A statistical analysis plan for the Adjunctive Corticosteroids for Tuberculous meningitis in HIV-positive adults (ACT HIV) clinical trial

TBM is the most severe form of tuberculosis. Clinical trial data are required to provide an evidence base for adjunctive dexamethasone in HIV-positive individuals with TBM, and to guide clinical practice. This document details the planned analyses at 12 months post randomisation for the ACT HIV clinical trial (NCT03092817); ‘a randomised double-blind placebo-controlled trial of adjunctive dexamethasone for the treatment of HIV co-infected adults with tuberculous meningitis (TBM)’. The primary endpoint of the ACT HIV trial is death (from any cause) over the first 12 months after randomisation. This statistical analysis plan expands upon and updates the analysis plan outlined in the published study protocol.


Scope of document
This document details the planned analyses at 12 months post randomisation for the ACT HIV clinical trial (NCT03092817); 'a randomised double-blind placebo-controlled trial of adjunctive dexamethasone for the treatment of HIV co-infected adults with tuberculous meningitis (TBM)'. A further analysis will follow at 24 months post randomisation. This statistical analysis plan expands upon and updates the analysis plan outlined in the published study protocol 1 .

Background and rationale for study
The ACT HIV clinical trial is a randomised double-blind placebo-controlled trial of adjunctive dexamethasone for the treatment of tuberculous meningitis (TBM) in HIV-positive adults. TBM is the most severe form of tuberculosis, with mortality approaching 50% in people living with HIV 2-5 despite the best available appropriate anti-TB chemotherapy.
TBM develops when Mycobacterium tuberculosis, initially acquired via the respiratory route, disseminates via the blood to form a secondary focus in the brain. Rupture of this secondary focus into the subarachnoid space results in TBM, with the consequent host inflammatory response being important for bacterial killing, but also responsible for the pathological complications and often-fatal consequences of the infection.
Adjunctive anti-inflammatory corticosteroid therapy, which may control excessive host inflammation, has long been considered to have a role in TBM treatment 6  How dexamethasone might confer its clinical benefit on those with TBM remains uncertain. A recent study of cerebrospinal fluid (CSF) cytokines in TBM suggested that HIVpositive individuals have higher cytokine concentrations than HIV-negative individuals 7 . Whether adjunctive dexamethasone is more or less likely to be beneficial in HIV-positive individuals with TBM is not known. There are an estimated 100,000 cases of TBM globally each year 8 , many associated with HIV.
Clinical trial data are required to provide an evidence base for adjunctive dexamethasone in HIV-positive individuals with TBM, and to guide clinical practice.

Amendments from Version 1
The update clarifies that the per protocol population will not include individuals subsequently found to have not met all inclusion criteria, or to have met any exclusion criteria, at the time of enrolment. This had not been stated in the previous version.
Any further responses from the reviewers can be found at the end of the article

Primary endpoint
The primary endpoint of the ACT HIV trial is death (from any cause) over the first 12 months after randomisation.

Secondary endpoints
The secondary endpoints of the ACT HIV trial are as follows:

Individuals included in analysis Primary analysis
The primary analysis includes all randomised participants. As such, an intention-to-treat (ITT) analysis will be performed for the primary and secondary endpoints. Participants will remain included in this analysis even if no study drug was received after randomisation.

Per protocol analysis
The per protocol (PP) analysis includes all randomised patients with the exception of those subsequently found to have not met all inclusion criteria, or to have met any exclusion criteria, at the time of enrolment, those with a final diagnosis other than TBM (confirmed by microbiology, serology, or histopathology), and those who received less than 7 days of administration of the randomised study drug for reasons other than death, or less than 30 days of anti-tuberculosis drugs for any reason other than death. Trial inclusion and exclusion criteria are listed in the trial protocol 1 . A per protocol analysis will be performed for all primary and secondary endpoints.
Statistical software Data will be analysed using the program R 9 , using the most up to date version available at the time of final analysis. Analysis: Baseline characteristics will be summarised as median (lower and upper quartiles) for continuous data and frequency (percentage) for categorical data. The amount of missing data for each baseline characteristic will also be displayed. We will perform multiple imputation of a variable in case of >5% missing values, and it is assumed that data are missing at random. Otherwise, a complete case analysis will be performed.

Baseline characteristics
Use of the uniform case definition diagnostic score The published TBM diagnostic score 10 will be used and subjects will be categorised as 'definite', 'probable', 'possible', or 'not TBM'. Participants will only be categorised as 'not TBM' if they have a confirmed alternative diagnosis (alternative to TBM) or they recovered without TB drugs.

