Table 4.
Mixed-effects modeling: univariate and multivariate analysis of associations between changes in serum inflammatory markers and changes in severity of the 5 most-severe symptoms from day −18 pre-AuSCT to day +30 post-AuSCT (natural log scale)
| Univariate models (327 observations)a | Multivariate models (310 observations)b | ||||||
|---|---|---|---|---|---|---|---|
| Marker | Estimatec | SE | Pr > |t| | Marker | Estimate | SE | Pr > |t| |
| Pain | |||||||
| sTNF-RI | 0.447 | 0.174 | 0.011 | STNF-R1 | 0.488 | 0.167 | 0.004 |
| IL-10 | 0.182 | 0.054 | 0.0008 | IL-10 | 0.162 | 0.062 | 0.010 |
| IL-6 | 0.154 | 0.053 | 0.004 | ||||
| CRP | 0.117 | 0.049 | 0.018 | ||||
| Fatigue | |||||||
| SIL-1R2 | −0.190 | 0.051 | 0.003 | ||||
| SIL-6R | −0.893 | 0.198 | < 0.0001 | SIL-6R | −0.519 | 0.249 | 0.039 |
| IL-6 | 0.185 | 0.051 | 0.0003 | ||||
| IL-10 | 0.110 | 0.053 | 0.037 | ||||
| CRP | 0.131 | 0.048 | 0.007 | ||||
| Baseline mood score | 0.431 | 0.142 | 0.003 | ||||
| Disturbed sleep | |||||||
| MIP-1α | −0.387 | 0.104 | 0.0002 | MIP-1α | −0.387 | 0.104 | 0.0002 |
| Lack of appetite | |||||||
| CRP | 0.142 | 0.065 | 0.029 | ||||
| IL-6 | 0.166 | 0.069 | 0.017 | ||||
| SIL-6R | −0.533 | 0.261 | 0.042 | ||||
| Drowsiness | |||||||
| IL-6 | 0.170 | 0.054 | 0.002 | IL-6 | 0.147 | 0.056 | 0.009 |
| SIL-6R | −0.538 | 0.209 | 0.011 | ||||
| Baseline mood score | 0.269 | 0.116 | 0.021 | ||||
| Top 5 most-severe symptoms | |||||||
| IL-6 | 0.156 | 0.037 | < 0.0001 | IL-6 | 0.196 | 0.059 | 0.001 |
| MIP-1α | −0.155 | 0.063 | 0.014 | MIP-1α | −0.154 | 0.061 | 0.013 |
| SIL-6R | −0.359 | 0.145 | 0.014 | ||||
| SIL-1R2 | −0.101 | 0.037 | 0.007 | ||||
| IL-10 | 0.094 | 0.038 | 0.015 | ||||
| CRP | 0.087 | 0.035 | 0.012 | ||||
| Baseline mood score | 0.264 | 0.089 | 0.003 | ||||
Abbreviations: AuSCT, autologous hematopoietic stem cell transplant; SE, standard error of the mean.
For univariate models, each model presents the significance of the single inflammatory marker associated with the symptom outcome, as well as the estimate value, adjusted for quadratic time, age, gender, staging, BMI, baseline mood score, stem cell dose, and bortezomib-based induction regimen.
For multivariate models, each model presents the significance of inflammatory markers associated with the symptom outcome, as well as the estimate value, adjusted for other inflammatory markers from univariate analysis, quadratic time, age, gender, staging, BMI, baseline mood score, stem cell dose, and bortezomib-based induction regimen. The significant level of each model was set as 0.05/n (n, number of cytokines included in the model).
The estimate is a parameter estimated for a predictor. For natural log transformed independent variables, the estimate shows how the outcome changes when the predictor changes by any ratio. For example, the estimate of natural log IL-6 levels for the mean severity of the top 5 symptoms was 0.196, which meant that when the level of IL-6 doubled (or increased 100%), there was a corresponding 0.196*log2 = 0.14-unit increase in the mean symptom score.