Table 2.
Top performing model predicting total and fluid composite and subtest scores on the NIH Toolbox Cognition Battery
| Test | cycles | fc (Hz) | q | th | min (s) | max (s) | merge (s) | FFT (Hz) | Duration (s) | Density (spm) | Amplitude (uV) | Pearson’s r | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Value | 95% CI | Value | 95% CI | Value | 95% CI | Value | 95% CI | Value | 95% CI | ||||||||
| Total Comp. | 7 | 14 | 0.7 | 6 | 0.4 | 3.5 | 1 | −2.40 | [−4.26,−0.61] | 2.63 | [0.27, 4.56] | 3.06 | [1.14, 5.18] | 1.05 | [0.21, 2.49] | 0.41 | [0.07, 0.50] |
| Fluid Comp. | 6 | 15.5 | 0.3 | 5.5 | 0.4 | 2.6 | 0.7 | −6.48 | [−9.30, −3.72] | 0.49 | [−2.41, 3.46] | 4.30 | [1.88, 6.93] | 1.60 | [0.13, 3.73] | 0.50 | [0.18, 0.60] |
| Fluid Comp.* | 5 | 15 | 0 | 5 | 0.4 | 3.9 | 0.7 | −5.08 | [−7.42, −2.57] | −2.51 | [−4.98, 0.29] | 7.15 | [4.67, 9.57] | 2.31 | [−0.05, 4.64] | 0.50 | [0.20,0.59] |
| PSM | 7 | 16 | 0.6 | 3.5 | 0.9 | 3.1 | 0.8 | −7.19 | [−10.00, −4.76] | 2.67 | [0.54, 4.21] | 2.13 | [−0.52, 4.29] | 1.79 | [0.65, 3.31] | 0.46 | [0.11, 0.56] |
| DCCS | 5 | 15.5 | 0 | 4.5 | 0.4 | 3.9 | 0.7 | −4.13 | [−6.00, −2.12] | −2.90 | [−4.94, −0.32] | 3.29 | [1.52, 5.01] | 2.28 | [−0.85, 4.11] | 0.45 | [0.12, 0.57] |
| PCPS | 5 | 15.5 | 0 | 4.5 | 0.4 | 3.9 | 0.7 | −6.86 | [−10.17, −3.18] | −5.54 | [−9.00, −1.84] | 7.42 | [4.06, 10.49] | 4.25 | [1.07, 7.46] | 0.47 | [0.15, 0.57] |
| Flanker ICA | 12 | 12.5 | 0 | 5 | 0.2 | 2.6 | 0.5 | 2.02 | [0.93, 3.02] | 0.19 | [−1.03, 1.52] | 3.21 | [1.80, 4.59] | 0.55 | [−0.16, 1.14] | 0.36 | [0.15, 0.42] |
| LSWM | 11 | 13 | 0.7 | 2.5 | 0.3 | 2.9 | 1 | −0.46 | [−2.25, 1.24] | 4.55 | [1.88,7.11] | 1.36 | [−1.21, 3.82] | 0.25 | [−1.58, 1.63] | 0.37 | [0.11, 0.46] |
Linear multiple regression was used to predict cognition from 4 sleep spindle features: amplitude, density, duration, and FFT (mean spindle frequency). Pearson’s correlation was then performed to compare measured cognitive scores with cognitive scores predicted by the optimized regression model. Sleep spindle features were generated using Luna.
comp, composite; fc, central frequency; DCCS, Dimensional Change Card Sort; ICA, Inhibitory Control & Attention; L, Lower; LSWM, List Sorting Working Memory; q, quality metric; PCPS, Pattern Comparison Processing Speed; PSM, Picture Sequence Memory; th, threshold; U, Upper.
*Top performing model from the final Luna run, which showed slight improvement in model performance.