Table 4.
Group-level discrepancies computed on the sample data
| Measure | Device mean (SD) | Reference mean (SD) | Bias [95% CI] | Lower LOA [95% CI] | Upper LOA [95% CI] |
|---|---|---|---|---|---|
| TST (min) | 339.32 (60.61) | 332.57 (67.25) | 56.42 − 0.15 × ref b0 = [10.03, 104.04] b1 = [−0.28, −0.03]* | Bias − 39.25 [23.06, 62.90] | Bias + 39.25 [23.06, 62.90] |
| SE (%) | 88.07 (4.53) | 86.08 (6.90) | 58.10 − 0.65 × ref b0 = [31.53, 98.51] b1 = [−1.11, −0.35]* | Bias − 2.46 × (24.90 − 0.26 × ref) c0 = [28.37, 64.51] c1 = [−0.71, −0.30]* | Bias + 2.46 × (24.9 − 0.26 × ref) c0 = [28.37, 64.51] c1 = [−0.71, −0.30]* |
| SOL (min) | 11.36 (14.81) | 13.75 (11.60) | 7.15 − 0.69 × ref b0 = [2.29, 20.15] b1 = [−2.53, −0.30]* | Bias − 2.46 × (0.82 + 0.53 × ref) c0 = [−12.72, 2.95] c1 = [0.30, 1.54]* | Bias + 2.46 × (0.82 + 0.53 × ref) c0 = [−12.72, 2.95] c1 = [0.30, 1.54]* |
| WASO (min) | 33.82 (12.76) | 38.18 (19.79) | 26.13 − 0.80 × ref b0 = [8.10, 49.05] b1 = [−1.56, −0.23]* | Bias − ref × 0.94 [0.69, 1.36] | Bias + ref × 0.94 [0.69, 1.36] |
| “Light” (min) | 234.64 (56.85) | 200.11 (48.52) | 34.54 (52.86) [7.75, 61.43]* | −69.08 [−95.47, −42.25] | 138.15 [111.43, 164.29] |
| “Deep” (min) | 50.04 (22.58) | 75.61 (23.71) | 23.70 − 0.65 × ref b0 = [−3.37, 61.43] b1 = [−1.16, −0.22]* | Bias − 2.46 × (2.66 + 0.18 × ref) c0 = [−37.43, 12.02] c1 = [0.07, 0.71]* | Bias + 2.46 × (2.66 + 0.18 × ref) c0 = [−37.43, 12.02] c1 = [0.07, 0.71]* |
| REM (min) | 54.64 (27.39) | 56.86 (23.11) | −2.21 (36.23) [−19.79, 16.21] | Bias − ref × 1.24 [1.04, 1.84] | Bias + ref × 1.24 [1.04, 1.84] |
TST, total sleep time; SE, sleep efficiency; SOL, sleep onset latency; WASO, wake after sleep onset; “light”, PSG-derived N1 + N2; “deep”, PSG-derived N3; REM, rapid eye movement sleep; SD, standard deviation; CI, confidence intervals; LOA, limit of agreement; ref, reference-derived measures (i.e. PSG, used to quantify the size of measurement).
*Cases showing a significant bias, proportional bias or heteroscedasticity. When a proportional bias was detected, a linear model predicting the discrepancies by the corresponding PSG measures was specified, and 95% CI were reported for the model’s intercept (b0) and slope (b1), as indicated in equation (1). When heteroscedasticity was detected, a linear model predicting the absolute residuals of the previous model by PSG-derived measures was specified, and 95% CI were reported for the model’s intercept (c0) and slope (c1), as indicated in equation (3).