Randomized Controlled Trial (RCT) - Complete Guide for Your Seminar

What is an RCT? The Big Picture

"It is really an epidemiologic experiment. Since its introduction, the RCT has questioned the validity of such widely used treatments as oral hypoglycaemic agents, varicose vein stripping, tonsillectomy, hospitalization of all patients with myocardial infarction, multiphasic screening, and toxicity and applicability of many preventive and therapeutic procedures."

• Park's Textbook of Preventive and Social Medicine

The Classic RCT Design (Flowchart)

The 6 Basic Steps of an RCT

Step 1 - Drawing Up a Protocol

• Aims and objectives
• Questions to be answered
• Criteria for selection of study and control groups
• Sample size
• Procedures for allocation of subjects
• Treatments to be applied - when, where, how, and to what kind of patients
• Standardization of working procedures
• Responsibilities of all parties involved
• How the outcome will be evaluated

Step 2 - Selecting the Populations

1. Informed consent - They must agree after being fully told about the purpose, procedures, and possible risks
2. Representativeness - They should represent the reference population
3. Eligibility/susceptibility - They must actually be susceptible to or suffering from the condition being studied (e.g., testing a new anti-anaemia drug requires participants to actually be anaemic)

Step 3 - Randomization (The Heart of the RCT)

"Randomization is the 'heart' of a controlled trial." - Park's

• It eliminates selection bias - the investigator has no control over who goes where
• It ensures known and unknown confounders are equally distributed between groups - this is something even careful matching cannot guarantee

Allocation Concealment (Often Confused with Blinding!)

• Allocation concealment = hiding the sequence of group assignment from investigators at the time of enrolling participants
• It prevents investigators from knowing which arm the next participant will be assigned to, so they cannot consciously or unconsciously manipulate who gets enrolled
• Without adequate concealment, an RCT should be treated as non-randomized
• Trials with inadequate concealment overestimate treatment effect by up to 40% on average

Step 4 - Manipulation/Intervention (The Control Group)

• Concurrent controls = both groups studied simultaneously - preferred, as they face the same environment and time period
• Historical controls = comparing with data from past patients - less reliable due to changes in environment, diagnostics, and care over time

Blinding (Masking) - Preventing Observer Bias

• Patients expect a new drug to work - if they know they're in the intervention arm, this itself can cause improvement (placebo effect)
• Investigators who know a patient's group may unconsciously manage them differently or assess outcomes differently
• Double-blinding is impossible in surgical trials (you can't hide a surgery); in such cases, use hard outcomes resistant to bias, or a blinded third-party assessor

Step 5 - Follow-up

• Both groups must be followed for the same period under identical conditions
• Losses to follow-up are a major source of bias - the groups must be monitored to retain participants
• Follow-up schedules, timing, and outcome measurement criteria must be pre-specified in the protocol

Step 6 - Assessment of Outcome

• Defined clearly in the protocol before the trial starts
• Measurable and objective
• Assessed without knowledge of group assignment (blinded assessment)

• Primary endpoint - the main outcome the trial is powered to detect
• Secondary endpoints - additional outcomes tracked
• Hard endpoints - objective, unambiguous (death, MI, bone fracture)
• Soft endpoints - subjective, more prone to bias (pain score, quality of life)

Types of RCTs

1. Clinical Trials

2. Preventive Trials

3. Field Trials

4. Community Trials

Study Designs Within RCTs

Parallel Design (Most Common)

Crossover Design

Intention-to-Treat (ITT) Analysis

"In a randomized controlled trial, it is important to conduct the main analyses following the 'intention to treat' principle. This means that the study population should be analyzed in terms of their original randomization assignment and not according to what treatment they actually received." - Smith & Tanagho's General Urology

Ethical Considerations

Equipoise

When Should a Trial be Stopped?

