# Lecture 26: Clinical Trials and Study Design
## Statistics / Biostatistics
---
## Learning Objectives
By the end of this lecture, students will be able to:
1. Describe the phases of clinical trials (Phase I-IV)
2. Explain the principles of randomization, blinding, and allocation concealment
3. Distinguish between superiority, non-inferiority, and equivalence trials
4. Describe intention-to-treat vs. per-protocol analysis
5. Identify common sources of bias in clinical trials and their mitigation
---
## Lecture Content
### I. What Is a Clinical Trial?
A clinical trial is a prospective experimental study in which an intervention is assigned to human participants to evaluate its effects on health outcomes. It is the gold standard for establishing causality in medicine and is governed by strict ethical and regulatory frameworks, including the Declaration of Helsinki and Good Clinical Practice (GCP) guidelines. Institutional Review Board (IRB) or Research Ethics Board (REB) approval is required, along with informed consent from all participants. To reduce publication bias, clinical trials must be registered prospectively on platforms such as ClinicalTrials.gov.
### II. Phases of Clinical Trials
**Phase I** trials focus on safety and dosing, enrolling a small sample of 20-80 participants (often healthy volunteers). The goal is to determine the maximum tolerated dose, characterize pharmacokinetics, and establish the safety profile. Phase I trials are not designed to test efficacy.
**Phase II** trials provide preliminary evidence of efficacy and further characterize side effects. They enroll a moderate sample of 100-300 participants who have the disease of interest and may be single-arm or randomized.
**Phase III** trials provide definitive evidence of efficacy and safety. They enroll large samples (hundreds to thousands) and are typically randomized, often double-blind, and controlled against the standard of care or placebo. Phase III results are required for regulatory approval by agencies such as the FDA, Health Canada, and the EMA.
**Phase IV** trials are conducted after drug approval as post-marketing surveillance. They monitor long-term safety, rare adverse effects, and effectiveness in broader populations than those studied in Phase III. Phase IV findings can lead to drug withdrawal if serious issues emerge.
A horizontal timeline showing the four phases of clinical trials. Phase I (small group icon, ~20-80 people): "Safety, dosing." Phase II (medium group, ~100-300): "Preliminary efficacy." Phase III (large group, ~1000+): "Definitive RCT." Phase IV (population icon): "Post-marketing surveillance." Below the timeline, approximate duration for each phase is indicated (Phase I: 1-2 years, Phase II: 1-3 years, Phase III: 2-5 years, Phase IV: ongoing). An annotation shows that only about 10% of Phase I drugs eventually gain approval.
### III. Randomization
Randomization ensures that treatment groups are comparable at baseline, minimizing confounding. **Simple randomization** gives each participant an equal probability of being assigned to each group (like a coin flip) but may produce imbalanced groups, especially with small samples. **Block randomization** randomizes within blocks of fixed size to ensure approximately equal numbers in each group at all times -- for example, within every 4 patients, 2 go to treatment and 2 to control. **Stratified randomization** performs randomization separately within strata defined by important prognostic factors (such as sex or disease severity), ensuring balance on these factors. **Adaptive randomization** adjusts allocation probabilities during the trial based on accumulating data or covariate balance.
**Allocation concealment** is the process of hiding the randomization sequence from those enrolling participants. It prevents selection bias, which could occur if knowing the next assignment influenced who was enrolled. Allocation concealment is distinct from blinding.
### IV. Blinding
**Single-blind** trials keep participants unaware of their assignment. **Double-blind** trials keep both participants and investigators or assessors unaware, and this is considered the ideal for minimizing bias. **Triple-blind** trials additionally blind the data analysts. **Open-label** trials reveal assignments to all parties, which is sometimes necessary (for example, when comparing surgical versus medical interventions) but increases the risk of performance and assessment bias.
A **placebo** is an inert treatment that looks identical to the active treatment. It controls for the placebo effect and ensures blinding is maintained. Ethical considerations require that proven effective treatment not be withheld from participants.
