Statistical Analysis Plan for the INTEnsive Care Bundle with Blood Pressure Reduction in Acute Cerebral Hemorrhage Trial: A Stepped-Wedge Cluster Randomized Controlled Trial

Randomized controlled trials (RCTs) are the gold standard methodology upon which evidence-based recommendations are made for decisions over therapeutic interventions and causal inference between an exposure and outcome of interest. However, RCTs are complex, demanding, and costly, and they require a long time to conduct and report [1]. These issues are particularly pertinent to the critical illness of intracerebral hemorrhage (ICH), where the relatively low rates, heterogeneous manifestations, coexisting multimorbidity, variable pathways of management, and high mortality and disability present important challenges to recruiting the large numbers of patients required for achieving statistical power to reliably assess treatments with invariably modest effects [2, 3]. Another drawback is selection bias from the inclusion of patients with a better prognosis than others, who receive an enhanced background level of care, which can raise concerns over the external validity of the results to the broader clinical population.

A key methodological aspect of RCTs is the random allocation of individuals to different interventions to ensure the groups generated are similar (i.e., balanced) in terms of known and unknown characteristics (i.e., prognostic and confounding variables). A robust randomization allows a reliable assessment of the treatment effect by comparing key (primary) and other important (secondary) outcome measures between the groups. In some circumstances, however, random allocation of individuals may not be suitable, as in the evaluation of complex interventions such as those involving organizational change, health systems, or models of care. In these situations, interventions naturally occur across groups of individuals, such as in medical clinics, health care facilities, or hospital wards. It can be particularly challenging, and sometimes scientifically unsound, to have interventions delivered across groups in one way to some individuals and in another way to others, as this can lead to the phenomenon of “contamination” and compromise the estimated treatment effect under investigation. Where interventions naturally occur in groups, it is better for the unit of random allocation to be the group or cluster, rather than the individual. These types of RCTs are called cluster RCTs.

An increasingly popular type of cluster RCT is the stepped-wedge cluster randomized trial (SW-CRT) design [4], in which clusters switch from control to intervention following a prespecified random sequence. SW-CRT are novel pragmatic designs, which are increasingly being used in the evaluation of service delivery types of interventions, combining the need for robust evaluations with logistical constraints. The SW-CRT design offers solutions to the logistical and ethical challenges of exposing investigators and participants to organizational change and provides efficiency gains in terms of statistical power over a parallel cluster RCT; since all the clusters eventually receive the intervention, there is better “buy-in” from investigators in allowing them all the opportunity to test the intervention.

The third INTEnsive care bundle with blood pressure Reduction in Acute Cerebral hemorrhage Trial (INTERACT3) aims to determine the effectiveness of a goal-directed care bundle of time-critical protocols involving early physiological control (intensive blood pressure [BP] lowering, glycemic control, and early treatment of pyrexia) and reversal of anticoagulation therapy, compared with usual the standard of care, on functional outcome in a broad range of patients with ICH [5].

The care bundle includes the modification of hypertension, hyperglycemia, and pyrexia, common abnormal physiological variables that occur early after the onset of ICH, which together with prior use of anticoagulation therapy, independently predict a poor outcome from this serious condition. Whilst guidelines are near consistent in recommending early, rapid, and smooth, BP lowering therapy after ICH, as a reasonable interpretation of the variable randomized evidence, uncertainty persists as to the optimal approach and size of the treatment effect, particularly in those patients with large ICH and raised intracranial pressure, or who undergo surgical evacuation, and have been purposefully been excluded from RCTs [6]. Limited hard outcome data exist for the benefits of glycemic control, use of antipyrexics, and rapid reversal of anticoagulation in ICH and are difficult to obtain through conventional individual patient RCTs. Combined, these variables have shown a strong association with poor outcomes after ICH [7], and another care bundle with intensive BP lowering has shown promise for improving clinical performance and outcomes in ICH [8].

Implementing the care-bundle intervention requires a switch from care as usual to a new model of care. Given that this needs to occur at the hospital level, INTERACT3 was designed as a SW-CRT with clusters (hospitals) randomized to different sequences (4 phases/3 steps) in the direction from all control (standard of care) to all intervention (care bundle). As soon as a site was due to commence the intervention phase in INTERACT3, the project team used a range of implementation methods to introduce the pre-defined goal-directed care bundle for application to consecutively eligible patients as a policy of usual care for the rapid correction (<1 h) of physiological variables as soon as any abnormality is recognized, and for this to be maintained for 7 days or hospital discharge (or death, if sooner). The interventions are: intensive BP lowering to a systolic target of <140 mm Hg; glucose control to a target of 6.1–7.8 mmol/L and 7.8–10.0 mmol/L for patients without and with diabetes mellitus, respectively; treatment of pyrexia by any means to a target body temperature ≤37.5°C; and reversal of anticoagulation to an international normalized ratio (INR) < 1.5, using vitamin K and prothrombin complex concentrate or fresh frozen plasma.

Evaluation of this multifaceted package of simple interventions as an effectiveness-implementation hybrid approach (i.e., a priori dual focus on assessing effectiveness and implementation) [9] has obvious advantages: (i) the agents for physiological control and reversal of anticoagulation are low cost, widely available, and scalable; (ii) there are defined treatment targets recommended in guidelines and hospital policies; (iii) potential adverse effects are well known, generally mild, and able to be rapidly corrected; (iv) as the intervention can be considered low-risk, it allowed use of a process of guardian consent to implement the care bundle to eligible patients as part of routine care; and (v) it provided an ability to rapidly include large numbers of patients with large ICH and/or require neurosurgical intervention, where there is limited RCT data on the effectiveness of intensive BP lowering and the other interventions under investigation, and who potentially have the most to gain from them.