• If one group is clearly doing better mid-trial (interim analysis shows significant benefit)
• If unexpected harm is observed in one group
• Data Safety Monitoring Boards (DSMB) monitor ongoing trials for this purpose

Helsinki Declaration Principles

• Informed consent is mandatory
• Participant welfare takes precedence over research interests
• Use of placebo is only ethical when no proven treatment exists

Strengths and Limitations of RCT

Strengths

• Eliminates confounding - known and unknown
• Highest level of evidence (Level I) in the evidence hierarchy
• Can establish causality, not just association
• Allows blinding to reduce bias
• Produces clean, interpretable results

Limitations

• Expensive and time-consuming
• Ethical constraints - can't randomize when one treatment is clearly better or when withholding treatment would be harmful
• Hawthorne effect - participants may change behavior because they know they're being studied
• May have poor external validity (generalizability) - trial populations are often carefully selected and may not reflect real-world patients
• Not suitable for rare diseases or outcomes that take decades to develop
• Cannot be used for studying causes of disease (only interventions)

CONSORT Guidelines (For Reporting RCTs)

• How many were assessed for eligibility
• How many were excluded and why
• How many were randomized
• How many in each arm received the allocated intervention
• How many were lost to follow-up
• How many were analyzed

Quick Comparison: RCT vs. Other Study Designs

One-Line Memory Aids for Seminar

• Randomization = equal chance for everyone = eliminates selection bias
• Allocation concealment = hiding the sequence = prevents enrollment manipulation
• Blinding = hiding who got what = prevents performance and assessment bias
• ITT = analyze as randomized, not as treated
• Equipoise = ethical justification for randomization
• CONSORT = the reporting standard

Honest Answer

What I can offer instead:

Biases in a Randomized Controlled Trial - Complete Guide

The Big Framework First

"It is not a major simplification to consider all these types of biases in three categories: selection bias, information bias, and confounding."

• Scott-Brown's Otorhinolaryngology Head & Neck Surgery

Stage-by-Stage Breakdown of Biases in an RCT

STAGE 1 - Design & Enrollment Biases

1. Selection Bias

• External selection bias (Volunteer/Healthy Worker effect): RCTs typically enroll patients who are healthier, more motivated, and more compliant than the average patient. So the results may overestimate what would happen in routine clinical practice. This is why RCT results show efficacy (works under ideal conditions) but may not reflect effectiveness (works in the real world).

  "Trials usually enrol patients who tend to be different (often healthier) from the target population and the results tend to overestimate the effects compared to what they would be in routine practice. This contributes to the differences between the efficacy of a therapy observed in the highly controlled circumstances in selected clinical settings and the effectiveness of a treatment in actual practice."

  • Scott-Brown's

• Internal selection bias (failure of randomization): If randomization is done poorly, the two arms may not be comparable at baseline - one arm may have systematically sicker or healthier patients. This is what randomization is designed to prevent.

2. Allocation Concealment Bias

"Empirical studies have shown that trials with inadequate concealment overestimate the treatment effect by as much as 40% on average."

• Scott-Brown's (from previous session)

STAGE 2 - During the Trial (Performance Biases)

3. Performance Bias (Co-intervention Bias)

"Performance bias: systematic differences in care being given to study patients other than the preplanned interventions being evaluated."

• Barash's Clinical Anesthesia

4. Placebo Effect

"(a) Placebo effect: Treatment benefit due to perception of treatment"

• The Harriet Lane Handbook

5. Nocebo Effect

"(b) Nocebo effect: Adverse effects due to perception of treatment"

• The Harriet Lane Handbook

6. Hawthorne Effect

"(c) Hawthorne effect: Participant change of behavior from being studied"

• The Harriet Lane Handbook

7. Observer-Expectancy Bias (Pygmalion Bias)

"(d) Observer-expectancy bias: Researcher's belief in efficacy of a treatment affects their actions"

• The Harriet Lane Handbook

8. Contamination Bias

STAGE 3 - At Follow-up (Attrition Biases)

9. Attrition Bias (Loss to Follow-up Bias)

"Attrition bias: systematic differences in the withdrawal of patients from each of the two intervention groups."

• Barash's Clinical Anesthesia

"In a randomized controlled trial, it is important to conduct the main analyses following the 'intention to treat' principle. This means that the study population should be analyzed in terms of their original randomization assignment and not according to what treatment they actually received."

