Download presentation
Presentation is loading. Please wait.
Published byDella Roberta Kelley Modified over 10 years ago
1
Design & Interpretation of Randomized Trials: A Clinician’s Perspective Francis KL Chan Department of Medicine & Therapeutics CUHK
2
Common problems of RCTs Originality Hypothesis Allocation concealment & randomization Evaluation of baseline data “Intention-to-treat” analysis Subgroup analysis
3
Is the study original? Ground breaking research? Does this work add to the literature in any way? Bigger, longer? More rigorous methodology? Results add to a meta-analysis of previous studies? Different population (age, sex, ethnic groups)?
4
Hypothesis & End point Many RCTs did not explicitly state their study hypotheses “The aim of this study was to compare the efficacy of a new treatment with the standard treatment…” Hypothesis 1: Treatment A is superior to the standard treatment Hypothesis 2: Treatment A is equivalent to the standard treatment
5
Hypothesis? Sample size estimation None!
6
Failure to detect a difference = Equivalence?
7
Superiority Trial The new treatment (µ N ) is superior to the standard treatment (µ S ) if the difference exceeds by a clinically important amount ( ). Test hypothesis (H): µ N - µ S >
8
Equivalence trial The new treatment is equivalent to the standard treatment if the maximal allowable difference does not exceed by a clinically important amount.
9
Equivalence trial - 0 + Equivalent Difference Not equivalent Favors new treatment Not equivalent Favors standard treatment New agent is not inferior to the standard
10
Assume non-inferiority if the lower limit of 95% CI is less than –5%, N=904 per group!
11
Allocation concealment & randomization Concealment of allocation (investigators and patients not knowing the assigned treatment before randomization) Was treatment assigned by an independent staff? What was the method of allocation concealment? contact with central office blinded packages sealed (opaque) envelopes
12
Allocation concealment
13
Comparison of baseline data Chan et al. Lancet 1997 Does P>0.05 indicate comparability of treatment groups?
14
Baseline data AzathioprinePlacebo Mean age54.754.9 Serum bilirubin ( mol/L) 37.230.9 Stage I disease %1412 Stage II disease %4443 Stage III disease %15 Stage IV disease %2730 Christensen et al. Gastro 1985 Effect of azathioprine on the survival of patients with primary biliary cirrhosis
15
Baseline data AzathioprinePlacebo Mean age54.754.9 Serum bilirubin ( mol/L) 37.230.9 Stage I disease %1412 Stage II disease %4443 Stage III disease %15 Stage IV disease %2730 Christensen et al. Gastro 1985 Effect of azathioprine on the survival of patients with primary biliary cirrhosis
16
Unadjusted P=0.10 Adjusted for bilirubin P=0.01
17
P=0.04 P=0.02 Columbus Investigators. NEJM 1997
18
Significant imbalance may not affect outcome Comparison of baseline data Non-significant imbalance may affect outcome Significance tests for baseline differences are inappropriate.
19
Significance tests for baseline differences Chan et al. Lancet 1997 INAPPROPRIATE
20
Significant imbalance may not affect outcome Comparison of baseline data Non-significant imbalance may affect outcome Significance tests for baseline differences are inappropriate. Table of baseline data should focus on factors affecting outcome.
21
45 baseline factors!
22
Significant imbalance may not affect outcome Comparison of baseline data Non-significant imbalance may affect outcome Significance tests for baseline differences are inappropriate. Table of baseline data should focus on factors affecting outcome. Analysis adjusted for baseline factors that are known to strongly influence the outcome (Covariate-adjusted analysis). Analysis of covariance for a quantitative outcome Logistic regression for a binary response Cox’s-proportional hazard model for time-to-event data
23
“Intention-To-Treat” Analysis “…results were analyzed according to the ITT principle.” Question: How were missing outcomes/ protocol violators handled in the so called “ITT” analysis?
24
“Intention-To-Treat” Analysis Endpt Savage et al. NEJM 1997
25
Minimize missing response on primary outcome Recommendations for ITT Analysis Follow up subjects who withdraw early Investigate & report the effect of missing response Report all deviations and missing response
26
Subgroup Analysis Randomised trial of home-based psychosocial nursing intervention for patients recovering from myocardial infarction. Frasure-Smith et al. Lancet 1997 “…The poor overall outcome for women, and the possible harmful impact of the intervention on women, underlie the need for…”
27
Effect of antenatal dexamethasone administration on the prevention of respiratory distress syndrome. Am J Obstet Gynecol 1981;141:276-87. Subgroup Analysis SteroidPlaceboP Pre-ecclampsia 21.2% (7/33) 27.3% (9/33) 0.57 No pre-ecclampsia 7.9% (21/267) 14.1% (37/262) 0.021
28
Effect of antenatal dexamethasone administration on the prevention of respiratory distress syndrome. Am J Obstet Gynecol 1981;141:276-87. Subgroup Analysis SteroidPlaceboP Pre-ecclampsia 21.2% (7/33) 27.3% (9/33) 0.57 No pre-ecclampsia 7.9% (21/267) 14.1% (37/262) 0.021 Difference 6.1%
29
Effect of antenatal dexamethasone administration on the prevention of respiratory distress syndrome. Am J Obstet Gynecol 1981;141:276-87. Subgroup Analysis SteroidPlaceboP Pre-ecclampsia 21.2% (7/33) 27.3% (9/33) 0.57 No pre-ecclampsia 7.9% (21/267) 14.1% (37/262) 0.021 Difference 6.1% Difference 6.2% P value depends on effect size & SE
30
Evaluation of Subgroup Analysis Tests of interaction (assess whether a treatment effect differs between subgroups) rather than subgroup P values Diff in Subgroup A – Diff in Subgroup B SE of the above Diff Z =
31
Trial of vitamin D supplements in pregnancy to prevent infant hypocalcemia. BMJ 1980;281:11-4. Interaction Test Difference = 0.42Difference = 0.15 0.42 – 0.15 = 0.27 SE of this Diff = 0.22 Z = Diff / SE = 1.23 P = 0.2 No evidence that the effect of Vit D is different between bottle-fed and breast-fed infants
32
General points regarding subgroup analysis Emphasis should remain on overall comparison More convincing if confined to a limited number of pre-specified subgroup hypothesis Rely on interaction tests, not P values View subgroup findings as exploratory (to be confirmed in subsequent trials)
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.