	0.46 (0.0)	0.31 (0.17)	0.55 (0.07)	0..44 (0.05)	0.50 (0.00)	− 1.80 (0.46)
Activity
Increased (n = 12,890) versus neutral (n = 21,661)	Random Forest model-user independent	0.46 (0.08)	0.43 (0.10)	0.55 (0.07)	0.41 (0.09)	0.48 (0.07)	− 1.38 (0.20)
	Random Forest model-user dependent	0.67 (0.13)	0.55 (0.25)	0.55 (0.26)	0.58 (0.24)	0.61 (0.10)	1.85 (5.92)
	Majority vote-user independent	0.48 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.52 (0.00)
	Majority vote-user dependent	0.57 (0.00)	0.45 (0.40)	0.56 (0.49)	0.44 (0.49)	0.49 (0.01)	− 0.02 (0.15)
	Random model-user independent	0.50 (0.00)	0.45 (0.07)	0.50 (0.01)	0.49 (0.01)	0.50 (0.00)	− 1.76 (0.07)
	Random model-user dependent	0.53 (0.00)	0.50 (0.13)	0.56 (0.05)	0.45 (0.05)	0.50 (0.00)	− 1.44 (0.63)
Decreased (n = 10,288) versus neutral (n = 21,661)	Random Forest model-user independent	0.50 (0.03)	0.42 (0.06)	0.58 (0.04)	0.47 (0.05)	0.54 (0.02)	− 1.43 (0.42)
	Random Forest model-user dependent	0.66 (0.11)	0.53 (0.18)	0.53 (0.20)	0.65 (0.21)	0.62 (0.10)	3.13 (6.40)
	Majority vote-user independent	0.69 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.04 (0.00)
	Majority vote-user dependent	0.71 (0.00)	0.27 (0.37)	0.35 (0.47)	0.65 (0.47)	0.49 (0.02)	− 0.02 (0.15)
	Random model-user independent	0.50 (0.00)	0.41 (0.07)	0.51 (0.01)	0.49 (0.01)	0.50 (0.00)	− 1.77 (0.10)
	Random modeluser dependent	0.55 (0.00)	0.45 (0.12)	0.55 (0.05)	0.44 (0.05)	0.50 (0.00)	− 1.39 (0.66)
Sleep
Insomnia (n = 8474) versus normal sleep (n = 36,140)	Random Forest model-user independent	0.70 (0.05)	0.13 (0.03)	0.16 (0.10)	0.82 (0.09)	0.48 (0.04)	− 1.49 (0.43)
	Random Forest model-user dependent	0.70 (0.16)	0.33 (0.18)	0.39 (0.22)	0.73 (0.17)	0.59 (0.08)	0.28 (2.66)
	Majority vote-user independent	0.85 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.13 (1.17)
	Majority vote-user dependent	0.90 (0.00)	0.05 (0.17)	0.06 (0.24)	0.93 (0.24)	0.49 (0.01)	− 3.36 (1.19)
	Random model-user independent	0.50 (0.00)	0.25 (0.07)	0.50 (0.01)	0.49 (0.00)	0.50 (0.00)	− 2.03 (0.13)
	Random model-user dependent	0.48 (0.00)	0.32 (0.13)	0.55 (0.05)	0.44 (0.04)	0.50 (0.00)	− 1.77 (0.48)
A broader definition of mania
Combined increased mood, activity and decreased sleep
Combined increased mood, activity and decreased sleep (n = 471) versus rest (n = 43,243)	Random Forest model-user independent	0.77 (0.10)	0.03 (0.02)	0.29 (0.17)	0.77 (0.11)	0.58 (0.06)	− 1.68 (0.28)
	Random Forest model-user dependent	0.77 (0.15)	0.17 (0.16)	0.41 (0.21)	0.78 (0.16)	0.67 (0.11)	2.15 (6.02)
	Majority vote-user independent	0.98 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.50 (0.00)	− 3.44 (0.00)
	Majority vote-user dependent	0.99 (0.00)	0.00 (0.00)	0.00 (0.00)	1.00 (0.00)	0.48 (0.02)	− 3.43 (1.30)
	Random model-user independent	0.49 (0.00)	0.13 (0.05)	0.50 (0.00)	0.49 (0.00)	0.50 (0.00)	− 2.39 (0.18)
	Random model-user dependent	0.46 (0.00)	0.46 (0.18)	0.56 (0.02)	0.44 (0.04)	0.50 (0.00)	− 1.78 (0.24)

The number of observations is the recorded samples before any resampling or cross-validation partitioning
^aEuthymia was defined as a mood score of − 0.5, 0, 0.5. Depression was defined as a mood score < − 0.5, and mania was defined as a mood score > 0.5. Increased activity was defined as a score > 0, and decreased activity was defined as a score < 0. Insomnia was defined as total sleep duration < 360 min. A broader definition of mania was made by combining increased mood, increased activity and decreased sleep
^bBayesian inference framework

Back to article page