### V. Control Groups
**Placebo controls** use an inert comparator and are appropriate when no standard treatment exists or in add-on designs. **Active controls** compare the new treatment to the existing standard of care and are required ethically when effective treatments exist. **No treatment or waitlist controls** provide no intervention and are ethically limited. **Historical controls** compare to previously treated patients but are subject to temporal bias and are generally discouraged.
### VI. Superiority, Non-Inferiority, and Equivalence Trials
A **superiority trial** aims to show that the new treatment is better than the comparator, with H0 stating no difference and H1 stating the new treatment is superior. This is the standard approach for most RCTs.
A **non-inferiority trial** aims to show that the new treatment is not worse than the comparator by more than a pre-specified margin (delta). It is used when the new treatment has other advantages such as fewer side effects, lower cost, or easier administration. The non-inferiority margin must be clinically justified.
An **equivalence trial** aims to show that both treatments have similar effects within a margin, with H0 stating treatments differ by at least delta and H1 stating they differ by less than delta. This design is common in bioequivalence studies for generic drugs.
A number line diagram showing the difference in outcomes (new - standard). Panel A (Superiority): The 95% CI is entirely to the right of zero -- superiority demonstrated. Panel B (Non-inferiority): The 95% CI is entirely to the right of the negative non-inferiority margin (-delta) -- non-inferiority demonstrated. Panel C (Equivalence): The 95% CI falls entirely within (-delta, +delta) -- equivalence demonstrated. Panel D (Inconclusive): The 95% CI crosses the margin. Each panel is annotated with the interpretation. The zero point and margins are clearly marked.
### VII. Intention-to-Treat vs. Per-Protocol Analysis
**Intention-to-treat (ITT)** analysis includes all randomized participants in their assigned groups, regardless of compliance, withdrawal, or crossover. It preserves the benefits of randomization, provides a conservative estimate of treatment effect (diluted by non-compliance), and is the standard primary analysis in RCTs.
**Per-protocol (PP)** analysis includes only participants who completed the trial as planned. It may overestimate the treatment effect because non-compliers who may have done worse are excluded. Per-protocol analysis is useful as a sensitivity analysis and is important in non-inferiority trials, where ITT is anti-conservative.
**Modified ITT** excludes participants who never received any treatment or had no post-baseline data, offering a middle ground between the two approaches.
### VIII. Bias in Clinical Trials
Several types of bias can threaten the validity of clinical trials. **Selection bias** results from non-random or imbalanced allocation and is mitigated by proper randomization and allocation concealment. **Performance bias** arises from differences in care beyond the intervention and is mitigated by blinding. **Detection or measurement bias** results from biased outcome assessment and is mitigated by blinding the assessors. **Attrition bias** results from differential dropout between groups and is mitigated by ITT analysis and minimizing loss to follow-up. **Reporting bias** involves the selective reporting of favorable outcomes and is mitigated by trial registration and pre-specified analysis plans.
### IX. Sample Size and Interim Analyses
Sample size is calculated before the trial begins, based on the expected effect size, alpha, desired power, variability, and expected dropout rate. **Interim analyses** are pre-planned examinations of accumulating data that can stop the trial early for efficacy (overwhelming benefit), futility (no chance of benefit), or safety concerns. Because multiple looks at the data inflate the Type I error rate, alpha must be adjusted using methods such as O'Brien-Fleming boundaries or Lan-DeMets spending functions. Interim analyses are overseen by a Data Safety Monitoring Board (DSMB).
A CONSORT flow diagram template. Starting with "Assessed for eligibility (n = ...)," it branches into "Excluded (n = ...)" with reasons. The remaining participants are "Randomized (n = ...)," split into Treatment arm and Control arm. Each arm shows "Allocated to intervention," "Received allocated intervention," "Did not receive" (with reasons), "Lost to follow-up" (with reasons), "Discontinued" (with reasons), and "Analyzed" (ITT and PP numbers). The diagram clearly tracks participant flow from enrollment to analysis.
---