• Smith & Tanagho's General Urology

STAGE 4 - Outcome Assessment (Detection Biases)

10. Detection Bias (Ascertainment Bias / Observer Bias)

"Detection bias: systematic differences in the ascertainment and recording of outcomes."

• Barash's Clinical Anesthesia

11. Information Bias (Measurement Bias)

"Information bias results from incorrect information about the determinant or the outcome or both. The important question that has to be answered is: 'Has information been gathered in the same way?'"

• Scott-Brown's

• Recall bias - one group remembers symptoms/events differently than the other
• Interviewer bias - interviewers probe one group more thoroughly
• Instrument bias - measuring devices are not calibrated or standardized

STAGE 5 - Analysis & Reporting Biases

12. Reporting Bias (Selective Outcome Reporting)

"Reporting bias" is listed as a key domain in the AHRQ grading framework for study quality.

• Kaplan & Sadock's Comprehensive Textbook of Psychiatry

13. Publication Bias

14. Sponsor-Related Bias (Funding Bias)

"Sponsor-related bias" is explicitly listed as a bias domain in quality assessment frameworks.

• Kaplan & Sadock's

15. Analysis Bias (Per-Protocol vs. ITT)

"(b) As-treated analysis: Analyzing outcomes based on receipt of intervention, not initial assignment - almost always incorrect due to participant crossover not being random."

• The Harriet Lane Handbook

Summary Table

The Cochrane Risk of Bias Tool - What Examiners Love to Ask

1. Random sequence generation - was the sequence truly random?
2. Allocation concealment - was it properly hidden?
3. Blinding of participants and personnel - was performance bias prevented?
4. Blinding of outcome assessment - was detection bias prevented?
5. Incomplete outcome data - was attrition handled properly (ITT)?
6. Selective outcome reporting - were all pre-specified outcomes reported?

RCT Data Collection to Analysis - A Complete Walkthrough

The Worked Example We'll Use Throughout

Trial: A double-blind RCT testing whether a new antihypertensive drug (Drug X) reduces the rate of major cardiovascular events (heart attack or stroke) compared to placebo over 2 years, in adults aged 40-70 with Stage 1 hypertension.

• 200 participants randomized: 100 to Drug X, 100 to Placebo
• Primary outcome: Occurrence of major cardiovascular event (Yes/No) - a dichotomous outcome
• Secondary outcome: Mean reduction in systolic BP at 12 months - a continuous outcome

PART 1 - DATA COLLECTION

Step 1: What Data is Collected?

Step 2: Standardization of Data Collection

"The protocol aims at preventing bias and to reduce the sources of error in the study."

• Park's Textbook

Step 3: Handling Dropouts and Non-Compliance

PART 2 - THE 2×2 CONTINGENCY TABLE (Foundation of Analysis)

"Outcomes of each treatment arm can be reported individually... more clinically useful metrics would enable us to describe the relative effect of one treatment compared to the other."

• Rockwood & Green's Fractures in Adults

PART 3 - CALCULATING THE KEY MEASURES OF EFFECT

1. Event Rates in Each Group

• Treatment Event Rate (TER) = Events in treatment group / Total in treatment group = 10/100 = 0.10 = 10%
• Control Event Rate (CER) = Events in control group / Total in control group = 30/100 = 0.30 = 30%

2. Relative Risk (RR) - Risk Ratio

"Risk ratios enable us to say treatment A reduces the odds (or risk) of an outcome by x% compared with treatment B."

• Rockwood & Green's

3. Relative Risk Reduction (RRR)

4. Absolute Risk Reduction (ARR) - Risk Difference

5. Number Needed to Treat (NNT) ★ Most Clinically Useful

"The number needed to treat is the number of patients who require treatment to prevent one negative event."

• Rockwood & Green's

6. Odds Ratio (OR)

"The odds of infection in treatment compared with controls is 0.11 [in the reference example]."

