This SAP corresponds to the version v1.1 31th October 2022 according to the EMPIRICAL study protocol v2.0 10-11-2020. The first version of the SAP (v1.0) was released on 26th November 2020 and updated to v1.1 on 31th October 2022 after the interim analysis report dated on 11th October 2022. The changes on the original version were made following independent DSMB members suggestions to facilitate results visualizations including summary tables for factorial analysis and reporting number of events in figures.
The SAP will ensure that the analysis is not data driven or selectively reported. The results of the primary analysis are expected to occur in July 2025, after all participants enrolled have completed the 360 days of follow-up.
The EMPIRICAL trial is a 2x2 factorial, superiority, unlabeled randomized controlled trial. The 1:1:1:1 randomization was stratified using block sizes by center and severity. A factorial clinical trial was proposed for this clinical trial to examine two or more different interventions in the same trial. Each intervention has a different mechanism of impact on the primary endpoint. In this case, we considered it to be more efficient to run one large factorial trial addressing two questions rather than two separate trials addressing a single trial/question individually. Furthermore, if an unanticipated interaction between the interventions exists, a factorial design allows such interactions to be identified, and the relative contribution of each can be explored (12). The 2x2 factorial design allows two primary comparisons: (i) the effect of valganciclovir on mortality and (ii) the effect of TB-Treatment on mortality.
The study is a phase II-III study as follows:
- The empirical treatment for CMV (valganciclovir) is phase II according to The United States Food and Drug Administration (FDA) definitions, as the purpose is to investigate the efficacy and side effects of this treatment in up to several hundred people with the disease/condition (presumed CMV pneumonia), lasting several months to two years.
- The empirical treatment for TB (isoniazid, rifampicin, pyrazinamide and ethambutol) is phase III according to FDA definitions since the trial tries to demonstrate whether a product offers a treatment benefit to a specific population (in this case, HIV-infected infants with unknown-onset severe pneumonia). Phase III studies typically involve 300 to 3,000 participants.
The inclusion and exclusion criteria as well as the endpoints of the study and clinical definitions were described in the protocol (11).
The principal analysis will be based on intention-to-treat (ITT). We will analyze patients in the groups they were randomized to, regardless of treatment received after randomization. A sensitivity analysis will be performed, including a per-protocol analysis considering only data of participants who follow the randomization protocol correctly and excluding those with randomization errors (participants received an IMP different than the one in their allocated arm).
In the case of a new TB diagnosis after randomization, participants will be treated with TB-T according to SoC practices and local protocols. These participants will be considered in their allocation arms in both ITT analysis and per-protocol analysis because their follow their intervention allocation (IMP+SoC or SoC).
Primary objectives
To compare the impact on 15-day and 1-year mortality of combined systematic empirical treatment against TB and CMV plus SoC versus SoC in HIV-infected infants with severe pneumonia.
Secondary objectives
Secondary objectives are detailed in the EMPIRICAL trial protocol (11).
Hypothesis
Empirical treatment against CMV with oral valganciclovir and empirical TB-Treatment together with standard pneumonia treatment improve survival in HIV-infected infants with severe pneumonia, with a low-risk/benefit profile.
Primary analysis
To describe the participants included in the study, a flow diagram was constructed according to the Consolidated Standards of Reporting Trials Statement (13). This flow diagram includes the numbers of participants screened, the numbers eligible, the reasons for ineligibility, and the number of patients randomized and analyzed for the primary outcome, as shown in Figure 1.
To describe the clinical trial population, the data will be displayed as a whole and summarized by each randomization arm. Baseline demographic characteristics will be included to summarize all subjects. In summary, tables of continuous variables, medians and interquartile ranges will be assessed for nonparametric variables, and means and standard deviations will be presented for parametric variables. The Shapiro‒Wilk test will be performed to test normality. In summary, tables of categorical variables, counts and percentages will be used. The denominator for each percentage will be the number of subjects within the population group without considering missing observations unless otherwise specified using the compareGroups R package (14). Table 1 shows the demographic information by the randomization arm. Table 2 shows the characteristics of the participants who presented the primary endpoint (mortality) during the study.
To analyze the probability of survival among treatment arms, a factorial design was used. Two comparisons will be conducted: (i) participants who were allocated to arms with TB-Treatment (TB-Treatment + SoC; and TB-Treatment + Valganciclovir + SoC) compared to those who were not allocated to arms with TB treatment (Valganciclovir + SoC; and only SoC); and (ii) participants who were allocated to arms with Valganciclovir (Valganciclovir + SoC and TB-Treatment; + Valganciclovir + SoC) compared to those who were not allocated to arms with valganciclovir (TB-Treatment + SoC; and only SoC). An interaction between the two main treatments is expected to be synergistic because each treatment focuses on different targets of possible causes of mortality, and due to deleterious effect of CMV in the immune response to TB. Rather than the common assumption in factorial trials of no interaction between the two comparisons, we estimated a better performance in patients who received both treatments. For that reason, both at-the-margins and inside-the-table will be presented as recommended in factorial designs (12), Table 3 and Table 4.
The Kaplan-Meier survival curves will be plotted for the 4 arms separately, as presented in Figure 2. A proportional Cox regression model will be fitted for the primary outcome (mortality). A multivariate regression analysis will be performed, including CD4 T-cell count, HIV viral load, age at HIV diagnosis, oxygen support use, and nutrition status as possible covariates. The HIV viral load and CD4 percentage will be included in the model as time-dependent covariates using the survival R package (15). The best final model will be selected using recursive feature elimination via random forest and 10-fold cross-validation with 5 repeats implemented in the caret R package (16). Univariable and multivariable hazard ratio estimates and 95% confidence intervals will be presented in a forest plot, as shown in Figure 3. An initial regression model will be fitted, including the interaction and the associated p value. If there is no evidence of a statistically significant interaction term (i.e., p ≥ 0.05), a factorial analysis will be used to determine the success of the trial. If the interaction term is statistically significant, the effect of each of the four interventions will be tested.