Primary endpoint
The primary endpoint of this trial is death over the first 12 months after randomisation. The primary analysis is a Cox proportional hazards regression model with treatment as the only covariate. Since TBM MRC severity grade at enrolment (I, II, or III) is an important risk factor for mortality and was used as stratification variable in the randomisation, we will additionally perform a secondary analysis with MRC grade added as stratum variable. The proportional hazards assumption will be formally tested based on scaled Schoenfeld residuals and visually assessed by a plot of the scaled Schoenfeld residuals versus transformed time. Deaths, by study drug arm, and the results of the analyses will be shown as per Table 3 and  Table 4.
Survivors known to be alive at 12 months will be censored at that time-point and subjects who withdrew or were lost to  follow-up before 12 months will be censored at the date they were last known to be alive. Subjects who withdrew or were lost to follow-up before 12 months are estimated to be less than 5% of enrolled participants.
The homogeneity of the treatment effect on overall survival across subgroups will be assessed by formal tests of interaction between treatment and the following pre-defined grouping variables: TBM MRC severity grade at enrolment (I, II, or III), diagnostic category (definite, probable, possible), LTA4H genotype (CC/CT/TT), drug resistance pattern (MDR TB or rifampicin mono-resistant TB, isoniazid resistant non-MDR, no or other resistance), ART status at enrolment (ART naïve, ≤3 months of ART, >3 months of ART), and CD4 cell count at enrolment (<50, 51-100, 101-200, >200). We will estimate and report the treatment effect by fitting a separate Cox model per subgroup. We will additionally fit a Cox model that includes CD4 count at enrollment as a continuous variable via a restricted cubic spline with four knots and an interaction term with treatment arm. X.XX

X.XX
The primary endpoint will be overall survival, i.e., time from randomisation to death, over the first 12 months of follow-up. This table reports the results from the Cox proportional hazards regression model. The primary effect measure will be the resulting hazard ratio comparing dexamethasone vs. placebo with a corresponding two-sided 95% confidence interval. In "All patients", treatment will be the only covariate. We will additionally report the hazard ratio if the modified MRC grade is included as stratum variable. The test for proportional hazards will use the Kaplan-Meier as time transformation.
MRC=Modified Research Council. ART status of a patient will be unknown if i) they are on ART treatment and the start ART date is missing completely; or ii) their ART status is unknown. ART status of a patient will be undetermined if start ART date is present but distinction between ≤3 months of ART, and >3 months of ART, cannot be made due to limited date information. Kaplan-Meier plots and explicit survival estimates at 3, 6, and 12 months of follow-up will be calculated for the full sample and in the subgroups defined above. We will compare the treatment arms via the absolute risk of death at 12 months (using a Wald-type test based on Kaplan-Meier estimates at 12 months and associated standard errors using Greenwood's formula) and the restricted mean survival until 12 months (using the survRM2 12 package in R).

Secondary endpoints
Secondary outcomes 2 to 7 below are time-to-event outcomes. For each, we will compute and plot the Kaplan-Meier estimates (outcomes 2, 4) or the competing risks Aalen-Johansen estimates (outcomes 3, 5, 6 and 7) and report the values of the estimates at 3, 6 and 12 months. We will also fit a Cox proportional hazards model (results are interpreted as relative cause-specific hazards in the presence of competing risks). Analyses will Table 4. Hazard ratios for death (from any cause) by sub-group over the first 12 months after randomisation. XX *Heterogeneity will be tested with a Cox regression model that includes an interaction between treatment effect and subgroup. # Results will be given for sub-group with positive mycobacterial culture on baseline CSF. Pre-XDR and XDR are defined following World Health Organisation definitions 11 .

No. of deaths
The primary endpoint will be overall survival, i.e., time from randomisation to death, over the first 12 months of follow-up. This table reports the results from the Cox proportional hazards regression model. The primary effect measure will be the resulting hazard ratio comparing dexamethasone vs. placebo with a corresponding two-sided 95% confidence interval.
In these subgroup analyses, a separate Cox model will be fitted for each value of the subgroup. The "Test for heterogeneity" will be based on the likelihood ratio test that includes subgroup as covariate and compares the models with subgroup as main effect only and with subgroup as treatment effect modifier, with TBM MRC severity grade at enrolment (I, II, or III) as covariates. be performed for the full samples (ITT and PP). Subgroup analyses are specified below per secondary endpoint. For the subgroup analyses, we follow the same procedure as for the primary outcome: we fit separate models per subgroup and we fit a model in which we test for interaction by subgroup.
1. Neurological disability at 12 months from randomisation Neurological disability will be assessed by the modified Rankin score at months 1, 2, 3, 6, 9, and 12 from randomisation ( Table 5). The main secondary endpoint is the 12-month assessment and subjects who died before 12 months will be treated as having a score of 6 ('dead'). Rankin score assessments will be included from clinical assessments performed monthly (+/-7 days), with the exception of the 12-month clinical assessment for which an acceptable range of -10 days/+1 month will be applied. All clinical assessment timings are based on days from randomisation (i.e., 'day 0' is labeled as the first day study drug is received, with study drug received immediately after randomisation).
Neurological disability (as assessed by the ordinal modified Rankin scale) at 12 months will be compared between the two arms with a proportional odds logistic regression model with the treatment assignment as the main covariate and adjustment for TBM MRC severity grade. The result will be summarised as a cumulative odds ratio with corresponding 95% confidence interval and p-value. Individuals who withdrew or were lost to follow-up before 12 months are excluded.

