Skip to main content

Table 3 – Simulation results for scenarios where a non-random subset is re-randomised, outcomes differ across randomisation periods, or treatment effects carryover into subsequent randomisation periodsa

From: A re-randomisation design for clinical trials

  Treatment effect = 0 Treatment effect = 0.4
  Estimated treatment effect Type I error rate (%) Estimated treatment effect Difference in power vs. parallel group trialb (%)
Scenario 1: Sicker patients are re-randomised     
 • Unadjusted for patient effects 0.0 5.1 0.4 −0.4
 • Adjusted for patient effectsc 0.0 5.2 0.4 +10.8
Scenario 2: Patients who experienced a poor outcome are more likely to be re-randomised     
 • Unadjusted for patient effects 0.0 4.4 0.4 −1.2
 • Adjusted for patient effectsc 0.0 4.8 0.4 +6.1
Scenario 3: Patients who received the intervention are more likely to be re-randomised     
 • Unadjusted for patient effects 0.0 5.0 0.4 −0.3
 • Adjusted for patient effectsc 0.0 4.8 0.4 +7.3
Scenario 4: Patients who received the control are more likely to be re-randomised     
 • Unadjusted for patient effects 0.0 4.7 0.4 −0.3
 • Adjusted for patient effectsc 0.0 5.4 0.4 +7.0
Scenario 5: Patients’ health status changes for their subsequent re-randomisationd     
 • Unadjusted for patient effects 0.0 5.3 0.4 −1.0
 • Adjusted for patient effectsc 0.0 5.1 0.4 +13.9
Scenario 6: The intervention effect carries over into subsequent randomisation periods     
 • Unadjusted for patient effects NA NA 0.4 −2.9
 • Adjusted for patient effectsc NA NA 0.3 −22.9
 • Adjusted for patient effectsc, and adjusted for number of previous allocations to the intervention NA NA 0.4 +13.3
Scenario 7: The intervention and control effects carry over differentially into subsequent randomisation periods     
 • Unadjusted for patient effects NA NA 0.4 −10.3
 • Adjusted for patient effectsc NA NA 0.3 −29.8
 • Adjusted for patient effectsc, and adjusted for number of previous allocations to the intervention NA NA 0.5 +18.4
 • Adjusted for patient effectsc, and adjusted for both the number of previous allocations to the intervention and number of previous allocations to the control NA NA 0.4 +10.0
  1. aThe ICC was set to 0.50 for all scenarios. The number of observations was 200 for scenario 1 (100 patients randomised once, 25 patients randomised 4 times), 200 on average for scenario 2 (approximately 100 patients randomised once, 50 randomised twice), 195 on average for scenarios 3 and 4 (approximately 65 patients randomised once, 65 patients randomised twice), and 200 for scenarios 5, 6, and 7 (50 patients randomised four times)
  2. bPower for the parallel group trial was set at 80 %
  3. cAnalyses adjusted for patient-effects using a mixed-effects linear regression model, with a random intercept for patient
  4. dBoth analyses adjusted for randomisation period as an indicator variable