AVERT-DOSE Statistical Analysis Plan

This paper outlines the statistical analysis plan for the AVERT-DOSE trial, a Phase 3 multi-arm, multi-stage, response-adaptive randomized study designed to identify the optimal early mobility training regimen for stroke patients.

The Gist

Imagine you've just had a stroke. You are in the hospital, and the doctors need to decide: When is the best time to start getting you moving again?

Should you start walking and sitting up immediately (within hours)? Or should you wait a bit longer to let your brain rest?

The AVERT-DOSE study is a massive, international experiment designed to answer this question. It's like a high-stakes cooking competition where chefs (doctors and therapists) are trying to find the perfect "recipe" for recovery.

Here is the simple breakdown of what this paper is about, using some everyday analogies.

1. The Big Question: How Much "Exercise" is Too Much?

In the past, there was a debate. Some thought getting patients up and moving immediately was great. Others worried it might be dangerous or too tiring for a fresh stroke.

This study didn't just compare "Moving" vs. "Staying in Bed." Instead, it tested four different "doses" of movement:

• The Reference Dose: A standard, low amount of movement (the "control group").
• Three Experimental Doses: Different amounts of sitting, standing, and walking, ranging from moderate to high intensity.

Think of it like testing four different volumes on a radio. The researchers wanted to find the "Goldilocks" volume—not too quiet (too little movement) and not too loud (too much movement)—but the perfect setting for the brain to recover.

2. The "Smart" Game: How the Trial Worked

This wasn't a boring, static experiment. It was a smart, adaptive trial.

Imagine a video game where the difficulty adjusts automatically based on how well you are playing.

• The Setup: The study started with four teams (arms).
• The Smart Switch: As the trial went on, a computer system watched the results. If one "dose" of movement started looking like it was helping patients recover better than the others, the computer would automatically send more new patients to that winning team.
• The Goal: This ensured that the trial didn't waste time testing bad ideas. It focused resources on the treatments that seemed to be working best.

Note: The study was originally planned to be huge (2,700 people) and run in two stages. However, due to the pandemic and funding running out, it had to stop early with about 1,000 people. The "smart" system was adjusted to work with this smaller group, but the goal remained the same: find the best dose.

3. Who Was in the Study?

They recruited people who had just had a stroke (within 48 hours). They split them into two groups based on how severe the stroke was:

• Mild Stroke: People who were still quite alert and mobile.
• Moderate Stroke: People who had more significant weakness.

They made sure the groups were balanced by age, where they lived, and what emergency treatments they got (like clot-busting drugs), so the comparison was fair.

4. What Were They Measuring? (The Scoreboard)

The main goal was to see who could live independently three months after the stroke.

• The "Win": Being able to go about daily life without needing help (a score of 0–2 on the "Modified Rankin Scale," which is like a disability report card).
• The "Loss": Being dependent on others or passing away.

They also looked at secondary scores:

• Could the person walk 50 meters without help?
• How fast could they walk?
• Did they have fewer complications (like falls or infections)?
• How did they feel about their quality of life?

5. The Rules of the Game (Statistics)

The researchers were very careful to be fair and accurate.

• Blinding: The people checking the results at 3 months didn't know which "dose" of movement the patient had received. This prevents bias (like a judge favoring a team they know).
• No "Per Protocol" Cheating: They analyzed everyone who started the trial, even if they missed a few sessions. This is like judging a race based on everyone who lined up at the start, not just the ones who finished perfectly.
• Handling Missing Data: If a patient couldn't be reached for the final check-in, the statisticians used clever math (multiple imputation) to estimate what their result likely would have been, based on similar patients.

6. The Bottom Line

This paper is the Statistical Analysis Plan. It's the rulebook written before the final results were revealed. It explains exactly how the researchers will crunch the numbers to ensure the answer is trustworthy.

Why does this matter?
If this study finds that a specific amount of early movement leads to more people walking out of the hospital on their own, it could change how stroke units work worldwide. It could mean thousands more people regain their independence, saving money on care and, most importantly, giving people their lives back sooner.

In a nutshell: The AVERT-DOSE study is a smart, data-driven search for the perfect amount of "get-up-and-go" therapy to help stroke survivors recover faster and better.

Technical Summary

Based on the provided document, here is a detailed technical summary of the AVERT-DOSE Statistical Analysis Plan (SAP) V1.0.

1. Problem Statement

The AVERT-DOSE trial addresses the critical clinical uncertainty regarding the optimal dose of early mobility training for patients with acute ischemic stroke. While early mobilization is standard care, the specific amount, frequency, and intensity required to maximize functional recovery without causing harm remain undefined. Previous trials (like the original AVERT trial) suggested that very early mobilization might be harmful in some contexts, but the optimal "dose" for different stroke severities (mild vs. moderate) was unknown. The study aims to determine which specific early mobility regimen yields the best functional outcomes for patients with mild and moderate stroke.

