Box 910 Disadvantages of stopping a trial as a consequence of an interim analysis before the planned final analysis

♦ The risk of a false positive is increased. This problem and possible solutions have been discussed above.

♦ The trial is likely to have an overestimate of the difference between the treatments being compared. Although a number of solutions have been proposed for this problem, none provide a completely coherent and satisfactory solution.

♦ The trial will be relatively small and the estimated difference will generally be large. In most diseases, large differences between new and old treatments are relatively rare and therefore any such claimed difference will lack credibility, and often be insufficient to influence practice.

♦ The trial will tend to be relatively small and thus the confidence intervals around the estimated difference are likely to be very wide, giving considerable uncertainty as to its true value.

♦ For a time-to-event outcome measure the results at interim analyses will usually represent results in the early part of the Kaplan-Meier curve, longer term results may be quite different.

Table 9.17 Results of the EORTC randomized trial looking at the role of hormone therapy and chemotherapy in addition to surgery in the treatment of women with breast cancer when the trial was stopped (May 1985) and when the trial was published (March 1988). Reprinted with permission from John Wiley & Sons Ltd from reference [42]

Date Patients Deaths Results

Hormone therapy Chemotherapy p-value HR 95%CI p-value HR 95% CI

May 399 86 0.16 0.75 0.49-1.13 0.004 0.53 0.35-0.82

1985

March 410 202 0.06 0.77 0.58-1.02 0.23 0.84 0.64-1.12

1988

our view, no firm conclusions can be drawn from such a small study presented in this way'. To support this view, further trials of adding chemotherapy to radiotherapy were initiated after this report.

Example. The European Organization for the Research and Treatment for Cancer (EORTC) conducted a factorial randomized trial looking at the role of hormone therapy and chemotherapy in the treatment of women with locally advanced breast cancer [42]. The data monitoring committee was not independent. Instead, the accumulating data from the trial were inspected by the trial statistician on a 6-monthly basis, and if they were 'statistically significant' they were discussed with the study coordinator and possibly with the main clinicians participating in the trial. The trial started in December 1979 and was closed soon after May 1985, because of'slow accrual.' The slow accrual was probably a consequence of the results at the time being shown to the participating clinicians. The results were published in March 1988, as shown in Table 9.17.

It can be seen in Table 9.17 that the results when published appear to be quite different from those when the trial was effectively closed in May 1985. In May 1985, the effect of chemotherapy was encouraging with a large observed effect and an extreme p-value. In contrast at that time the results for hormone therapy looked less encouraging. When the trial was published the results for hormone therapy were actually more encouraging than those for chemotherapy - a complete reversal of the results seen at the interim analysis which led to the trial being stopped.

0 0

Post a comment