Collaboration Of Controlled Randomised trials of Oral Antithrombotic agents after intraCranial Haemorrhage (COCROACH)

Lead Investigator: Rustam Al-Shahi Salman, University of Edinburgh
Title of Proposal Research: Collaboration Of Controlled Randomised trials of Oral Antithrombotic agents after intraCranial Haemorrhage (COCROACH)
Vivli Data Request: 7581
Funding Source: The COCROACH team are employed by the University of Edinburgh, which received a grant from a charity (the British Heart Foundation) to pay for the statistician’s time on the project.
Potential Conflicts of Interest: Professor Rustam Al-Shahi Salman) coordinates the Collaboration Of Controlled Randomised trials of Oral Antithrombotic agents after intraCranial Haemorrhage (COCROACH; https://www.ed.ac.uk/clinical-brain-sciences/research/so-start/for-collaborators) and is the senior author of the Cochrane Systematic Review of antithrombotic drugs after stroke due to intracerebral haemorrhage (https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD012144 .pub2/full/ta#) There are no financial or other competing interests with this application.

Summary of the Proposed Research:

Stroke due to bleeding in the brain, known as brain haemorrhage, affects at least 3 million people in the world each year. People who survive brain haemorrhage are at risk of suffering another “major vascular event”. These major vascular events include heart attacks, strokes, and death due to clotting or bleeding problems. The risk of having a major vascular event is up about 8% each year after brain haemorrhage. Randomised controlled trials found that people with an iregular heart beat, known as atrial fibrillation (or AF for short) benefit from blood-thinning anticoagulant drugs.

Why are we doing this research?
Doctors aren’t sure whether anticoagulant drugs can be used for people with AF who have had a brain haemorrhage. A few, small randomised controlled trials of anticoagulant drugs have been done in patients with brain haemorrhage and AF. These include our SoSTART trial (203 patients), the APACHE-AF trial (101 patients), and the NASPAF-ICH trial (30 patients). However, these trials were small and none reached a conclusive result.

What will we do?
We will perform two different types of analysis by combining data from all completed randomised controlled trials of oral anticoagulation for AF after brain haemorrhage to get the most precise and reliable answer. As well as the three trials mentioned above, we want to include data from 80 patients with brain haemorrhage and AF, who were randomised between Edoxaban and placebo in the ELDERCARE-AF trial. That’s why we are requesting data via Vivli from the ELDERCARE-AF trial.

What will this research mean?
These analyses will provide the most precise and reliable information about the effect of oral anticoagulation for AF after brain haemorrhage, which will inform the care of patients with this problem. We will publish the results in the Cochrane Library and in a peer-reviewed journal.

Statistical Analysis Plan:

1. Overall statistical principles

All analyses will be based on the intention to treat (ITT) principle with patients analysed according to allocated treatment, irrespective of whether they adhered to the allocated treatment, in the group to which they were allocated.
In general terms, categorical data will be presented using counts and percentages, whilst continuous variables will be presented – according to their distribution – using the mean, median, standard deviation (SD), minimum, maximum, inter quartile range (IQR) and number of patients with an observation (n).

Distributional assumptions underlying the statistical analyses will be assessed by visual inspection of residual plots. If the distributional assumptions for the parametric approach are not satisfied, further data transformation (to alleviate substantial skewness [i.e. log-transformation] or to stabilise the variance), or other suitable methods will be considered. This will be documented in the statistical results report together with the reasoning supporting the action taken, if applicable.

In general, a p-value of <0.05 will be considered as statistically significant, and will be presented alongside confidence intervals to facilitate interpretation.

All analyses and data manipulations will be carried out using either SAS software9 version 9.4 (or later) or Stata 15.10

Analysis populations: We will include adult participants with symptomatic spontaneous intracranial haemorrhage (intracerebral haeomrrhage [ICH], intraventricular haemorrhage [IVH], subarachnoid haemorrhage [SAH], or subdural haemorrhage [SDH]) who underwent treatment allocation at random in eligible randomised controlled trials (RCTs) and did not opt out of data sharing. Wherever possible, we will include all randomised participants, irrespective of treatment adherence, in line with the ITT principle.

2. List of analyses

Study characteristics
We will tabulate which trials are included, and characteristics of these studies: participants, interventions, comparators, outcomes, duration of follow-up, funding source, and risk of bias within each RCT. We will quantify the number of participants in each RCT, as a whole, and split by treatment, as well as the total number of participants across all trials (as a whole and split by treatment).

Individual patient data (IPD) integrity
We will report any issues identified in checking IPD, or state if there were none. We will check the range, completeness, and internal consistency of data items supplied by each RCT to help identify and rectify any major errors, inconsistencies or biases in the data, as well as promoting better understanding of individual RCTs. Any questions about the data supplied will be directed to the chief investigator(s) responsible for that RCT, in confidence. We will keep a detailed log of all changes and transformations made to participant data, to enable transparency and reproducibility. We will check that each RCT’s published baseline characteristics and aggregate results on at least the primary outcome (and any other outcomes that are included in this IPDMA) can be reproduced using the individual participant data supplied, with reasonable accuracy, allowing for any data that are missing because participants opted out of data sharing. If these checks do not identify any major concerns, we will include the RCT’s participant-level data in quantitative syntheses.

