Statistical analysis plan for the randomized controlled trial Tenecteplase in Wake-up Ischaemic Stroke Trial (TWIST)

The primary aim of TWIST is to test the hypothesis that treatment with IV tenecteplase within 4.5 h of wakening is superior to standard care (no IV thrombolysis) and improves functional outcome at 3 months in patients with wake-up ischemic stroke selected by NCCT brain imaging.

Patients aged 18 years and older with new stroke symptoms acquired during sleep and a clinical diagnosis of stroke with limb weakness and National Institute of Health Stroke Scale (NIHSS) score ≥3, or aphasia, are eligible for inclusion. Detailed inclusion and exclusion criteria have been published previously [4] and are listed in the supplementary material. In brief, the main exclusion criteria are NIHSS score >25 or NIHSS consciousness score >2, infarct size larger than >1/3 of the middle cerebral artery territory, or intracranial hemorrhage on NCCT.

A remote, web-based, computer-generated randomization procedure is used. All online submissions are secured by the use of password site entry and data encryption procedures via the TWIST online webpage. Investigators record patient details via a secure web interface before randomization takes place. The randomization procedure includes a standard minimization algorithm which ensures that the treatment groups are balanced across all centers in all countries for key prognostic factors: age, NIHSS score, and time since wake-up. Patients are allocated with a probability of 0.80 to the treatment group which minimizes the difference between the groups with regards to the key prognostic factors. Patients will be randomly allocated to open-label IV tenecteplase plus best standard care or to best standard care alone (i.e., no IV thrombolysis).

Details on the background and study design are documented in the published trial protocol [4]. Both patients and investigators are unblinded to treatment allocation. Primary outcome assessments are performed by telephone interview at 3 months (90±7 days) by trained study personnel, blinded to treatment allocation. The results of 3-month follow-up interviews are securely stored and kept separated from the investigators.

Only the Data Monitoring Committee (DMC) has access to interim data and results. An independent statistician, who otherwise is not involved in the trial, prepares biannual unblinded reports of the interim data and results to the DMC. The DMC will work on the principle that a difference of at least 3 standard errors in an interim analysis of a major outcome event may be needed to justify halting, or modifying the study before the end of planned recruitment (Haybittle-Peto rule) [6]. This criterion has the practical advantage that the exact number of interim analyses would be of little importance, and so no fixed schedule is proposed. A less-strict rule might be used if there was evidence that tenecteplase was unsafe. If the DMC judges credible evidence of harm, or overwhelming evidence of efficacy, the committee will advise the chairman of the Steering Committee. Unless this happens, the Steering Committee will remain blinded to the interim results.

The primary outcome is functional outcome at 3 months on an ordinal scale (0–6), defined by the modified Rankin Scale (mRS) score. The use of a categorical shift in mRS at 3 months after stroke has been advocated to provide a more comprehensive index of the clinical impact of acute stroke treatment [7]. This accounts for the fact that thrombolytic treatment is often not completely curative, but has the potential to improve patient outcome over the whole range of functional measurements. Table 1 outlines all relevant outcomes for the primary report.

Additional clinical, radiological, and health care system-related outcomes are included in the supplemental material. Details about definition and adjudication of clinical events and expert imaging readings are also listed in the supplementary material and Additional file 1.

The primary endpoint in TWIST is mRS across the full ordinal scale (shift analysis). We originally based our sample size estimation on the results of a Cochrane systematic review of the effect of rt-PA within 4.5 h of stroke onset, assessed as a binary endpoint (favorable outcome mRS 0–2 versus mRS 3-6) [11]. As a sample size estimation based on ordinal logistic regression analysis is more appropriate, we revised the sample size estimation in 2020. The revised sample size estimation was based on observations from recent studies of thrombolytic treatment in patients with wake-up stroke. In the largest randomized controlled trial of wake-up stroke, WAKE-UP¹, the difference between thrombolysed and non-thrombolysed patients was 11.5% for a favorable outcome defined as mRS 0-1. A difference of 11.5% was also found in a recent meta-analysis [12] of six observational studies on patients with unknown stroke onset time, where the favorable outcome was defined as mRS 0–2. The MRI-based inclusion criteria in WAKE-UP compared to the CT-based inclusion in TWIST could lead to a smaller treatment effect in TWIST. We assume a treatment effect of 10% absolute difference in a binary endpoint setting (mRS 0–1 versus mRS 2–6) and a distribution between mRS categories similar to that of the WAKE-UP trial [1]. Accordingly, we anticipated a favorable outcome in 42% of participants in the non-thrombolysed group versus 52% in the thrombolysed group, which corresponds to an odds ratio (OR) of 1.50, and mRS distribution across the control group in six levels (categories 5 and 6 merged) of 15%, 27%, 23%, 17%, 13%, and 5%, respectively. With a power of 80%, a two-sided significance level of 5%, and an effect size specified as an OR of 1.50 from an ordinal logistic regression model for the ordinal outcome in the control group, the estimated sample size was 600. The revised target was therefore increased from 500 to 600 patients, i.e., 300 patients in each treatment arm.

