Table 4.

Group-level proportional error matrix in the sample of healthy sleepers and adolescents with insomnia symptoms.

			Fitbit Charge 3™
			Wake	“light”	“deep”	REM	kappa	PABAK
PSG	Wake	Healthy sleepers	.68 (.18) [.60, .75]	.21 (.15) [.16, .27]	.01 (.02) [.00, .02]	.10 (.11) [.06, .14]	.52 (.14) [.47, .58]	.83 (.10) [.79, .87]
		Insomnia symptoms	.81 (.12) [.74, .87]	.12 (.08) [.08, .17]	.00 (.01) [.00, .01]	.07 (.07) [.03, .10]	.59 (.14) [.52, .66]	.86 (.03) [.84, .87]
	“light”	Healthy sleepers	.08 (.05) [.06, .09]	.78 (.07) [.75, .81]	.07 (.04) [.05, .09]	.07 (.05) [.05, .09]	.48 (.14) [.43, .54]	.48 (.14) [.42, .54]
		Insomnia symptoms	.08 (.02) [.06, .09]	.71 (.13) [.64, .78]	.08 (.08) [.04, .12]	.13 (.10) [.08, .19]	.42 (.18) [.32, .52]	.43 (.18) [.33, .52]
	“deep”	Healthy sleepers	.02 (.02) [.01, .02]	.39 (.24) [.29, .48]	.59 (.25) [.50, .69]	.00 (.01) [.00, .01]	.48 (.14) [.43, .54]	.48 (.14) [.42, .54]
		Insomnia symptoms	.02 (.02) [.01, .03]	.40 (.17) [.31, .49]	.57 (.17) [.48, .66]	.01 (.03) [.00, .03]	.42 (.18) [.32, .51]	.43 (.18) [.33, .52]
	REM	Healthy sleepers	.03 (.03) [.02, .04]	.20 (.17) [.14, .27]	.01 (.03) [.00, .02]	.76 (.18) [.68, .83]	.56 (.25) [.47, .66]	.74 (.14) [.68, .79]
		Insomnia symptoms	.05 (.06) [.02, .09]	.26 (.17) [.17, .36]	.00 (.01) [.00, .01]	.69 (.19) [.58, .79]	.56 (.18) [.47, .66]	.73 (.12) [.68, .80]

“light”, PSG-based N1 + N2; “deep”, PSG-based N3; REM, Rapid-Eye-Movement sleep; PSG, polysomnography; PABAK, prevalence-adjusted bias-adjusted kappa. Results are reported as mean (standard deviation) [95% confidence intervals]. Bold type indicates sleep-stage sensitivity (device’s ability to correctly detect a given stage), whereas the remaining cells indicate percentages of misclassifications. Confidence intervals were computed using percentile bootstrap for the group with insomnia symptoms due to skewed distributions and small sample size.