• Rockwood & Green's

Summary of All Measures for Our Example

PART 4 - CONTINUOUS OUTCOME ANALYSIS

• Mean difference = 18 - 8 = 10 mmHg (Drug X reduced SBP by 10 mmHg more than placebo)
• Test statistical significance using an independent samples t-test (if data is normally distributed) or Mann-Whitney U test (if non-normal)

PART 5 - STATISTICAL SIGNIFICANCE (p-value and Confidence Intervals)

p-value

• The p-value tells you the probability of getting a result this extreme by chance alone, assuming the null hypothesis (no difference) is true
• Conventional threshold: p < 0.05 = statistically significant
• In our example, suppose the p-value for the primary outcome is p = 0.002 → highly significant

95% Confidence Interval (CI)

"The CI defines the range in which the true value exists with 95% confidence. For example, if a study reports that nails reduced the risk of infection by 50% compared with plates... and has an associated CI of 25% to 75%, one may say with 95% confidence that the true treatment effect lies between 25% and 75%."

• Rockwood & Green's

• CI does not include 1.0 → confirms the result is statistically significant
• If CI included 1.0 (e.g., [0.85 - 1.20]), the result would not be significant

• For RR/OR: if CI crosses 1.0 → not significant
• For ARR/Mean difference: if CI crosses 0 → not significant

PART 6 - INTENTION-TO-TREAT (ITT) vs PER-PROTOCOL ANALYSIS

ITT Analysis (The Standard, Preferred Method)

• Actually took their medication
• Dropped out
• Switched groups

"The advantage of an intention-to-treat analysis is that the question that is being addressed corresponds exactly with the one clinicians and patients try to answer in clinical practice."

• Scott-Brown's

Per-Protocol Analysis (Secondary, Exploratory)

"The problem with per-protocol analysis is that if many patients do not receive the treatment to which they were randomized, the study no longer represents an experimental study."

• Scott-Brown's

PART 7 - SUBGROUP ANALYSIS

"Subgroup analyses are comparisons between randomized groups in a subset of patients with specific characteristics... They can be dangerous and misleading, but they can also provide important extra insight."

• Scott-Brown's

• Multiple comparisons inflate the chance of a false-positive result (Type I error)
• Each subgroup is smaller, so underpowered (risk of false-negative, Type II error)
• Rule: Only subgroup analyses pre-specified in the protocol before the trial started are valid. Post-hoc subgroup analyses are hypothesis-generating only

PART 8 - PRIMARY vs SECONDARY ENDPOINTS

"The analysis should focus on a single outcome - often referred to as the 'primary endpoint' - to avoid the problems of interpreting the outcome of multiple hypothesis tests. This primary endpoint should also be used for the power calculation in the design phase."

• Scott-Brown's

• Primary endpoint: Cardiovascular event rate (what the trial is powered to detect) → reported with RR, ARR, NNT, 95% CI, p-value
• Secondary endpoints: SBP reduction, hospitalization rate, quality of life scores → reported but interpreted with more caution; not used to claim the trial "succeeded" or "failed"

PART 9 - INTERPRETING RESULTS - THE COMPLETE PICTURE

1. Is the result statistically significant? (p-value, CI)
2. Is the result clinically meaningful? (NNT - does treating 5 patients to prevent 1 event seem worthwhile?)
3. Was ITT used? (protects against bias)
4. Were both arms truly comparable at baseline? (check the "Table 1" of any RCT paper)
5. What was the dropout rate? (>20% loss to follow-up seriously threatens validity)
6. Are the results generalizable? (efficacy in selected trial population vs. effectiveness in real-world practice)

"Assessment of outcome: The incidence of positive/negative results is rigorously compared in both groups, and the differences, if any, are tested for statistical significance."

• Park's Textbook of Preventive and Social Medicine

Final Summary Flowchart

DATA COLLECTION
    ↓
Build 2×2 Table (for binary outcomes)
    ↓
Calculate: TER → CER → RR → RRR → ARR → NNT → OR
    ↓
Calculate 95% CI and p-value
    ↓
For continuous outcomes: Mean difference + t-test
    ↓
Apply ITT Analysis (primary) + Per-Protocol (secondary)
    ↓
Report primary endpoint → secondary endpoints → subgroup analyses
    ↓
Interpret: Statistical significance + Clinical significance + Generalizability