Baseline data
Before any analyses of outcomes are undertaken, descriptive analyses will be performed for all the baseline variables both to describe the study populations, combined and separately by treatment group allocation and study, and to ensure suitability for analysis. No formal statistical testing of differences between studies will be performed. Baseline data will include:
• confirmation of eligibility
• medical history
• demographic characteristics
• brain imaging characteristics
• concomitant medications.

Primary outcome
The primary outcome is a composite outcome of major vascular events that are appropriate to the intervention-comparator dyad, quantified over the entire period of follow-up. In this iteration of the COCROACH IPDMA restricted to the first three completed RCTs primarily assessing the effects of oral anticoagulant agents, major vascular events are defined as stroke or cardiovascular death:
• Symptomatic non-fatal stroke = ischaemic stroke, haemorrhagic stroke [i.e. intracerebral haemorrhage (ICH), intraventricular haemorrhage (IVH), or subarachnoid haemorrhage (SAH)), or unknown pathological sub-type and alive at outcome date + 30 days
• Cardiovascular death = death within 30 days of symptomatic stroke (as above) or other intracranial haemorrhage, extracranial haemorrhage, myocardial infarction, or systemic embolism (arterial or pulmonary embolism); sudden cardiac death; death from another vascular cause (and not within 30 days of an outcome event); death of an unknown cause.

Secondary outcomes
Secondary outcomes will include major ischaemic events (including the individual components separately), major haemorrhagic events (including the individual components separately), death, and the modified Rankin Scale, all quantified over the entire period of follow-up.

Major ischaemic events
• Ischaemic stroke
• Systemic arterial embolism
• Pulmonary embolism
• Myocardial infarction

Major haemorrhagic events
• Spontaneous intracranial haemorrhage (ICH, IVH, SDH, or SAH)
• Major extracranial haemorrhage (gastrointestinal or other)

Death of any cause
• All deaths quantified over the entire period of follow up.

Death/dependence on the modified Rankin Scale
This secondary outcome was collected in the contributing RCTs and has been added to capture the impact of fatal and non-fatal outcome events, but mRS was not mentioned in the COCROACH protocol (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42021246133%2520).

Outcome analysis

Survival analysis
For outcomes measures as time-to-event (all outcomes except modified Rankin), outcome survival times are defined as the time to first outcome event during follow-up from randomisation. Follow-up will be censored at death (unrelated to an outcome event) or last available follow-up. Natural frequencies of first outcome events in follow-up, fatal vs. non-fatal events, and their annual event rates will be quantified by event type and treatment allocation group, and absolute rate differences between groups estimated along with HRs.

Proportional odds model
For the secondary analysis of the modified Rankin Scale (an ordinal measure), we will use a proportional odds model (POM). If the assumption of proportional odds is rejected, a suitable alternative will be considered instead.

Synthesis methods
Handling of missing data
If fewer than 10% of participants have a missing value for any baseline variable required as part of the two-stage analysis, in the first instance, records for participants with such a missing value will be removed from any formal statistical analysis that involves the variable with the missing value (complete case analysis). If more than 10% of participants have such missing data, we will impute missing values using mean imputation prior to analysis. In tabulations, numbers of missing observations will be provided, but percentages will not include them.

Use of a one-stage or two-stage approach
We will pool RCTs to perform a two-stage meta-analysis to estimate treatment effects and to investigate heterogeneity.

Method used to generate effect estimates separately within each study and combined across studies
We will use the ipdmetan command in STATA to generate effect estimates separately within each study (stage 1) then combined across studies (stage 2).

Modelling
We will use a fixed effect inverse-variance model for the primary analysis, with a sensitivity analysis using a random effects model to check the robustness of our findings (irrespective of the degree of heterogeneity). A fixed effect meta-analysis is considered valid under the assumption that all effect estimates are estimating the same underlying intervention effect. For the random effects meta-analysis the DerSimonian and Laird method will be used. Firth’s method will be used if there are zero cell counts.
We will test the assumption of proportional hazards using a log-log plot stratified by allocated treatment and may also consider the Schoenfield residuals test (estat phtest). If there is strong evidence of violation of this assumption, we will consider a non-proportional hazards model instead. If there is no evidence of violation, then a Cox regression model will be constructed for a time-to-event primary outcome. The Cox regression for each trial will be adjusted for any covariates used in its randomisation algorithm in so far as this is possible. If, due to small numbers of events, a trial has adjusted for a subset of randomisation algorithm covariates in its primary analysis, we will use the same subset as the trial used. We will use a two-stage fixed effects model in Stata to generate a hazard ratio (HR) per study and a pooled estimate. The results will be presented in a forest plot displaying adjusted HRs with 95% confidence intervals for each study and the pooled estimate. The random effects model results will also be presented to compare with the findings of the principal analyses.

