Choosing the Right Analysis Strategy in RCTs: From ITT to CACE [Concept of Trial analysis]
Introduction
Randomized controlled trials (RCTs) are the gold standard for causal inference in clinical medicine, offering unparalleled protection against confounding. But randomization is not the finish line—how we analyze trial data determines the truth we tell. From policy-level insights to bedside advice, each analytic approach—ITT, mITT, Per-Protocol, As-Treated, and CACE—serves a different clinical purpose.
In this expanded article, we unpack five major analysis strategies with a focus on clinical realism, methodological rigor, and decision-making consequences. We then dive deep into the rising star of trial analysis: the Complier Average Causal Effect (CACE).
1. Aligning Analysis with the Clinical Question
The fundamental law of trial analytics: choose your analytic strategy based on the question you want to answer. Here's how that looks in real-world terms:
| Clinical Question | Best Analysis | Use Case |
| “Does offering this treatment improve outcomes across all eligible patients?” | Intention-to-Treat (ITT) | Public health policy, guideline formulation |
| “What is the effect among those who would actually adhere to treatment?” | CACE or cautious Per-Protocol (PP) | Clinician counseling a motivated patient |
| “What happened to those who actually received treatment?” | As-Treated (AT) | Mimicking real-world behavioral patterns |
| “What if only those starting treatment are analyzed?” | Modified ITT (mITT) | Convenience, but high bias risk |
🔍 Secret Insight: Each method estimates a different causal contrast. Mixing them up is like comparing apples, oranges, and metaphoric fruit salads.
2. CACE: The Adherence-Adjusted Gold Standard
What Is CACE?
CACE estimates the treatment effect among “baseline compliers”—patients who would adhere to whichever group they were randomized to. Unlike PP or AT, CACE:
- Preserves randomization.
- Avoids selection bias.
- Reflects realistic clinical engagement.
It answers: "What is the causal effect in patients who would follow medical advice regardless of assignment?"
Four Principal Adherence Strata:
- Baseline Compliers – Do as assigned.
- Always Takers – Take treatment regardless.
- Never Takers – Avoid treatment regardless.
- Defiers – Do the opposite (usually assumed negligible).
CACE isolates stratum #1 using a latent class model framework grounded in principal stratification.
3. Estimating CACE: The Mechanics
Step-by-Step Derivation
- Estimate ITT effect (outcome difference between assigned groups).
- Estimate compliance rates:
- qt: % treated in the intervention arm.
- qc: % treated in the control arm (crossovers).
- Estimate proportion of compliers: qt - qc
- Apply the formula:
Interpretation
This yields a causal estimate for treatment efficacy in the subset who would follow clinical guidance—the most clinically actionable subgroup.
4. Comparing the Analytic Arsenal
| Method | Question Answered | Strengths | Limitations |
| ITT | “What is the effect of offering treatment?” | Preserves randomization, generalizable | Dilutes effect if adherence is low |
| mITT | “Effect among those who started treatment?” | Operationally simple | Selection bias, lacks causal clarity |
| PP | “Effect if protocol followed exactly?” | Reflects pure biological efficacy | Breaks randomization, high bias risk |
| AT | “Effect among those who got treatment?” | Real-world mimicry | Confounding risk—no randomization |
| CACE | “Effect among those who would comply?” | Preserves randomization, clinically focused | Complex modeling, needs assumptions |
🔍 Secret Insight: PP and AT analyses create pseudo-cohorts with broken randomization. Treat their results as observational unless corroborated by CACE.
5. Practical and Ethical Implications
Stakeholder-Specific Relevance
- Policymakers: Prioritize ITT. It captures total system impact, including dropout, refusal, and deviation.
- Clinicians: Rely on CACE for accurate counseling of engaged patients.
- Trialists:
- Pre-specify analytic plans (ITT primary, CACE secondary).
- Avoid cherry-picking PP/AT results unless stated as exploratory.
- Respect ethical imperatives: no post hoc exclusions unless justified symmetrically.
Conclusion: Let the Question Choose the Method
RCTs provide the blueprint for therapeutic truth, but analysis is the blueprint’s language. Speak it clearly. The ideal clinical trial doesn’t just test a drug—it tests a real-world decision. To serve both ethics and efficacy:
- Use ITT to anchor your primary findings.
- Use CACE to extract efficacy among the engaged.
- Avoid overinterpreting PP or AT as definitive.
In the evolving precision-care era, analysis strategy is a clinical instrument. Sharpen it wisely.
Key Takeaways
- 🧠 Analysis = Answer Framing. ITT ≠ CACE ≠ PP.
- 🔍 CACE reflects adherence without breaking randomization.
- 📊 mITT, PP, AT = caution: prone to bias, best for exploratory aims.
- 🧭 Use stakeholder-aligned logic: ITT for policy, CACE for patient care.
- 📑 Predefine strategy. Treat PP/AT as observational.