| | Basic model | Spline model | ΔAIC favors spline |
|---|
Biomarker | β | P | P (FDR) | AICbasic
| β1 | P1 | P1 (FDR) | β2 | P2 | P2 (FDR) | AICspline
|
|---|
PlGF | −0.0488 | 0.0091 | 0.0320 | 112.0 |
−0.13
|
0.0002
|
0.0015
|
0.149
|
0.0039
|
0.0355
|
109.7
| True |
Log(MCP1) |
0.185
|
0.0001
|
0.0009
|
142.1
| 0.18 | 0.0029 | 0.0125 | 0.00889 | 0.9090 | 0.9270 | 147.4 | False |
Log(MIP1) | 0.151 | 0.0095 | 0.0320 | 176.2 |
0.488
|
< 0.0001
|
< 0.0001
|
−0.652
|
< 0.0001
|
0.0002
|
160.1
| True |
IL15 |
−0.0534
|
0.0002
|
0.0015
|
95.2
| −0.0875 | 0.0017 | 0.0092 | 0.0641 | 0.1330 | 0.2565 | 99.5 | False |
IL-7 |
−0.062
|
0.0007
|
0.0039
|
100.8
| −0.0899 | 0.0032 | 0.0125 | 0.055 | 0.2300 | 0.3653 | 105.7 | False |
VEGF-A | 0.146 | 0.0045 | 0.0203 | 157.9 | −0.0588 | 0.3110 | 0.3817 |
0.415
|
< 0.0001
|
< 0.0001
|
143.2
| True |
Log(IL-6) |
0.285
|
< 0.0001
|
0.0005
|
117.8
| 0.265 | 0.0003 | 0.0022 | 0.0418 | 0.4850 | 0.6236 | 123.1 | False |
Log(IL-8) |
0.362
|
< 0.0001
|
< 0.0001
|
76.3
| 0.312 | 0.0000 | 0.0000 | 0.102 | 0.0254 | 0.0686 | 77.9 | False |
- Data is from linear mixed effects models for the eight biomarkers that changed over time, after correction for multiple comparisons (see Additional file 1: Table S1 for data on all biomarkers). Biomarkers were used as dependent variables (scaled and standardized to z-scores) and time (hours) was used as predictor. For each biomarker, we tested two models, with or without restricted cubic splines (using three knots) to model time. Without splines, time is modeled with one parameter (β), and with splines, times is modeled with two parameters (β1 and β2). For each biomarker, we calculated the Akaike information criterion (AIC) for the two models. AIC may be used to compare model fits, where a lower AIC is preferable and penalizes models with additional predictors (and thereby protects against overfitting). For biomarkers with AICbasic-AICspline < 2, we selected the basic model; otherwise we selected the spline model (selected model indicated with green shading). Data where p values are significant after correction for multiple comparisons [P (FDR)] are shown in italics. For example, for MIP1, the AIC selected the spline model, and both β1 (the linear component) and β2 (the cubic component) were significant, suggesting that MIP1 increased significantly during the first part of the study, and then decreased significantly during the second part of the study. In contrast, for IL-8, the AIC selected the non-spline model, and β was significant, suggesting that IL-8 increased continuously during the entire study duration. See Fig. 1 for visualizations of the significant effects
