Table 2.
Summary of main effects of design features from normal linear modelling of simulation outputs for all scenarios containing a genuine dexamethasone effect.
| Levels | Mean power change versus reference | P-value a | ||
|---|---|---|---|---|
| Scenario | Curve | (1) Sine curve: steep | −9% | <0.0001 |
| (2) Sine curve: slow | −5% | <0.0001 | ||
| (3) Non-monotone | Reference | – | ||
| Variance | (17.9 mL)2 | +17% | <0.0001 | |
| (22 mL)2 | +8% | <0.0001 | ||
| (26 mL)2 | Reference | – | ||
| Randomisation rate | 2 pt/month | -0.7% | 0.24 | |
| 4 pt/month | Reference | – | ||
| Maximum mean effect of dexamethasone | 16.4 mL | +44% | <0.0001 | |
| 8.2 mL | Reference | – | ||
| Mechanism | No interaction | +3% | 0.007 | |
| Treatment-BL interaction | Reference | – | ||
| Heteroscedasticity | Absent | +0.4% | 0.76 | |
| Present | Reference | – | ||
| Design option | Timing of adaptations (in terms of number of patients randomised) | 33 | 0% | 0.86 |
| 50 | +4% | 0.0010 | ||
| 66 | +6% | 0.0058 | ||
| 10; 35; 60 | +6% | 0.013 | ||
| 20; 45; 70 | +8% | <0.0001 | ||
| 49; 66; 83 | +9% | 0.0001 | ||
| 12; 24; 36; 48; 60 | +7% | 0.0003 | ||
| 16; 32; 50; 66; 84 | +9% | <0.0001 | ||
| 44; 55; 66; 77; 88 | +8% | <0.0001 | ||
| No adaptation | Reference | – | ||
| Utility function for adaptations | Predicted increase in precision of response at ED95 after one future randomisation | +2% | 0.033 | |
| Proportional to current probability that dose reduces MBL | Reference | – | ||
| Doses | Four active doses | +6% | <0.0001 | |
| Six active doses | Reference | – | ||
| Placebo allocation rate | 2/7 | +8% | <0.0001 | |
| 1/7 | Reference | – |
a
Each p-value is from the normal linear modelling of trial power and relates to the t-statistic comparing a given level to the reference category.