2.3. Statistical Analyses

All analyses were conducted in the subset of per-protocol (PP) participants who did not resume cART until after week 40 (Vacc-4x n = 72; placebo n = 32) (Fig. 1, Fig. S1) (Pollard et al., 2014). This time point was at least 12 weeks after cART interruption; thus, participants had experienced peak viremia rebound by then (Kutzler and Jacobsen, 2008). The VL and CD4 count endpoints were log₁₀-transformed VL and CD4 count at weeks 44, 48, and 52 during ATI, with and without subtracting log₁₀-transformed pre-ART VL and CD4 count, respectively. VE on VL (VE^VL) and CD4 count (VE^CD4) were defined as the mean difference of VL and CD4 count endpoints, respectively, between the Vacc-4x and placebo groups. A positive (or negative) immune predictor indicates a positive (or negative) association between immune response levels and Vacc-4x benefit. Vacc-4x benefit is indicated by having a lower VL or a higher CD4 count in the Vacc-4x group compared with placebo.

CONSORT diagram showing the availability of immunological samples for the vaccine effect prediction analysis of the phase 2 Vacc-4x clinical study.

For each VL and CD4 count endpoint, univariate and multivariable linear regression models of observed complete-case data were used to identify potential predictors of VE. Missing data due to insufficient specimens were assumed to be missing completely at random. Wald tests were used to evaluate statistical interactions between each immune variable predictor and treatment assignment. All reported results are from multiple regression models adjusted for at least one of the following baseline confounding variables identified as predictors of the VL or CD4 count endpoint in the Vacc-4x or placebo group: preART VL or CD4, median-dichotomized LDH (for VL endpoints), continuous LDH (for preART-adjusted VL endpoints), median-dichotomized WBC (for preART-adjusted VL endpoints), anti-C5/gp41^732–744 antibodies, LDH or WBC (for CD4 endpoints), or dichotomized WBC or anti-C5/gp41^732–744 (for preART-adjusted CD4 endpoints). Sex at birth was also considered in the analyses of the Vacc-4x group or the combined Vacc-4x and placebo groups. However, due to the sparsity of female participants, especially in the placebo group, analyses did not adjust for sex at birth when models were unstable. When considering what baseline covariates might contribute to the analysis of immune responses as predictors of vaccine effect (our study objective), we first examined all available baseline demographic and clinical factors as potential predictors of VL and/or CD4 count during ATI within each treatment group as previously described (Huang et al., 2016a). Specifically, we considered gender (male or female), age (both continuous and dichotomized), country (Germany, Spain, US, or UK and Italy combined), time on cART (continuous and dichotomized), time since HIV diagnosis (continuous and dichotomized), CD4 nadir (continuous and dichotomized), preART VL (continuous and dichotomized), preART CD4 count (continuous and dichotomized), as well as host genetics class I HLA type. Among these baseline factors, together with safety factors considered in a separate analysis (Huang et al., 2016b), age, gender, preART VL, LDH and WBC were identified as significant predictors of VL and/or CD4 count during ATI, which were subsequently included in the analyses of vaccine effect predictors reported here. Other baseline factors were found to be only marginally or not predictive and hence were not included in further analysis.

Q-values were calculated to account for multiple comparisons within the analysis of each endpoint and predictor type (continuous or categorical) (Benjamini and Hochberg, 1995). Multiple testing adjustments for interaction tests were applied across immune variables that were identified as a significant predictor in either the individual or combined treatment groups. Predictors with associated unadjusted p-value < 0.05 and q-value < 0.1 were considered strong evidence of signals for hypothesis-generating purposes. Longitudinal data analyses of VL and CD4 count endpoints over time between week 36 and week 52 were also performed. Analyses were performed using R (version 3.2.1) (R Core Team, 2015).