Coefficients (Coeff) are additive terms in the model for the natural logarithm of T-cell levels, fit by gee.⁴²  The model for the expected log measurement is of the form L = a + b₁x_{1 + … +}b₆x₆, where x₁, x₂, and x₃, each take the value 1 or 0 indicating whether or not the observation has the respective combination of donor status and reactivation status listed in the first three lines of the table; where x₄ is the day posttransplantation, x₅ is the propensity score, adjusting for baseline covariates, and x₆ is 1 if acute GVHD has occurred and 0 otherwise. The estimated coefficients b₁…b₆ are shown in the third column. The first three coefficients (D⁻/R⁺, Pre-reactivation; D⁺R⁺, Pre-reactivation; D⁻R⁺, Post-reactivation) are comparisons to the 225 D⁺/R⁺ post-reactivation measurements of IFN-γ–CD8, which were taken as the reference group (see Figure 2). The coefficient “Acute GVHD, grade > 1” is the adjusted average difference of those with and without acute GVHD. The coefficients “DPT” and “Propensity Score” multiply those variables in the model, and the intercept serves to calibrate the overall predicted level of IFN-γ–CD8 T cells. Effect ratios are the antilog of the estimated coefficients (eg, the antilog of b₁ is 0.13, meaning that “D⁻R⁺, Pre-reactivation” samples averaged 13% of the mean for the reference group). Robust z score is the coefficient estimate divided by its robust SE. An absolute (positive or negative) z score greater than 2 is statistically significant. The P value is obtained by referring the robust z score to a normal distribution and multiplying the tail area by 2, yielding the 2-sided Wald-test significance probability.

DPT indicates days posttransplantation.