Devita, Hellman, and Rosenberg's Cancer

LWBK1006-16 LWW-Govindan-Review December 12, 2011 18:55

Chapter 16•Design and Analysis of Clinical Trials 193

treatment assignment can reduce bias introduced in the process of treat-

ment, whereas randomization reduces bias in patient selection. Random-

ized assignment to treatment can be used with or without blinding.

Answer 16.13. The answer is A.

Randomization is not necessary for patients with a disease that will uni-

formly and rapidly progress to an end-stage. In such a setting of homo-

geneity, a historical control usually is sufficient.

Answer 16.14. The answer is D.

It is true that meta-analysis can combine the results of small randomized

studies and may address questions that cannot be answered by any indi-

vidual study alone. However, by no means can meta-analyses replace

carefully designed and adequately powered randomized trials. This is

because different trials can be heterogeneous in terms of diagnostic and

staging procedures, supportive care, and methods of patient evaluation

and follow-up. Such heterogeneity may obscure small-to-moderate ther-

apeutic effects and undermine the ground for any pooled analyses.

Answer 16.15. The answer is D.

This claim may be true for bioequivalence studies but not for therapeutic

equivalence trials. In the former, for example, the objective is to show

biological equivalence of two proprietary preparations of a drug, and a

relatively large type I error or large tolerable difference () is acceptable.

For trials targeting equivalence in therapeutic effect (e.g., mortality rate),

however, usually a much smalleris of interest, and this will require a

very large sample size.

Answer 16.16. The answer is D.

The trial has shown that 0 is a plausible value for the difference but not

that the difference equals zero.

Answer 16.17. The answer is B.

Because Bayesian methods can incorporate information from preclinical

studies and sources outside of the trial, they are appealing in planning

Phase I and II trials. In these trials, information regarding the experi-

mental treatment is sparse, and study design is based on various (usually

subjective) assumptions. However, Phase III trials intend to provide reli-

able and objective guidance for decision making in a given disease, so the

subjective nature of prior distributions in Bayesian methods limits their

popularity in planning Phase III trials.

Answer 16.18. The answer is B.

Omitting patients who die or withdraw without completing the study is

a serious source of bias, as is dropping patients for noncompliance or

deviation from protocol. These may be patients for whom the treatment

does not work or has serious adverse consequences. There is no sound