How (summary) survival curves will be generated
Survival times will be summarised descriptively for each treatment group separately in each RCT using Kaplan-Meier time-to-event curves. This will include numbers at risk, numbers with an outcome, and number censored in each interval. We will use the same axis range per study, with the maximum overall follow up.

Methods for quantifying statistical heterogeneity
Statistical heterogeneity manifests itself in the observed intervention effects being more different from each other than one would expect due to random error (chance) alone.11 For each of the primary and secondary outcomes, we will assess homogeneity between the treatment effects in different RCTs firstly by comparing the individual RCT confidence intervals. More formally, we will use Cochran’s Q test; the assumption of homogeneity will be rejected at p <0.10. We will also use the I2 statistic to estimate the proportion of total variation between RCT estimates that is due to heterogeneity rather than sampling error.11 A rough guide to I2 interpretation is as follows:
• 0% to 40%: might not be important
• 30% to 60%: may represent moderate heterogeneity*
• 50% to 90%: may represent substantial heterogeneity*
• 75% to 100%: considerable heterogeneity*

* The importance of the observed value of I2 depends on (1) magnitude and direction of effects, and (2) strength of evidence for heterogeneity (e.g. p-value from the Chi2 test, or a confidence interval for I2: uncertainty in the value of I2 is substantial when the number of studies is small).11 As the number of studies for this analysis is small, the latter is a particular consideration here.
If heterogeneity is identified, in the first instance the data will be checked to ensure they are correct. If heterogeneity is confirmed and there is considerable variation in results (and particularly if there is inconsistency in the direction of effect) it may be misleading to quote an average for the intervention effect. Any heterogeneity will be explored further using the subgroup analyses and the leave-one-out method.

Exploration of variation in effects in subgroup analyses

Providing there are sufficient data available, we will explore whether any observed treatment effect is consistent across well-defined participant subgroups. Analyses of the effects on the primary outcome in the subgroups were planned in the protocol as detailed in Section 2 of this statistical analysis plan (SAP) (see adjusted analysis covariates). In this iteration of the IPDMA, the following exploratory sub-group analyses of the primary outcome have been prioritised and will be explored depending on what is permitted by the available data:
• Age (<75 years vs. ≥75 years, also analysed as a continuous variable in a sensitivity analysis)
• Sex (male vs. female)
• Intracranial haemorrhage location (exclusively lobar intracerebral haemorrhage or SAH vs. other intracranial haemorrhage (non-lobar ICH, IVH, or SDH))
• Time between intracranial haemorrhage onset and randomisation (<8 weeks vs. ≥8 weeks)
• CHA2DS2-VASc score (<4 vs. ≥4).

We will explore the effects of the interventions in stratified analyses, examining statistical interactions between subgroups and the overall effect within each RCT (the “deft” approach) and then pool these interactions across RCTs, dependent on whether there are sufficient participants and outcomes per group. We will quantify absolute rate differences between strata in sub-groups within each trial only.

Risk of bias across studies
The Risk of Bias 2 (RoB2) tool will be used to assess bias per study in the following areas: (1) the randomisation process; (2) deviations from intended intervention; (3) missing outcome data; (4) measurement of the outcome and (5) selection of the reported result. The judgement for each domain is ‘low risk of bias’, ‘some concerns’, or ‘high risk of bias’. In addition, the same three judgement options are available for overall risk of bias. We will use the same tool to determine the final assessment of each RCT’s risk of bias, disregarding item (5) above, but also using information from RCT protocols and collaborators. We will also assess publication bias across RCTs by visual inspection for funnel plot asymmetry and with Egger’s regression test.

Requested Studies:

A Phase 3, Randomized, Double-blind, Placebo-controlled, Parallel-group, Multicenter Study of DU-176b in Patients With NVAF Aged 80 Years or Older Who Are Ineligible for Available Oral Anticoagulation Therapy
Data Contributor: Daiichi Sankyo, Inc
Study ID: NCT02801669
Sponsor ID: DU176b-C-J316

Public Disclosures:

Al-Shahi Salman R, Stephen J, Tierney JF, Lewis SC, Newby DE, Parry-Jones AR, White PM, Connolly SJ, Benavente OR, Dowlatshahi D, Cordonnier C, Viscoli CM, Sheth KN, Kamel H, Veltkamp R, Larsen KT, Hofmeijer J , Kerkhoff H, Schreuder FHBM, Shoamanesh A, Klijn CJM, Van der Worp HB. 2023. Effects of oral anticoagulation in people with atrial fibrillation after spontaneous intracranial haemorrhage (COCROACH): prospective, individual participant data meta-analysis of randomised trials. The Lancet Neurology. doi: 10.1016/S1474-4422(23)00315-0