randomized controlled trial

**Definition and design** - Prospective experimental study; the investigator assigns the exposure (drug, procedure, behavioral intervention). - Participants meet inclusion/exclusion criteria, give informed consent, and are randomized to treatment vs control. The control may receive placebo, standard of care, or an active comparator. - Groups are followed over time and outcomes are compared (e.g., relative risk, absolute risk reduction, hazard ratio).

**Key design features (high-yield)** - **Randomization**: Distributes known and unknown confounders equally between groups, which reduces selection bias and confounding. Its strength is that it controls for unmeasured confounders. Types include simple, block, stratified, and cluster randomization. - **Allocation concealment**: Prevents investigators from knowing the next assignment, which prevents selection bias at enrollment. - **Blinding (masking)**: Single-blind (the patient), double-blind (patient and investigator), triple-blind (also the data analysts). Blinding reduces performance bias, observer/detection bias, and the placebo effect. - **Placebo control**: Controls for the placebo effect. - **Intention-to-treat (ITT) analysis**: Analyzes participants in the groups they were randomized to, regardless of adherence or crossover. It preserves randomization, avoids attrition and crossover bias, and is conservative, so it may bias toward the null. It is the preferred primary analysis. - **Per-protocol analysis**: Includes only adherent participants. It can show larger effects but is prone to bias. - **As-treated analysis**: Analyzes participants by the treatment they actually received. - **Crossover trial**: Each participant receives both treatments in sequence, separated by a washout period, so each serves as their own control. It is best for stable chronic conditions. - **Equipoise**: Genuine uncertainty about which arm is better is the ethical requirement for conducting an RCT.

**Phases of clinical trials (FDA)** - Phase 0: Microdosing; pharmacokinetics in a few people. - Phase I: Small number of healthy volunteers (or patients, for oncology); tests safety, toxicity, and pharmacokinetics/pharmacodynamics. - Phase II: Small number of patients with the disease; tests efficacy and dosing and further assesses safety. - Phase III: Large, randomized, multicenter trials comparing the new treatment with the standard of care or placebo; these are the definitive efficacy trials used for approval. - Phase IV: Postmarketing surveillance; detects rare or long-term adverse effects.

**Strengths** - Highest internal validity of any primary study design for evaluating interventions. - Establishes causation and minimizes confounding.

**Limitations** - Expensive and time-consuming. - Limited generalizability (external validity) because of strict eligibility criteria. - Not feasible or ethical for harmful exposures (e.g., smoking); use cohort or case-control studies instead. - Inefficient for rare outcomes or diseases with long latency. - Loss to follow-up (attrition bias), nonadherence, and crossover can undermine results. - Underpowered trials risk type II error (β). Power = 1 – β.

**Biases still possible** - Attrition bias, nonadherence, and unblinding (e.g., recognizable side effects). - Hawthorne effect: participants alter behavior because they know they are observed. - Lead-time bias, which is more relevant to screening trials. - Early stopping for benefit may overestimate the effect.

**Evidence hierarchy (top to bottom)** Systematic review/meta-analysis of RCTs > RCT > cohort > case-control > cross-sectional > case series/case report > expert opinion.

**Exam pearls** - "Best study design to determine whether an intervention causes an outcome" = RCT. - "Best way to reduce confounding" = randomization (design stage). Other design-stage methods are matching and restriction. Stratification and multivariate regression are analysis-stage methods. - A study with a randomized exposure but no control group is not an RCT. - When a trial is stopped early for harm or overwhelming benefit, a data safety monitoring board (DSMB) is typically involved. - The CONSORT statement is the reporting standard for RCTs. - Statistical outputs: p < 0.05 and a 95% CI that excludes the null value (RR = 1 or difference = 0) indicate statistical significance. Absolute risk reduction (ARR) = risk in control − risk in treatment. Number needed to treat (NNT) = 1/ARR.