All analyses will be conducted on data from all randomly assigned patients according to the intention to treat (ITT) principle, i.e. patients will be analyzed in the group they were randomized to, no matter what treatment they received, and regardless of whether they deviated from the protocol in any way. All outcomes and analyses are prospectively categorized as primary, secondary, or exploratory.

Ordinal logistic regression is pre-specified as the method for the analysis of the primary outcome using the common odds ratio as the corresponding effect size measure, restricting adjustment to age, symptom severity (baseline NIHSS score), and time from wake-up to randomization. All covariate adjustment variables are continuous variables, assumed linearity and non-linearity allowed. Non-linearity will only be included if the addition of the squared variable significantly improves the model fit (judged by an improvement in the likelihood ratio test).

Differences in all outcomes between the two treatment groups will be tested independently at the two-tailed 0.05 level of significance. All estimates of treatment effects will be presented with 95% confidence intervals (CIs). No formal adjustments will be undertaken to reduce the overall type I error associated with both secondary and exploratory analyses including the subgroup analyses. Their purpose is to supplement evidence from the primary analysis to better characterize the treatment effect. Results from the secondary and exploratory analyses will be interpreted in this context. Pre-specified subgroup analyses (as outlined in the supplementary material) will be carried out irrespective of whether there is a significant treatment effect on the primary outcome.

Sensitivity analyses for the primary outcome will be undertaken to test the robustness of the primary analysis with regards to protocol violations, baseline imbalance, clustering effects, and missing data (details in supplementary material). Unadjusted analysis will be undertaken as sensitivity analysis and will be presented for all primary and secondary outcomes.

Analyses will be conducted primarily using SAS 9.4 statistical software. Proposed tables and figures for the main publication are presented in Additional file 2.

Rigorous efforts are made to minimize the amount of missing outcome data. Minimal loss to follow-up for the 3 months assessment of the primary outcome is anticipated. If functional status at 3 months is unknown for any patient, we will apply the following algorithm: If the patient was alive at 3 months and measurements are available after baseline, we will use the level of function recorded on day 7 (i.e. measured at day 7 or prior to discharge from hospital) to impute functional status at 3 months. Hence, 3 months mRS will be imputed for patients with status at day 7 or discharge. We have chosen this simple form of single imputation, as it usually classifies patients appropriately, where both day 7 and 3 months data are known, and any additional gain from more complex multiple imputation methods is likely to be small [13]. The mRS outcome is assumed to be missing-at-random. Important explanatory and auxiliary variables such as age, baseline NIHSS score, geographical region, time of randomization after wake-up and treatment group are being collected, and will be examined to assess the plausibility of the missing at random assumption. No missing data are expected in adjustment covariates as completeness is assured by the web based randomization procedure. Any missing components of the NIHSS score will be imputed by multiple imputation. Sensitivity analyses based on different hypothesis about the missingness pattern of the primary outcome will be conducted to test for the robustness of the primary outcome, including analysis of the “complete case population,” i.e., based on the data of completers, using only observed without accommodating missing data.

Flow of patients through the study will be displayed in a standard Consolidated Standards of Reporting Trials diagram. The report will include the number of patients included, withdrawn, lost to follow-up, the number who received the allocated treatment and the number of patients analyzed.

To assess balance, description of collected baseline characteristics will be presented for the tenecteplase and control groups. Discrete variables will be summarized as frequencies and percentages. Unless otherwise indicated, percentages will be calculated according to the number of patients for whom data are available. If there are more than 5% missing values for a variable, the denominator will be added as a footnote in the corresponding summary table. Continuous variables will be summarized using either mean and standard deviation, or median and interquartile range (IQR). Time intervals will be summarized by medians and IQRs. Some of these variables were used in the minimization algorithm determining randomization allocation, and good balance would thus demonstrate successful operation of the algorithm.