To analyze snapshot clinically relevant timepoints in the trial (15-day mortality and 1-year mortality), univariable and multivariable logistic regressions will be performed.
To describe the rate, the number of incident deaths (number of deaths) will be calculated in the cohort, and the total follow-up time (person-time) will be computed; death-free individuals in the cohort will be observed over the study period. The estimated mortality incidence will be obtained by dividing the number of deaths by the total duration of follow-up (person-year).
Secondary analysis
If there is no evidence of a statistically significant interaction effect for the primary outcome, then no interaction effect will be assumed for the secondary outcomes. Likewise, if evidence of a statistically significant interaction effect is identified for the primary outcome, then the secondary outcomes will be analyzed assuming an interaction effect is present.
To compare the cumulative days on oxygen therapy from randomization until discharge, the probability of recovery after oxygen treatment will be estimated using the Kaplan‒Meier estimator. A log rank test will be performed to test for differences between randomization arms. Similar analyses will be performed to test the differences in cumulative days on oxygen therapy 1 year after randomization.
Serious Adverse Events (SAEs) will be described among the different randomization arms in Table 5. The type of registrable SAE, action required and resolution times will be compared. The chi-square test and Fisher’s test will be performed for categorical variables. For normally distributed continuous variables, Student’s t test will be performed, and the U-Mann‒Whitney or Kruskal‒Wallis test will be used when nonparametric. All hypothesis testing will be carried out at the 5% significance level and p-values will be rounded to three decimal places. In summary tables, p-values less than 0.001 will be reported as <0.001 according to the compareGroups R package (14). To analyze the association between SAEs and treatment arms the person-month density incidence will be calculated using a Poisson regression model. Overdispersion will be checked and if overfitting of zeros in the model, a zero-inflated model will be performed. SAEs incidence rate ratios together with 95% CI will be presented. Volcano plots will be used to represent risk differences and significance for the most frequent and significant diseases using the ggplot2 R package (17), as shown in Figure 4.
The prevalence of CMV infection will be calculated taking into account 57.1 copies/mL as the cutoff for positivity. CMV-attributable pneumonia will be considered in cases with CMV viral load >4.1 log copies/mL (18). CMV viral decay will be analyzed using paired Wilcoxon signed-rank tests.The prevalence and incidence of TB will also be analyzed according to the protocol definitions. The diagnostic accuracy of the TB tests will be assessed using confusion matrices, and the reporting accuracy, sensitivity, specificity and positive and negative predictive values will be assessed using the caret R package (16). Results derived from theXpert Ultra test will be considered as the gold standard.
Sensitivity analysis
Sensitivity analysis will be conducted to assess the robustness of the primary trial results by repeating the primary Cox regression analysis while taking into account the out-of-randomization scheme, if any. The as-treated analysis will be performed considering patients who were prescribed a treatment different than it was allocated to. A second sensitivity analysis will be conducted for the ITT population to test missing-at-random assumptions if more than 20% of the data included in the primary endpoint model are missing.
Handling of missing data
To avoid loss of information and statistical power in the association analysis, missing data will be imputed using joint multivariate normal distribution multiple imputation implemented in mice R package (19). To prevent many assumptions, only variables with less than 20% missing information will be considered for imputation. To obtain a better understanding of the way missing data are distributed among variables in the study, correlation matrices, patching patterns and box plot analyses will be performed by means of several functions implemented in the MICE and VIM R packages (20).
Protocol deviations
Summary of serious protocol deviations will be reported in the biannually DSMB reports and final results of the study.
Interim analysis
An interim analysis was planned when 50% of the sample size or 50% of the recruitment time elapsed whatever happen earlier. The Symmetric O’Brien Fleming stopping boundary was applied to the primary endpoint analysis for harm and efficacy.
Final analysis
Final analysis is planned to be conducted on July 2025 after database lock on 31st January 2025.
Sample size
The target sample size is 624 randomized participants (156 per arm). Sample size calculations were calculated to reach the main target points of the trial, the short-term (15-day) mortality reduction due to valganciclovir treatment and the long-term (1-year) mortality reduction due to TB-Treatment. The clinical prior assumptions were to observe a reduction from 35% baseline short-term mortality in the SoC to 23% in the valganciclovir-treated patients and a reduction from 41.5% baseline long-term mortality in the SoC to 28.8% in the TB-treated patients, as detailed in the EMPIRICAL trial protocol (21). The sample size was estimated using the WebPower R package (22) based on 80% statistical power and a 5% two-sided statistical significance level.
Statistical packages
All analyses will be carried out using R statistical programming software (23).
Data Management Plan
The study data will be managed according to the Data Management Plan (DMP) of EMPIRICAL study. The study data collection will be recorded and managed using the electronic data capture software of REDCap (24) version 8.4.4. Data will be reviewed on regular basis using standardized data quality controls in an effort to increase the cleanliness of future data stored. The quality controls will be conducted for the first five patients on each site and afterward, every six months corresponding to each DSMB reporting period. An R code pipeline will be run checking for missing information, clinical inconsistences, safety inconsistences, ranges checks, and dates inconsistences. Queries will be orderly reported in Data Validation Sheets and send it to each hospital/recruiting site responsible for resolving them in REDCap electronic Case Reports Forms.