Skip to main content
. 2019 Nov 28;21(11):e16273. doi: 10.2196/16273

Table 2.

Results from qualifying publications involving both early-generation, nonsleep-staging and newer-generation, sleep-staging Fitbit models.

Author (year) Model Reference Resultsa
Beattie et al (2017) [22]b Surge Type III home PSGc

  • Normal mode vs PSG. Overestimated TSTd,e (46 min) and SEe,g (8.1%); underestimated WASOe,h (44 min) and SOLi (2 min, NSj); accuracy, 0.88 (SD 0.05); sensitivity, 0.98 (SD 0.02); specificity, 0.35 (SD 0.13)

  • Normal mode vs actigraphy. Overestimated SOLe (12 min), SEf (1.1%), and TST (5 min, NS); underestimated WASOe (17 min)

  • Sensitive mode vs PSG. Underestimated TSTe (86 min) and SEe (16.0%); overestimated SOLf (12 min) and WASOe (75 min); accuracy, 0.78 (SD 0.08); sensitivity, 0.78 (SD 0.09); specificity, 0.80 (SD 0.17)

  • Sensitive mode vs actigraphy. Underestimated TSTe (127 min) and SEe (22.9%); overestimated SOLe (25 min) and WASOe (102 min)
Brazendale et al (2019) [23] Charge HR Sleep log & actigraphy

  • Fitbit correlated with both actigraphy (r=.48)f and sleep log (r=.71)f for measuring TST
Brooke et al (2017) [24] Flex & Charge HR Sleep log

  • Both Fitbit Flex (r=.68, MAPEk=8.80%) and Fitbit Charge HR (r=.58, MAPE=11.5%) correlated with sleep log in measuring TSTe
Cook et al (2017) [25] Flex PSG &
actigraphy

  • Normal mode vs PSG. Overestimated TSTe (46 min) and SEe (8.1%); underestimated WASOe (44 min) and SOL (2 min, NS); accuracy, 0.88 (SD 0.05); sensitivity, 0.98 (SD 0.02); specificity, 0.35 (SD 0.13)

  • Normal mode vs actigraphy. Overestimated SOLe (12 min), SEf (1.1%), and TST (5 min, NS); underestimated WASOe (17 min)

  • Sensitive mode vs PSG. Underestimated TSTe (86 min) and SEe (16.0%); overestimated SOLf (12 min) and WASOe (75 min); accuracy, 0.78 (SD 0.08); sensitivity, 0.78 (SD 0.09); specificity, 0.80 (SD 0.17)

  • Sensitive mode vs actigraphy. Underestimated TSTe (127 min) and SEe (22.9%); overestimated SOLe (25 min) and WASOe (102 min)
Cook et al (2019) [26]b Alta HR PSG

  • Overestimated TSTf (12 min), SEf (2.0%), and deep sleepe (18 min); underestimated SOL (4 min, NS), WASO (8 min, NS), and light sleep (11 min, NS); accuracy, 0.90 (SD 0.04); sensitivity, 0.96 (SD 0.02); specificity, 0.58 (SD 0.16); accuracy in detecting light sleep, 0.73; deep sleep, 0.89; REMl sleep, 0.89
de Zambotti et al (2016) [27] Charge HR PSG

  • Overestimated TSTf (8 min) and SEe (1.8%); underestimated WASOf (6 min) and SOL (3 min, NS); accuracy, 0.91 (SD 0.05); sensitivity, 0.97 (SD 0.02); specificity, 0.42 (SD 0.16); predictive value for sleep, 0.93 (SD 0.05); predictive value for wake, 0.65 (SD 0.18)
de Zambotti et al (2018) [28]b Charge 2 PSG

  • Normal sleeper cohort. Overestimated TSTf (9 min) and light sleepe (34 min); underestimated SOLf (4 min), deep sleepe (24 min), WASO (5 min, NS), and REM sleep (1 min, NS); sensitivity, 0.96; specificity, 0.61; accuracy in detecting light sleep, 0.81; deep sleep, 0.49; REM sleep, 0.74

  • PLMSj cohort. Underestimated deep sleepf (28 min), SOL (7 min, NS), and WASO (1 min, NS); overestimated TST (8 min, NS), light sleep (36 min, NS), and REM sleep (0 min, NS); specificity, 0.62; accuracy detecting light sleep, 0.78; deep sleep, 0.36; REM sleep, 0.62
Dickinson et al (2016) [29] Charge HR Actigraphy

  • No systematic difference across days between Fitbit and actigraphy in measuring TST and SE
Hakim et al (2018) [30] Charge HR PSG

  • Overestimated TSTf (30 min); underestimated total wake timef (23 min)
Kang et al (2017) [31] Flex Unattended PSG

  • Good sleepers—normal mode. Overestimated TSTf (7 min), SE (1.8%, NS), and SOL (1 min, NS); underestimated WASO (7 min, NS); accuracy, 0.93; sensitivity, 0.97; specificity, 0.36

