1258 Is My Sleep Tracker Tracking My (REM) Sleep? Comparison of Two Consumer Smartwatch Devices to Polysomnogram, in a Sleep Clinic Referral Population

Introduction The popularity of consumer sleep-tracker smartwatch technology continues to grow, but uncertainty remains over data accuracy and applicability to patients with sleep disorders. Methods Subjects undergoing evaluation for sleep disorders at a single sleep center (Wyomissing, PA), during a 16-month period ending September 2025, were enrolled in the study on the night of their scheduled polysomnogram (PSG). Subjects wore Fitbit Inspire 3 (FBI3) and Garmin VivoSmart 5 (GVS5) sleep-tracker devices during the PSG. Summary sleep staging data (total rapid eye movement [REM] sleep and total sleep time [TST]) were analyzed by descriptive statistics, paired Student's t-Test, and equivalence testing (using Two One-Sided Tests [TOST]), versus gold-standard PSG. Results A total of 110 subjects enrolled; 92 provided complete data. Subjects were excluded from the analysis due to early termination of PSG (n=2) or sleep-tracker device failure (FBI3 [n=13]; GVS5 [n=8]). Total minutes of REM sleep mean +/- standard deviation (SD) were similar across devices: PSG 66.9 (SD 30.7), FBI3 70.0 (SD 27.5) and GVS5 70.0 (SD 44.5) minutes. Paired Student's t-Test showed no significant difference in PSG total REM sleep compared to FBI3 (t=0.95, p=0.35) or GVS5 (t=0.63, p=0.53). Testing of equivalence using TOST (employing prespecified 90% upper and lower confidence intervals of +/- 10 minutes), demonstrated that FBI3-measured REM sleep was equivalent (i.e., no substantial differences in means) to PSG-measured REM (mean difference –3.1 min [CI -8.5; +2.3]); GVS5 sleep-tracker did not meet equivalence (mean difference -3.1 min [-11.3; +5.1]). Sleep-tracker devices measured total sleep time with comparatively less accuracy: PSG 355.1 (SD 69.8) minutes; FBI3 365.9 (SD 77.1) minutes; GVS5 444.9 (SD 73.0) minutes. Although the FBI3 device-measured total sleep time did not differ significantly from PSG (Student's t-Test [t=1.50, p=0.138]), the GVS5 device highly overestimated TST (t=10.4, p< 0.01]). Conclusion Among sleep medicine patients, sleep-tracker smartwatches provided potentially clinically relevant estimates of total REM sleep, but showed wider variability in measurements of total sleep time, depending on the device, compared to the gold-standard polysomnogram. Support (if any) The Reading Hospital Foundation (Reading, PA) provided grant funding.