Randomized controlled trial Study Guide

📖 Core Concepts Randomized Controlled Trial (RCT) – an experiment that randomly assigns participants to at least one intervention group and a control group to evaluate efficacy or safety. Random Allocation – each participant has a known probability of being placed in any arm; balances known & unknown prognostic factors. Control Group – receives placebo, standard care, or an alternative treatment; essential for causal inference. Blinding (Masking) – concealing group assignment from participants, providers, and/or outcome assessors to reduce performance and detection bias. Intention‑to‑Treat (ITT) Analysis – all randomized participants are analysed in the groups to which they were allocated, preserving randomisation benefits. CONSORT – a 25‑item checklist (plus extensions) that standardises RCT reporting, ensuring transparency of design, execution, and results. 📌 Must Remember Superiority vs. Non‑inferiority vs. Equivalence Superiority: test if new ≠ control and is better (p < 0.05). Non‑inferiority: test if new is not worse than a predefined margin Δ. Equivalence: test if the difference lies within ±Δ. Sample‑size basics – larger N → higher power; under‑powered trials risk a Type II error (false negative). Type I error (α) is usually set at 0.05; Type II error (β) relates to power = 1 − β. Allocation concealment (e.g., opaque sealed envelopes, central randomisation) is distinct from blinding and must be maintained until assignment. Ethical rule – Clinical Equipoise – genuine uncertainty in the expert community is required to justify randomising patients. Trial registration – mandatory before enrolment (since July 2005) for journals following ICMJE policy. 🔄 Key Processes Design selection → Choose parallel, crossover, cluster, stepped‑wedge, or factorial based on intervention, population, and logistics. Randomisation method → Simple randomisation for large trials (>200). Restricted (permuted‑block or adaptive biased‑coin) for smaller trials to keep groups balanced. Allocation concealment → Implement sealed envelopes, central service, or pharmacy dispensing before enrolment. Blinding plan → Define who is blinded (participant, provider, assessor) and how it is achieved; document in CONSORT. Sample‑size calculation → Specify effect size, α, desired power (1‑β), and variance; adjust for anticipated dropout. Data collection & interim monitoring → Pre‑specify stopping rules for benefit, harm, or futility. Analysis → Conduct ITT analysis; choose appropriate model (logistic, ANCOVA, Cox) based on outcome type. 🔍 Key Comparisons Parallel‑group vs. Crossover Parallel: each participant stays in one arm; simpler, no carry‑over. Crossover: participants receive multiple interventions sequentially; higher power but requires wash‑out periods. Superiority vs. Non‑inferiority trials Superiority aims to prove better; null hypothesis = “no difference”. Non‑inferiority aims to prove not worse; null hypothesis = “difference ≥ Δ”. Simple randomisation vs. Permuted‑block Simple: independent ½ chance each; may produce imbalanced groups in small samples. Block: forces balance within each block; risk of predictability if block size known. ⚠️ Common Misunderstandings “Randomized trial” ≠ “Randomized controlled trial.” The latter always includes a control arm. Blinding ≠ Allocation concealment. Concealment prevents selection bias before assignment; blinding prevents performance/detection bias after assignment. Non‑inferiority does not prove superiority. A trial can show a new treatment is not worse, yet still be clinically inferior in practice. “Intention‑to‑treat” means ignoring protocol violations. It means analysing participants as randomised, regardless of adherence. 🧠 Mental Models / Intuition “Balancing scale” model: Randomisation is like placing participants on a perfectly balanced scale; any systematic difference after allocation is due to the intervention, not pre‑existing traits. “Layers of protection” model: 1️⃣ Randomisation → balances groups. 2️⃣ Allocation concealment → prevents selection bias. 3️⃣ Blinding → blocks performance & detection bias. 4️⃣ ITT analysis → preserves the randomised balance in the final estimate. 🚩 Exceptions & Edge Cases Blinding infeasible (e.g., physical therapy, surgery) → ensure outcome assessor blinding and use objective endpoints. Cluster RCTs → must adjust sample‑size for intra‑cluster correlation (ICC) and use hierarchical analysis. Stepped‑wedge design – useful when the intervention must eventually be rolled out to all clusters; analysis must account for time trends. Early stopping – can inflate effect size estimates; report the number of interim looks and adjustment method (e.g., O’Brien‑Fleming). 📍 When to Use Which Parallel‑group → standard drug efficacy studies, large homogeneous populations. Crossover → short‑acting, reversible interventions where each participant can serve as his/her own control. Cluster → interventions delivered at the group level (schools, villages) or when contamination between individuals is likely. Factorial → testing two or more interventions simultaneously and assessing interaction effects. Superiority → when you expect the new treatment to be better and want to claim improvement. Non‑inferiority → when the new treatment offers other benefits (cost, safety) and you need to show it isn’t substantially worse. Equivalence → when the goal is to demonstrate “no meaningful difference” (e.g., generic vs. brand drug). 👀 Patterns to Recognize Imbalance in baseline characteristics → may indicate flawed randomisation or inadequate concealment. High dropout rates combined with per‑protocol analysis → red flag for attrition bias. Selective reporting of only significant subgroup results → likely data dredging. CONSORT checklist missing items (e.g., no description of blinding) → potential bias. 🗂️ Exam Traps “Single‑blind” terminology – modern CONSORT expects you to specify who was blinded; a question that only says “single‑blind” is incomplete and may be wrong. Confusing superiority null hypothesis – the null is “no difference”; rejecting it supports superiority, not the other way around. Assuming simple randomisation works for any size – in small trials it can produce unequal groups; the correct answer will mention restricted methods. Equating “intention‑to‑treat” with “per‑protocol” – they are opposite approaches; expect a trap that swaps their definitions. Believing placebo use is always ethical – remember it is unethical when withholding proven effective treatment would cause harm. --- Use this guide to quickly scan each concept before the exam, and focus on the bolded decision rules and common traps to boost confidence and accuracy.