  • Insomniacs—normal mode. Overestimated TSTe (33 min) and SEe (7.9%); underestimated WASOe (31 min) and SOL (2.4%, NS); accuracy, 0.87; sensitivity, 0.97; specificity, 0.36

  • Good sleepers—sensitive mode. Accuracy, 0.66; sensitivity, 0.65; specificity, 0.82

  • Insomniacs—sensitive mode. Accuracy, 0.68; sensitivity, 0.64; specificity, 0.89
Kubala et al (2019) [32] Alta Actigraphy

  • Good sleepers. Overestimated TSTe (74 min); underestimated WASOf (16 min)

  • Poor sleepers. Overestimated TST (20 min, NS); underestimated WASO (13 min, NS)
Lee et al (2017) [33] Charge HR Actigraphy

  • Overestimated TSTe (22 min); correlation between Fitbit and actigraphy: sleep start timese (r=.87) and TSTe (r=.92)
Lee et al (2018) [34] Charge HR Sleep log

  • Correlation between Fitbit and sleep log: TSTe (r=.55, MAPE 14.2%) and TIBe,n (r=.48, MAPE 12.7%); SE and WASO not correlated
Liang and Chapa Martell (2018) [35] Charge 2 Sleep Scope (EEGo based)

  • Overestimated WASOe (25 min) and deep sleepe (40 min); underestimated TSTf (12 min), SOLe (11 min), REMe sleep (12 min), light sleepe (42 min), and SE (1.5%, NS)
Liu et al (2019) [36]b Alta HR Sleep log

  • Overestimated WASOf (13 min); underestimated TSTf (6 min), SOLf (5 min), and SEf (1.4%)
Mantua et al (2016) [37] Flex PSG
(ambulatory
system)

  • No significant difference in measuring TST and SE; TST correlatede (r=.97); SE not correlated (r=.21, NS); average percentage error: TST, 2.97%; SE, 11.57%
Maskevich et al (2017) [38] One PSG

  • Overestimated TSTe (88 min) and SEe (17.4%); underestimated WASOf (39 min) and SOL (17 min, NS); accuracy, 0.81 (0.68-0.93); sensitivity, 0.99 (0.97-1.00); specificity, 0.27 (0.12-0.55); predictive value for sleep, 0.99, and wake, 0.27
Meltzer et al (2015) [39] Ultra PSG &
actigraphy (2
different ones)

  • Fitbit—normal mode vs PSG. Underestimated WASOe (32 min); overestimated TSTe (41 min) and SEe (8%); accuracy, 0.84; sensitivity, 0.87; specificity, 0.52

  • Fitbit—sensitive mode vs PSG. Underestimated TSTe (105 min) and SEe (21%); overestimated WASOe (106 min); accuracy, 0.71; sensitivity, 0.70; specificity, 0.79
Montgomery-Downs et al (2012) [40] Classic PSG &
actigraphy

  • Fitbit vs PSG. Overestimated SEe (14.5%) and TSTe (67 min); sensitivity, 0.98 (0.92-1.00); specificity, 0.20 (0.02-0.78)

  • Fitbit vs actigraphy. Overestimated SEe (5.2%) and TSTe (24 min)
Osterbauer et al (2016) [41] Flex PSG

  • TST by Fitbit and PSG correlatedf (rhop=.99); WASO, SE, and awake minutes not correlated; sensitivity, 0.99; specificity, 0.10
Sargent et al (2018) [42] Charge HR PSG

  • TST by Fitbit vs PSG: NS; Fitbit automatically identified 60% of sleep periods, with a success rate of 80% when sleep was 9h, 90% when sleep was 8h, 70% when sleep was 7h, 50% when sleep was 2h, and 10% when sleep was 1h
Svensson et al (2019) [43]b Versa Sleep Scope (EEG based)

  • Overestimated TIB (9 min, NS), TST (7 min, NS), WASOe (14 min), and deep sleepe (36 min); underestimated SE (0.1%, NS), SOLe (14 min), REMe sleep (6 min), and light sleepe (20 min); accuracy, 0.89 (0.88-0.89); sensitivity, 0.92 (0.919-0.923); specificity, 0.54 (0.53-0.55)

aAccuracy, sensitivity, and specificity in detecting sleep epochs are reported unless otherwise specified.

bPublication consisting of newer-generation, sleep-staging Fitbit models.

cPSG: polysomnography.

dTST: total sleep time.

eP<.01.

fP<.05.

gSE: sleep efficiency.

hWASO: wake after sleep onset.

iSOL: sleep onset latency.

jNS: not significant.

kMAPE: mean absolute percent error.

lREM: rapid eye movement.

mPLMS: periodic limb movement in sleep.

nTIB: time in bed.

oEEG: electroencephalographic.

pSpearman correlation coefficient.