Table 7. Results for ordinal regression (elastic net)–coefficients.
All available predictors (5 levels of proteins and 10 ratios between these proteins in serum together with 4 protein levels and 6 ratios between proteins in QFT supernatants) were used to model the distribution of an ordinal random variable, Y, which equalled 1 for HC, 2 for LTBI and 3 for ATB. The non-zero coefficients are regarded as being informative of the (conditional) distribution of Y.
| Parameters | Coefficients (95% CI) | |
|---|---|---|
| logit(P[Y = 1|Y> = 1]) | logit(P[Y = 2|Y> = 2]) | |
| Serum | ||
| (Intercept) | 5.007 (2.320; 5.241) | 3.745 (2.144; 3.779) |
| IL-18 | -0.115 (-0.232; 0) | -0.073 (-0.232; 0) |
| IL-18BP | -0.237 (-0.311; -0.082) | -0.237 (-0.317; -0.093) |
| IL-37 | 0.129 (0; 0.047) | 0.046 (0; 0.002) |
| IFN-γ | -0.848 (-0.633; -0,020) | -0.848 (-0.633; -0.020) |
| IP-10 | -0.113 (-0.149; -0.062) | -0.157 (-0.175; -0.079) |
| IL-18/IL-18BP | -0.545 (-0.811; 0) | -1.025 (-0.858; 0) |
| IL-18/IL-37 | 0.129 (0;0.015) | 0.129 (0; 0.015) |
| IL-18/IFN-γ | -0.051 (-0.12; 0) | 0.079 (-0.007; 0.056) |
| IL-18/IP-10 | 0 (0; 0.154) | 0 (0; 0.088) |
| IL-18BP/IL-37 | 0 (-0.019; 0) | -0.153 (-0.081; 0) |
| IL-18BP/IFN-γ | 0 (-0.155; 0) | 0 (-0.209; 0) |
| IL-18BP/IP-10 | 0.222 (0.150; 0.431) | -0.005 (0; 0.430) |
| IL-37/IFN-γ | 0 (0; 0.023) | 0 (-0.007; 0.021) |
| IL-37/IP-10 | 0 (-0.034; 0) | -0.200 (-0.063; 0) |
| IFN-γ/IP-10 | 1.300 (0; 0.851) | 2.427 (0; 1.589) |
| QFT supernatant | ||
| IL-18 | -0.371 (-0.419; -0.063) | 0 (-0.123; 0) |
| IL-18BP | -0.097 (-0.540; -0.217) | -0.028 (-0.125; 0) |
| IL-37 | -0.086 (-0.067; 0) | 0.011 (-0.032; 0.004) |
| IFN-γ | -0.981 (-0.478; -0.153) | -0.027 (-0.027; 0) |
| IL-18/IL-18BP | 0.433 (0; 0) | -0.106 (-0.252; 0) |
| IL-18/IL-37 | 0 (-0.028; 0) | 0 (-0.035; 0) |
| IL-18/IFN-γ | 0.301 (0; 0.602) | -0.363 (-0.418; -0.021) |
| IL-18BP /IL-37 | -0.543 (-0.514; -0.074) | 0.089 (-0.022; 0.232) |
| IL-18BP/IFN-γ | 0.266 (0.411; 1. 116) | -0.659 (-0.947; -0.247) |
| IL-37/IFN-γ | 0.047 (0.031; 0.211) | -0.009 (0; 0.049) |
The CI's were estimated via boostrap procedure based on 300 repetitions with the size of the subsample equal to 80% of the original sample. This approach is proposed due to lack of closed form formulas for the distributions of the coefficients. Due to differences in sample size, there are instances where the coefficient lies outside of the CI indicating lack of robustness.