Estimating the size of the treatment effect

As emphasized at the beginning of this chapter, an important statistic that should always be calculated when a clinical trial is analysed is the estimate of the difference between

Table 9.14 Three approaches to spending a 0.05 significance level over the four analyses to be performed in a trial, made up of three interim analyses and one final analysis

Analysis number

Pocock

O'Brien and Fleming

Peto-Haybittle

1

0.018

0.001

0.001

2

0.018

0.004

0.001

3

0.018

0.019

0.001

4

0.018

0.043

0.0049

the treatments being compared. Although this is a relatively simple idea and is straightforward at the planned end of the trial, the calculation of an estimate becomes very complicated during interim analyses. We discuss below the reason.

Table 9.15 shows the CHART randomized clinical trial discussed above, and the results of the trial at each annual analysis. The rows show how the estimate has changed over the life of the trial, including the time at which reports were finally published. It can be seen that the estimate reached a peak at 20 per cent in 1992 more than twice its final value of 9 per cent reported in 1997. This trial was not monitored using any of the methods presented in Table 9.13, but using Bayesian methods and as a consequence was not stopped before its planned end.

Table 9.16 shows when the trial would have been stopped using the various approaches in Table 9.14. It can be seen that depending on the a-spending approach used, this trial may have been stopped in either 1992 or 1993 and reporting an estimated 2-year survival improvement of 20 per cent or 15 per cent, respectively. Both these estimates are much larger than the final 9 per cent that was reported. This shows that a trial which is stopped early for a positive result will typically have stopped on a 'random high' with the estimate of the difference between treatments being 'overestimated.' Proposals to adjust the estimate to allow for this overestimation have been made [39], but none provide a completely satisfactory solution and thus most trials which have stopped early present the simple unadjusted estimate. If this had been done with the CHART trial in 1992, then it would have reported an estimate of treatment effect more than two times larger than the final reported (and more reliable) estimate. Readers should bear this in mind reviewing the results of such trials. The attendant difficulties involved in assessing the accumulating

Table 9.15 Results of the CHART lung cancer trial at approximately annual intervals during the accrual of patients to the trial from 1992 to 1995, and at the time of reporting of the results of the trial in 1997. Reprinted with permission from Elsevier Science (The Lancet, 2001, 358, 375-81)

Year

Hazard

Estimated 2-year

p-value

ratio

survival improvement (%)

1992

0.55

20

0.007

1993

0.63

15

0.001

1994

0.70

12

0.004

1995

0.75

9

0.006

1997

0.76

9

0.004

Table 9.16 Estimates of when the CHART lung cancer trial would have stopped using three different approaches to monitoring together with the estimated improvement in 2-year survival

«-spending

Date trial would

Estimated 2-year

function

have been stopped

survival improvement (%)

Pocock

1992

20

O'Brien and Fleming

1993

15

Peto-Haybittle

1993

15

efficacy and safety results of a trial, and reaching incorrect premature conclusions has led to the formation of independent data monitoring committees. This committee, which is made up of individuals completely independent of the trial, will typically review the results of such trials in a confidential manner, and recommend continuation and early termination as appropriate. The structure and format of these committees are discussed further in Section 8.10.

Cigarette Crusher

Cigarette Crusher

Get All The Support And Guidance You Need To Permanently STOP Being A Slave To Nicotine And Cigarettes. This Book Is One Of The Most Valuable Resources In The World When It Comes To Easy Ways To Eliminate Smoking Addiction And Revitalize Your Body.

Get My Free Ebook


Post a comment