Design Clinical Trial Protocol

---
name: design-clinical-trial-protocol
description: Use when designing a clinical trial — including study type selection, randomization and blinding strategy, endpoint definition, sample size calculation, and regulatory compliance requirements for a study protocol.
source: "ICH E6(R2) Good Clinical Practice Guidelines (2016); ICH E9 Statistical Principles for Clinical Trials (1998); CONSORT 2010 Statement (Schulz et al., BMJ 2010); FDA Guidance for Industry: E8(R1) General Considerations for Clinical Studies (2022)"
tags: [clinical-trial, rct, randomization, blinding, endpoints, gcp, research-design, biostatistics]
---

# Design Clinical Trial Protocol

Design a rigorous clinical trial protocol by defining the study question, selecting the appropriate design, specifying endpoints, calculating required sample size, and planning for regulatory compliance from the outset.

## Why This Is Best Practice

**Why best:** A rigorous protocol — clear PICO question, appropriate design, adequate randomization/blinding, and a pre-specified analysis plan — produces valid, reproducible causal estimates and satisfies the GCP/CONSORT standards regulators and journals require.
**Adopted by:** FDA, EMA, PMDA, and all major regulatory agencies require clinical trials submitted for drug approval to follow ICH E6 GCP standards. CONSORT 2010 reporting guidelines are mandated by >600 journals including NEJM, Lancet, JAMA, and BMJ. WHO International Clinical Trials Registry Platform (ICTRP) requires prospective registration of all clinical trials.
**Impact:** Schulz et al. (2010) CONSORT analysis showed that trials with adequate allocation concealment had 30-40% smaller (more accurate) effect estimates than poorly concealed trials, demonstrating that design quality directly affects result validity. Prospective registration (ClinicalTrials.gov, ISRCTN) — compared to retrospective registration — reduces outcome-switching bias, which Dwan et al. (2008) found in up to 62% of published trials.

## Steps

### 1. Define the research question using PICO

Specify:
- **Population:** precise inclusion and exclusion criteria (demographics, disease stage, comorbidities, washout requirements)
- **Intervention:** treatment, dose, route, duration
- **Comparator:** placebo, standard of care, active comparator
- **Outcomes:** primary endpoint (one only; drives the sample size calculation) and secondary endpoints (pre-specified, not data-dredged)

The primary endpoint must be:
- Clinically meaningful (not a surrogate unless validated)
- Measurable with a validated instrument
- Assessable within the planned follow-up period

### 2. Select the study design

Match design to the research question and phase:
- **Phase I:** dose escalation, safety, pharmacokinetics — 3+3 design or model-based (BOIN, CRM); n=20-80
- **Phase II:** preliminary efficacy, dose selection — randomized or single-arm; n=50-300
- **Phase III:** confirmatory efficacy — randomized, double-blind, controlled; n=100-10,000+
- **Phase IV:** post-approval safety and effectiveness — observational or randomized

For interventional trials: randomized controlled trial (RCT) is the gold standard for causal inference.

Special designs:
- **Adaptive design:** pre-specified rules for sample size re-estimation or dose dropping at interim; requires FDA pre-agreement
- **Crossover design:** participants receive both treatments sequentially; efficient but only for stable, reversible conditions
- **Cluster randomization:** groups (hospitals, schools) rather than individuals randomized; requires ICCs in sample size calculation

### 3. Plan randomization and blinding

**Randomization strategy:**
- Simple randomization: acceptable for large trials (n>200); vulnerable to imbalance in smaller trials
- Block randomization (recommended): blocks of 4-6 ensure balance over time; block size should be concealed from investigators
- Stratified randomization: balance key prognostic factors (disease severity, site, age) across arms; use for multi-site trials or trials with strong covariates
- Use a validated randomization system (REDCap, IVRS/IWRS)

**Blinding level:**
- Double-blind: participants AND assessors unaware of allocation — required for subjective endpoints
- Single-blind: assessor-blinded only; acceptable for objective endpoints (mortality)
- Open-label: no blinding; acceptable when blinding is impossible; use blinded outcome adjudication

### 4. Calculate sample size

Sample size calculation must be documented in the protocol:
```
For two-group parallel trial (continuous endpoint):
n per group = 2σ²(Zα/2 + Zβ)² / δ²

Where:
σ = standard deviation of outcome
δ = minimum clinically important difference (MCID)
Zα/2 = 1.96 (two-sided α = 0.05)
Zβ = 0.84 (power = 80%) or 1.28 (power = 90%)
```

Add 10-20% for expected dropout/attrition.

Source the MCID and σ from pilot data, literature, or regulatory precedent — not arbitrary choices.

### 5. Plan the statistical analysis

Pre-specify in a Statistical Analysis Plan (SAP) before unblinding:
- **Primary analysis population:** Intent-to-Treat (ITT; all randomized) as the primary; Per-Protocol as sensitivity analysis
- **Handling missing data:** pre-specify multiple imputation or mixed-model for repeated measures (MMRM)
- **Multiplicity adjustment:** if multiple primary endpoints or interim analyses, specify Bonferroni, Holm, or alpha-spending function (O'Brien-Fleming)
- **Interim analyses:** pre-specify stopping rules (futility and efficacy boundaries) with Data Safety Monitoring Board (DSMB) review

### 6. Address regulatory and ethical requirements

Before the first participant is enrolled:
- **Protocol registration:** ClinicalTrials.gov (US), EudraCT (EU), ISRCTN — before enrollment begins
- **IRB/Ethics Committee approval:** submit protocol, informed consent form, investigator brochure
- **Informed consent:** document process; ensure comprehension beyond signature
- **DSMB:** required for trials with mortality endpoints or vulnerable populations
- **IND/CTA filing:** required for investigational drugs before Phase I

## Common Mistakes

- **Composite primary endpoint designed to inflate event rate rather than measure a clinically coherent outcome:** Components must be similarly clinically important (MACE = MI + stroke + CV death is coherent; combining death with "hospitalization for any reason" is not).
- **Sample size not powered for the primary endpoint:** Powering for secondary endpoints produces a statistically underpowered primary analysis.
- **Retrospective registration:** Trials registered after enrollment begins have lower credibility and are rejected by CONSORT-compliant journals.

## When NOT to Use

- Rare diseases with <50 available patients: consider N-of-1 designs, Bayesian adaptive designs, or master protocols (basket/umbrella trials).