Statistical Analysis Plan for the Cardiac Magnetic Resonance GUIDEd Management of Mild-moderate Left Ventricular Systolic Dysfunction (CMR GUIDE) trial

This paper outlines the pre-specified statistical analysis plan for the CMR GUIDE trial, which aims to evaluate whether implantable cardioverter defibrillators guided by ventricular scar detection reduce sudden cardiac death or significant ventricular arrhythmias in patients with mild-to-moderate left ventricular systolic dysfunction using a Fine and Gray competing risks model.

The Gist

The Big Picture: A High-Stakes Game of "Hearts and Scars"

Imagine the heart as a house. Sometimes, the walls of this house get damaged, leaving behind "scars" (fibrosis) from past injuries. In this study, researchers are looking at houses where the walls are a bit weak (mild-to-moderate systolic dysfunction) and have visible scars.

The big question the researchers are asking is: If we find a scar in the wall, should we install a "fire alarm and sprinkler system" (an Implantable Cardioverter Defibrillator, or ICD) immediately to prevent a fire (sudden cardiac death), or should we just install a "security camera" (an Implantable Loop Recorder, or ILR) to watch and wait?

This document is the rulebook for how the researchers will count the scores at the end of the game. It doesn't tell us who won yet; it tells us exactly how they will decide who won.

---

1. The Players and the Setup

• The Players: Adults with heart weakness (either from blocked arteries or other causes) who have scars on their hearts but haven't had a major heart attack recently.
• The Teams:
  • Team ICD: These players get the "fire alarm" implanted. If their heart starts beating dangerously fast, the device shocks it back to normal.
  • Team ILR: These players get the "security camera." It records what the heart is doing but doesn't shock it. It's purely for observation.
• The Goal: To see if the "fire alarm" team has fewer "fires" (sudden cardiac death or dangerous heart rhythms that cause fainting) than the "security camera" team.

2. The Scoreboard (The Primary Outcome)

The main thing they are counting is a composite score. Think of it like a "Game Over" condition. A player loses if they experience:

1. Sudden Cardiac Death: The heart stops suddenly.
2. The "Faint" Event: A dangerous heart rhythm that causes the person to pass out or have dangerously low blood pressure.

How they count:
They aren't just counting who died first. They are using a special mathematical tool called a Fine and Gray model.

• The Analogy: Imagine a race where some runners trip and fall (the event we care about), but others get pulled off the track for a different reason, like a flat tire (dying from something else, like cancer).
• The researchers want to know: "If we ignore the flat tires, how much faster did the 'fire alarm' team finish the race compared to the 'camera' team?" This method ensures that dying from a non-heart cause doesn't mess up the math.

3. The Rules of the Game (Analysis Plan)

The "Intention-to-Treat" Rule

This is a golden rule in medical trials. It means: "You play for the team you were assigned to, even if you switch teams later."

• If a player was assigned to the "Camera" team but later decided to get a "Fire Alarm," they still count as a "Camera" player in the final score. This keeps the comparison fair and realistic, just like real life where people change their minds.

The "Tie-Breaker" (Win Ratio)

Sometimes, a simple "who died first" isn't enough to tell the whole story. So, the researchers have a Win Ratio system, like a tournament bracket. They pair up one player from the "Fire Alarm" team with one from the "Camera" team and compare them step-by-step:

1. Who lived longer? (If one died and the other didn't, the survivor wins).
2. Who had the fewer fainting spells? (If both lived, who had the fewer dangerous rhythms?).
3. Who had fewer hospital visits for heart failure?
4. Who felt better? (Based on a quality-of-life questionnaire).

They go down this list until one player is clearly "better" than the other. The final score is the ratio of "Fire Alarm winners" to "Camera winners."

4. Checking the Details (Secondary & Safety)

The researchers also plan to look at other things, like:

• Quality of Life: Did the patients feel better or worse? (Using questionnaires like the "Minnesota Living with Heart Failure Questionnaire," which is like a report card on how much heart trouble is ruining your day).
• Hospital Visits: How many times did they have to go to the hospital?
• Side Effects: Did the devices cause any problems?

They have a special rule for these extra checks called Multiplicity Adjustment.

• The Analogy: If you flip a coin 100 times, you might get 10 heads in a row just by luck. If you check 100 different things in a study, you might find a "win" just by chance. To stop this, they use a mathematical filter (the Holm-Sidak correction) to make sure the results they find are real and not just lucky guesses.

5. What About Missing Data?

If a patient drops out of the study or stops coming for check-ups, the researchers have a rule: They stop counting that person the moment they were last seen healthy.

• They won't guess what happened to them. They assume if they didn't report a "fire," there wasn't one. This is a strict, honest way of handling missing information.

Summary

This document is the instruction manual for the math. It says:

1. We will compare the "Fire Alarm" group vs. the "Security Camera" group.
2. We will count sudden deaths and fainting spells as our main score.
3. We will use special math to handle people who die from other causes.
4. We will use a "tournament style" comparison to see which group did better overall.
5. We will be very strict about not finding fake results by checking too many things.

The goal is to provide a clear, unbiased answer to the question: Does installing a defibrillator based on heart scars actually save lives compared to just watching and waiting?

Technical Summary

Based on the provided document, here is a detailed technical summary of the CMR GUIDE Statistical Analysis Plan (SAP).

1. Problem Statement

The clinical problem addressed is the management of patients with mild-to-moderate left ventricular systolic dysfunction (LVSD) (LVEF 36–50%) who are at risk for sudden cardiac death (SCD) but do not meet current guidelines for Implantable Cardioverter Defibrillator (ICD) implantation based solely on ejection fraction.

