From: Overestimation of benefit when clinical trials stop early: a simulation study
Characteristic | Typical truncated trials (n=47,174) | Typical truncated trials if carried to completion (n=242,829) | Large truncated trials (n=295,602) |
---|---|---|---|
Number of participants | 541 (456–626) | 771 (648–896) | 9,616 (8070–11,160) |
Average follow-up (months) | 27.1 (23.0–31.2) | 49.9 (42.0–57.9) | 40.2 (33.8–46.7) |
Placebo event rate (per 100 person-years) | 8.6 (7.3–10.0) | 7.7 (6.5–9.0) | 2.3 (1.9–2.7) |
Z-value for stopping | 2.782 | 1.960 | 2.358 |
Number of events | 72 (53–96) | 194 (142–259) | 588 (429–784) |
Number of trials that overestimate benefit | 42,191 (89.4) | 181,218 (74.6) | 195,854 (66.3) |
Number of trials that underestimate benefit | 4983 (10.6) | 61,611 (25.4) | 99,748 (33.7) |
True RRR (%) | 38.6 (27.9–50.9) | 27.0 (18.7–36.6) | 26.0 (18.9–35.0) |
Observed RRR (%) | 54.8 (47.5–63.6) | 32.3 (25.6–41.2) | 28.1 (21.9–36.7) |
Absolute RRR overestimate (%) | 14.9 (6.4–24.6) | 5.3 (−0.1 to 11.4) | 2.3 (−1.3–6.3) |
Observed RRR/true RRR | 1.37 (1.12–1.83) | 1.18 (0.99–1.51) | 1.08 (0.96–1.28) |
Number of trials with negative true benefit (i.e. harm) | 265 (0.6) | 2172 (0.9) | 456 (0.2) |
Number of trials with observed RRR/true RRR in range | |||
1.0–1.2 | 10,559 (22.4) | 63,803 (26.3) | 99,369 (33.6) |
1.2–1.5 | 12,194 (25.8) | 53,165 (21.9) | 56,779 (19.2) |
> 1.5 | 19,173 (40.6) | 62,078 (25.6) | 39,250 (13.3) |