2. Methodology

Study Design

• Type: Phase 3, Multi-Arm, Multi-Stage (MAMS), Covariate-Adjusted, Response-Adaptive (CARA), Randomized Controlled Trial (RCT).
• Population: Adults (≥18 years) with acute ischemic stroke (first or recurrent), admitted within 48 hours of onset.
  • Strata: Two severity strata based on National Institute of Health Stroke Scale (NIHSS):
    • Mild: NIHSS 0–7
    • Moderate: NIHSS 8–16
• Interventions: Four arms total (three experimental dose regimens + one pre-specified reference arm).
  • Interventions involve task-specific, upright/out-of-bed training (sitting, standing, walking) delivered by trained staff within 48 hours of stroke onset, continuing for 14 days or until discharge.
  • Doses vary in amount, frequency, task difficulty, and intensity.
• Adaptive Design Modification:
  • Original Plan: A two-stage design with adaptive sample size re-estimation and dropping of inferior arms.
  • Actual Execution: Due to funding constraints and the COVID-19 pandemic, the trial was stopped early. The design was modified (in a blinded manner) to a four-arm single-stage CARA trial.
  • Final Sample Size: ~1,000 participants (650 mild, 350 moderate), significantly lower than the original target of 2,700–3,600.

Randomization (CARA Algorithm)

Randomization is performed independently for each stratum using a complex algorithm combining two components:

1. Covariate-Adjusted: Uses Pocock and Simon biased coin minimization to balance Age (18-64, 65-80, 81+), Region (Aus, Asia, UK/Europe, South America), and Hyperacute Treatment (Thrombolysis/EVT status) across arms.
2. Response-Adaptive: Uses an Urn Model to dynamically shift allocation probabilities.
   • As data accumulates, arms showing more favorable outcomes (mRS 0–2) receive more participants.
   • Arms showing poor outcomes (mRS 6) receive fewer.
   • The reference arm's allocation probability is fixed at 25%, while the remaining 75% is distributed among experimental arms based on performance.

Primary and Secondary Outcomes

• Primary Estimand: Proportion of participants achieving little or no disability (mRS 0–2) at 3 months.
  • Strategy: Treatment policy strategy (handling intercurrent events via Missing At Random assumptions).
• Secondary Efficacy:
  • Ordinal shift across full mRS scale at 3 months.
  • Proportion achieving unassisted walking 50m within 3 months.
  • Time to unassisted walking 50m (competing risk analysis with death).
  • Walking speed (10m walk test) at 3 months.
  • Health-related quality of life (EQ-5D-5L).
• Safety: Mortality (14 days and 3 months), Serious Adverse Events (SAEs).

Statistical Analysis Plan

• Primary Analysis: Binary logistic regression with treatment arms as predictors, adjusting for baseline NIHSS, age, region, and hyperacute treatment.
  • Effect Measure: Adjusted Odds Ratio (aOR) and Standardized Risk Difference (sRD).
  • Hypothesis Testing: Rerandomization tests (10,000 replications) to generate p-values, adhering to FDA guidelines for covariate adjustment.
• Missing Data: Primary analysis assumes Missing At Random (MAR) and uses Multiple Imputation with Chained Equations (MICE). Sensitivity analyses (Tipping Point) will be conducted.
• Multiplicity: No formal multiplicity adjustments for the primary comparisons (each experimental arm vs. reference tested at two-tailed $\alpha=0.05$).
• Software: Stata and R.

3. Key Contributions

• Adaptive Design Implementation: This is one of the few large-scale stroke trials to successfully implement a Covariate-Adjusted Response-Adaptive (CARA) design in a multi-arm setting, allowing for ethical allocation of patients to potentially superior treatments during the trial.
• Resilience to Disruption: The plan documents a rigorous, blinded modification of the trial design from a complex two-stage MAMS to a single-stage CARA in response to external funding and pandemic disruptions, preserving scientific integrity.
• Estimand Framework: The analysis strictly adheres to the ICH E9 (R1) Estimand framework, clearly defining population, intercurrent event strategies, and summary measures.
• Advanced Statistical Rigor: The plan utilizes sophisticated methods including:
  • Rerandomization tests for hypothesis testing.
  • Competing risk models (cause-specific hazards) for time-to-walking analysis (accounting for death as a competing risk).
  • Multiple imputation for missing data.
  • Frequentist confidence estimation for clinical meaningfulness (NNT=20 threshold).

4. Results

• Note: This document is a Statistical Analysis Plan (SAP), not a results paper. It outlines how the data will be analyzed, not the outcomes of the analysis.
• Status: The document states the last participant was recruited on December 4, 2025, with a final sample size of 1,000. The analysis has not yet been performed or reported in this text.
• Expected Output: The plan details the generation of specific tables (CONSORT flow, baseline characteristics, primary/secondary outcomes) and figures (Grotta bars for mRS distribution, forest plots for subgroups) for the main manuscript.

5. Significance

• Clinical Impact: The results of AVERT-DOSE will provide high-level evidence to define the "optimal dose" of early mobilization. This could fundamentally change stroke rehabilitation guidelines, potentially preventing harm from over-ambitious early mobilization or under-treating patients who could benefit from higher doses.
• Methodological Benchmark: The successful execution and reporting of a CARA design in a global, multi-center stroke trial serves as a blueprint for future adaptive trials in neurology and rehabilitation, demonstrating how to handle complex randomization and design changes without unblinding or bias.
• Policy Relevance: By identifying the specific dose-response relationship, the findings will help healthcare systems optimize resource allocation (nursing/physiotherapy time) for stroke units, balancing cost-effectiveness with patient outcomes.

Conclusion: The AVERT-DOSE SAP represents a robust, pre-specified framework for analyzing a complex, adaptive stroke trial. Despite the reduction in sample size due to external factors, the statistical methodology remains rigorous, ensuring that the final results will provide valuable insights into the safety and efficacy of varying early mobility doses for stroke recovery.