Time in Bed Comparisons Between Two Automated Algorithms Applied to activPAL Data and Observations in a Polysomnography Lab

Objective: This study compared estimates of time in bed (TIB) and TIB start and end from two algorithms applied to activPAL data with observations of TIB in a polysomnography (PSG) lab. Methods: Twenty-five healthy volunteers (age: 32.4 ± 7.4 years, body mass index: 25.2 ± 3.7 kg/m2) wore a thigh-worn activPAL accelerometer during one-night laboratory-based PSG assessment. TIB and TIB start and end estimates were generated using automated algorithms within Processing PAL and PAL Technologies (PAL Batch software, version 8) applications. Agreement was determined using pairwise 95% equivalence tests (±10% equivalence zone), mean percentage absolute error, intraclass correlation coefficients (ICCs), and 95% limits of agreement. Results: The results for the algorithms were similar. TIB start and end were within the proposed ±10% equivalence zone of the PSG-lab observations, but TIB was not when using either algorithm. Mean percentage absolute errors for both algorithms were approximately 10%, 3%, and 1% for TIB and TIB start and end, respectively. Reliability between both algorithms and PSG lab was poor for TIB (ICC ≥ .39) and TIB start (ICC ≥ .38) and good for TIB end (ICC ≥ .81). Both algorithms recorded more TIB by about 42 ± 60 min and detected earlier TIB start and later TIB end by about 18 ± 62 and 14 ± 9 min, respectively, compared with PSG. Negative fixed biases (PSG–activPAL) were observed for TIB and TIB end from both algorithms (p < .05). For both algorithms, 95% limits of agreement were ±120 min for TIB and ±125 min for TIB start. Removing two outliers improved the agreement between both algorithms and PSG-lab observations. Conclusions: Processing PAL and PAL Technologies algorithms recorded more TIB by ∼40 min compared with PSG-lab TIB. Researchers should be aware of this discrepancy when using these algorithms and utilize predefined quality control protocols to enhance data quality. Future studies should continue refining these algorithms.