ARCHE - Randomized Controlled Trials (RCTs)

Introduction

A randomized controlled trial is an experimental study design in which participants are randomly assigned to receive either an intervention (treatment) or a control (placebo or standard treatment). RCTs are considered the gold standard for testing causal effects of treatments or interventions because randomization, if done properly, balances both known and unknown confounders between groups. Key aspects of RCTs include random sequence generation, allocation concealment, blinding, and intention-to-treat analysis. When appraising an RCT, we focus on how well it prevented bias through these methods.

Key Elements to Appraise

Randomization: Check that the study explicitly states it was randomized. Importantly, how was randomization done? A credible method (e.g. computer random number generator or random number table) should be used, Beware of inadequate methods (like alternation, or date of birth) that introduce bias.

Allocation Concealment: This ensures that the person enrolling participants cannot predict the next assignment. Look for statements like “allocation was concealed using sealed opaque envelopes” or central randomization. If allocation was not concealed, recruiters could (even unintentionally) introduce selection bias. The JBI checklist highlights allocation concealment as a separate critical item,

Baseline Comparability: Good RCTs report baseline characteristics of each group. Check if groups are similar on key factors (age, severity of disease, etc.). Significant imbalances (especially in small trials) may affect outcomes; the investigators should either stratify randomization or adjust in analysis. The checklist asks “were treatment groups similar at baseline?”,

Blinding (Masking): Determine whether participants, healthcare providers, and outcome assessors were blinded to the assigned groups. Blinding reduces performance and detection bias. For example, if patients know they got a new drug, they might report symptoms differently, or if doctors know, they might treat them differently. The checklist explicitly covers blinding of participants, providers, and assessors,

Follow-up and Attrition: Check how many participants completed the trial. High dropout rates, or large differences in dropout between groups, can bias the results. The study should report reasons for loss and ideally use intention-to-treat (ITT) analysis (analyzing everyone as assigned, regardless of dropout), JBI asks if “the follow-up was complete and, if not, differences were described”, and whether ITT was used,

Intervention Fidelity: Apart from the primary intervention, were other treatments controlled? The groups should be treated identically except for the intervention itself. This includes same monitoring, co-interventions, and instructions. The JBI checklist mentions that other interventions should be similar,

Outcome Measures: Outcomes should be pre-specified, objective (when possible), and measured equally in both groups. For example, “blood pressure measured by trained staff using a standard protocol” is good. If outcomes are subjective (like pain scores), blinding is especially important. Check if outcome assessors were blinded.

Statistical Analysis: The study should present appropriate analyses. This includes intention-to-treat as mentioned, appropriate handling of missing data (e.g. last observation carried forward or sensitivity analysis), and using suitable statistical tests. P-values with confidence intervals should be given. The JBI checklist asks if appropriate statistical analysis was used,

Reporting: Look for a flow diagram of participants (e.g. CONSORT diagram) showing numbers assessed, randomized, followed-up, and analyzed. Verify that all primary and secondary outcomes listed in the methods are reported in the results.

Common Mistakes / Red Flags

Inadequate Randomization: If randomization method isn’t described or is flawed (like assigning based on clinic number or date of visit), selection bias can occur. The Cochrane handbook equates non-random methods with high risk of bias.

Lack of Concealment: If recruiters could see the next assignment (e.g. open list), they might (even unconsciously) enroll different patients, breaking randomization. This is a subtle but major source of bias.

No Blinding: While not all trials can blind (e.g. surgery vs no surgery), lack of blinding should be noted. If participants or assessors knew assignments, results (especially subjective outcomes) might be biased.

High or Differential Attrition: Many dropouts (e.g. >20%) or big differences between groups suggest attrition bias. Always check the dropout rate and whether reasons were similar in both groups,

Per-Protocol Only Analysis: If the study only reports results for participants who completed the treatment (per-protocol) and ignores those who dropped out or switched treatments, this can bias the effect. ITT is preferred and is explicitly checked,

Multiple Outcomes/P-Hacking: Be wary if many outcomes are measured but only the “significant” ones are reported. Also, watch out for very small p-values that might be due to multiple testing without adjustment.

Small Sample Size: Some trials are underpowered. If the paper lacks a sample size calculation or justification, small negative studies may not truly rule out a difference.

Imbalance or Errors in Data: If baseline characteristics are very different, or if randomization failed in some way, results must be interpreted with caution. Check for any unusual exclusions after randomization.

Example

Imagine an RCT testing a new antihypertensive drug vs placebo in 200 patients. First, confirm randomization: the paper should say something like “Patients were randomized 1:1 to drug or placebo using a computer-generated list.” This addresses sequence generation. Check allocation concealment: for instance, “Treatment assignments were concealed in sealed, opaque envelopes” shows care. Look at baseline tables: age, sex, baseline blood pressure should be similar in both groups. Blinding: ideally, the paper notes it was double-blind (neither patients nor doctors knew who got drug or placebo). If it was open-label, that’s a limitation. See how many completed the study: if 10% dropped out equally in both arms and they used ITT (analyzing all 200 in their original groups, ), that’s good. Also, outcome measurement should be described: “BP measured by staff unaware of group” is ideal. Finally, review results: do they report mean BP with 95% CIs, and a p-value? Do they mention any side effects? An example sentence might be: “At 6 months, systolic BP decreased by 15 mmHg in the drug group vs 5 mmHg in placebo (mean difference 10 mmHg, 95% CI 6-14, p<0.001).”

Quick Checklist

Proper randomization method described (e.g., computer-generated),
Allocation concealment used to prevent selection bias,
Groups similar at baseline; blinding of participants, providers, and assessors,
Complete follow-up with minimal attrition; ITT analysis used,
Identical treatment apart from intervention; objective outcomes measured equally,
Appropriate statistical analysis with CIs and p-values,
CONSORT-compliant reporting with flow diagram.

Take-Home Points