First new neurological event or death over the first 12 months after randomisation
A new neurological event is defined as a fall in GCS by ≥2 points for ≥2 days from the highest previously recorded GCS (including baseline) or the onset of any of the following clinical adverse events: cerebellar symptoms, focal neurological signs, or onset of seizures. A description of all the events will be given, summarising how they met the criteria.
Analyses will additionally be performed in the subgroups defined by TBM MRC severity grade, diagnostic category, ART status at enrolment, time from randomisation to start of ART, and baseline CD4 count.

First IRIS event over the first 6 months after randomisation
The criteria for neurological IRIS diagnosis is defined in the study protocol 1 . The neurological IRIS rate will be defined as the number of IRIS events divided by the observed person-time of follow-up in each treatment group. A description of all the IRIS events will be given, describing how they met the IRIS diagnostic criteria.
The number of IRIS events in each group will be summarised. Death will be interpreted as a competing risk. Analyses will additionally be performed in the subgroups defined by TBM MRC grade, diagnostic category, ART status at enrolment, time from randomisation to start of ART, and baseline CD4 count.
4. AIDS-defining event or death over the first 12 months after randomisation Acquired immunodeficiency syndrome (AIDS)-defining illnesses will be defined as per the World Health Organisation (WHO) classification. WHO define AIDS as a clinical diagnosis (presumptive or definitive) of any stage 4 HIV condition 13 .
5. HIV-associated malignancy over the first 12 months after randomisation HIV-associated malignancy is defined as new diagnosis of one or more of the three major HIV-associated malignancies; Kaposi sarcoma, high grade B-cell non-Hodgkin lymphoma or invasive cervical cancer.
The number of HIV-associated malignancy events in each group will be summarised. Death will be interpreted as a competing risk. Analyses will additionally be performed in the subgroups defined by ART status at enrolment.

Use of open-label corticosteroid treatment for any reason, and at any time, after randomisation
The number of open-label corticosteroid treatment events in each group will be summarised. The reasons for corticosteroid treatment will be listed. Death will be interpreted as a competing risk.

Requirement for shunt surgery by 12 months
The number of shunts performed in each group will be summarised. Death will be interpreted as a competing risk.

Adverse events until 12 months
Serious adverse events (SAE) are defined in the study protocol 1 . SAE will be sub-grouped into categories. SAE will be grouped and graded as per Common Terminology Criteria for Adverse Events (CTCAE) 14 .
The number of patients with any serious adverse event will be summarised and compared between the two treatment arms based on the chi-squared test, or Fisher's exact test in case the expected count under the null hypothesis in at least one of the cells is smaller than one 15 . Specific adverse events will be summarised, but not formally compared. The total number of serious adverse event episodes per patient will also be summarised and informally compared based on a quasi-Poisson regression model with treatment as the only covariate.
The following subgroups of adverse events will also be separately summarised: clinical grade 3&4 adverse events; serious adverse events possibly, probably, or definitely related to the study drug; adverse events leading to TB treatment or ARV interruptions. Grade 3&4 laboratory abnormalities will be summarised in the same way as clinical adverse events. Adverse events will be shown as per Table 6-Table10.
Explanatory analysis via multivariate model A multivariate model of variables independently associated with death will include: age, enrolment GCS, TBM diagnostic category (definite, probable, possible), study drug allocation    Symptoms that lead to some restriction in lifestyle, but do not interfere with the patients ability to look after themselves 3 Symptoms that restrict lifestyle and prevent totally independent living 4 Symptoms that clearly prevent independent living, although the patient does not need constant care and attention.

5
Totally dependent, requiring constant help day and night.
6 Death (dexamethasone vs. placebo), CD4 count, CSF total leucocytes, CSF total neutrophils, CSF blood:glucose ratio, CSF lactate. Numeric variables will be analysed using restricted cubic splines. We will not include any interactions.

Data availability
No data is associated with this article.

Acknowledgments
The  N.ae = the total number of episodes of that particular adverse event The ' clinical grade 3&4 adverse events of all types' grouping will include only the group of adverse events that cannot be separated into grades 3 or 4 (recorded as 'grade 3 or 4' at the beginning of the trial). These events will not also be represented in the above 'grade 3' and 'grade 4' rows.