Although conceptually simple, though, the SW-CRT design also presents several methodological issues [10, 11] that need to be considered in the analysis and interpretation of results. The main issue lies with the fact that SW-CRT can be confounded by time. This is because more intervention phases happen in the later part of the trial, making the effect of the intervention potentially confounded by secular trends, such as changes in background care. This issue is particularly relevant to INTERACT3 due to the COVID pandemic evolving during the roll-out of the intervention phases in China and activation of sites elsewhere in the world. COVID is well recognized to have impacted on the organization of health services and the availability of investigators for research. It is therefore plausible that it could have altered the characteristics of patients, and the timing and performance of the care bundle they received, in the hospitals that have participated in INTERACT3. To assess the potential impact of COVID on the INTERACT3 intervention, we plan to conduct a range of sensitivity analyses, including models with varying time trends per cluster, controlled multiple imputations, and adjustment for the baseline characteristics of patients.

We describe the statistical analysis plan (SAP) for INTERACT3, which has been completed prior to data lock (expected November 2022) and to which investigators will adhere to in analysis of the results (online suppl. material; see www.karger.com/doi/10.1159/000526384 for all online suppl. material). The SAP was approved and signed off by the Trial Steering Committee (TSC) in January 2022. Overall, 122 hospitals have participated in the study, based in Brazil, China, Chile, India, Mexico, Nigeria, Pakistan, Peru, Sri Lanka, and Vietnam, with over 7,000 adult (age ≥18 years) patients with acute ICH enrolled by December 31, 2021.

The study is being conducted in compliance with local and international regulatory and ethical requirements. In addition to gaining approvals from the ethics committee (or institutional review board) at each participating hospital prior any trial activities were undertaken, the Medicine and Biological Sciences Research Ethics Committee of the University of Leicester approved conduct of the study (Ethics Reference: 26596-tgr2-ls) on October 30, 2020 in relation to a requirement of the funding from the Medical Research Council (MRC) for the study to have oversight from an independent TSC.

The study uses a mixed consent process whereby cluster guardian consent is obtained from an appropriate executive officer in participating hospitals to allow patients with acute ICH to receive the randomized care bundle as “usual care” and for use of individual written informed consent for data on patients to be collected through in-person assessments and medical records and for personal information to be released for research purposes to allow centralized telephone follow-up at 6 months following admission.

Final patient follow-up occurred in June 2022. All analyses outlined in the SAP will occur in late 2022 to allow the main results to be announced in early 2023.

Critical appraisal of the results of INTERACT3 will follow standard principles and criteria as outlined in the updated CONSORT statement for reporting of SW-CRTs [12]. Special attention will be given to assessing the results in relation to the impact of COVID, data missingness, and variable cluster size and systems of care. We acknowledge the advantages offered by the SW-CRT design may be offset by inefficiencies that have arisen from the random assignment of clusters rather than individuals to treatments over a protracted period, and in relation to complexities in analyses from the conduct of the study during a once-in-a-century pandemic.

We sincerely thank the INTERACT3 investigators, coordinators, and project staff for their hard work and dedication to the study. We also thank the patients and their families for participating in this study, and the INTERACT3 Trial Steering Committee who commented on this report.

Ethics approval was obtained at each site before being activated. According to a funding request from the Medical Research Council (MRC), additional approval had been obtained from the Research Ethics Committee of the University of Leicester, UK. The study uses a mixed consent process, whereby cluster guardian consent is obtained from an appropriate executive officer in participating hospitals to allow patients with acute ICH to receive the randomized care bundle as “usual care” for use of individual written informed consent for data on patients to be collected through in-person assessments and medical records and for personal information to be released for research purposes to allow centralized telephone follow-up at 6 months following admission.

Dr. Lili Song and Prof. Craig Anderson report receiving grant funding from Takeda and Credit paid to their institution and receiving lecture fees and travel reimbursement from Takeda. Mr. Laurent Billot, Dr. Xin Hu, Dr. Lu Ma, Dr. Menglu Ouyang, Dr. Xiaoying Chen, and Prof. Chao You have no relevant disclosures.

This study is jointly funded by the Department for International Development, Global Challenges Research Fund, and the Medical Research Council (MRC) via a research grant (MR/T005009/1); Program Grant from the National Health and Medical Research Council (NHMRC) of Australia (APP1149987); West China Hospital Outstanding Discipline Development 1-3-5 Program (ZY2016102). Other funding is from Sichuan Credit Pharmaceutical Co., Ltd and Takeda (China) International Trading Co., Ltd.

Mr. Laurent Billot wrote the first draft of the SAP. Prof. Craig Anderson wrote the first draft of the cover manuscript. Dr. Lili Song, Dr. Xin Hu, Dr. Lu Ma, Dr. Menglu Ouyang, Dr. Xiaoying Chen, and Prof. Chao You provided critical revisions and approved the final version for submission.

Data sharing will be available from 12 months after publication of the main results. Investigators are to make a formal request for data sharing through the Global Research Committee of The George Institute, and access will be controlled by the principal investigators with the approval of the Trial Steering Committee. Online supplementary material: Statistical analysis plan.

INTERACT3 trial has been registered on ClinicalTrials.gov (NCT03209258) and Chinese Clinical Trial Registry (ChiCTR-IOC-17011787).