Current guidelines primarily rely on LVEF thresholds (<35%) for ICD implantation. However, a significant portion of patients with LVEF between 36–50% still experience SCD. The CMR GUIDE trial investigates whether using Cardiac Magnetic Resonance (CMR) imaging to detect ventricular scar/fibrosis (via Late Gadolinium Enhancement, LGE) can better stratify risk and guide ICD implantation. The core hypothesis is that an ICD strategy guided by the presence of LGE scar will reduce SCD or hemodynamically significant ventricular arrhythmia (HSVA) compared to standard care (using an Implantable Loop Recorder, ILR, for monitoring).

2. Methodology

Study Design

• Type: Prospective, multi-centre, randomised, controlled trial (RCT).
• Population: 353 recruited patients (finalized recruitment) with ischemic (ICM) or non-ischemic (NICM) cardiomyopathy, LVEF 36–50%, and documented ventricular scar/fibrosis on CMR.
• Intervention:
  • ICD Arm: Single-chamber ICD implanted with specific programming (high-rate VF zone, delayed therapy VT zone) to minimize inappropriate shocks.
  • Control Arm: Implantable Loop Recorder (ILR) for rhythm monitoring.
• Primary Outcome: A composite of Sudden Cardiac Death (SCD) or Hemodynamically Significant Ventricular Arrhythmia (HSVA) (defined as VT/VF causing syncope or hypotension SBP <90 mmHg).
• Follow-up: Minimum 3 years (up to 7 years total study duration).

Statistical Analysis Plan (SAP) Key Features

The SAP (Version 1.2) pre-specifies the analytical framework to ensure rigor and minimize bias.

• Primary Analysis Method:

  • Model: Fine and Gray competing risks regression model.
  • Rationale: This model accounts for the fact that non-sudden cardiac deaths are competing risks that prevent the observation of the primary event.
  • Estimand: Sub-distribution Hazard Ratio (HR) with 95% Confidence Intervals (CI).
  • Censoring: Patients are censored at the last known alive and event-free date.
  • Sensitivity: A Cox proportional hazards model (cause-specific HR) will be used as a supporting analysis.

• Handling of Competing Risks:

  • For the primary outcome, non-SCD deaths are competing risks.
  • For secondary outcomes like HSVA, all deaths are treated as competing risks.
  • For HF hospitalizations, non-HF deaths are competing risks.

• Multiplicity Adjustment:

  • Primary Outcome: No adjustment (unadjusted $\alpha = 0.05$).
  • Key Secondary Outcomes: A sequential Holm-Sidak correction is applied to control the family-wise error rate for: All-cause mortality, MLHFQ score, HF hospitalizations, and Cardiovascular mortality.
  • Exploratory Outcomes: No multiplicity adjustment; results are reported as point estimates and CIs.

• Subgroup Analyses:

  • Pre-specified subgroups include Age, Sex, LVEF strata, Cardiomyopathy type (ICM vs. NICM), NYHA class, and QRS duration.
  • Interaction terms will be tested to assess heterogeneity of treatment effects.

• Win Ratio Analysis:

  • A hierarchical Win Ratio approach is employed to jointly analyze multiple outcomes, prioritizing clinical importance.
  • Hierarchy:
    1. Time to SCD.
    2. Time to HSVA.
    3. Number of HF hospitalizations.
    4. Minnesota Living with Heart Failure Questionnaire (MLHFQ) score at 24 months.
  • Pairs of patients (one from each arm) are compared stepwise until a "winner" is determined.

• Data Handling:

  • Analysis Set: Intention-to-Treat (ITT) / Treatment Policy strategy.
  • Missing Data: No imputation for the primary endpoint; analysis assumes all reported events are captured.
  • Software: SAS Enterprise Guide (v7.1+) and R (v4.0.0+).

3. Key Contributions

• Refined Risk Stratification: The study moves beyond LVEF alone, proposing a CMR-guided approach to identify "high-risk" patients within the "grey zone" of mild-moderate LV dysfunction.
• Robust Statistical Framework: The SAP introduces a sophisticated handling of competing risks (Fine and Gray) and a hierarchical Win Ratio analysis, which is increasingly recognized as a superior method for composite endpoints in cardiology trials where event types vary in severity.
• Addressing the "Grey Zone": By focusing on LVEF 36–50% with scar, the trial addresses a specific gap in current guidelines where ICD benefit is unproven.
• Standardized Device Programming: The ICD arm utilizes specific programming (MADIT-RIT inspired) to mitigate the risk of inappropriate shocks, a known limitation in previous ICD trials.

4. Results

• Note: This document is a Statistical Analysis Plan (SAP) and a preprint (dated April 2026). It does not contain the final clinical results of the trial.
• Status: The document outlines the plan for analyzing data from 353 recruited patients. It confirms that recruitment is complete (mid-2015 to mid-2022) and follow-up is ongoing/expected to conclude by mid-2025.
• Power: The study was powered to detect a Hazard Ratio of 0.45 with 80% power, assuming an annual event rate of 7.5% in the control group. The SAP notes that even with lower event rates (4%), the study retains >80% power.

5. Significance

• Clinical Impact: If the CMR-guided strategy proves effective, it could fundamentally change guidelines for ICD implantation, sparing patients without scar from unnecessary device implantation while protecting those with scar who are currently excluded based on LVEF alone.
• Methodological Benchmark: The SAP serves as a methodological reference for future trials dealing with competing risks and composite endpoints, particularly the integration of Win Ratio analysis in cardiac device trials.
• Safety and Efficacy Balance: By explicitly planning to analyze HSVA and syncope alongside mortality, the trial aims to balance the life-saving potential of ICDs against the risks of inappropriate shocks and procedural complications.

Disclaimer: As noted in the document, this is a preprint that has not been peer-reviewed and should not be used to guide clinical practice at this stage. The results are pending final analysis and publication.