HYPOTHESIS TESTING IN CLINICAL TRIALS

https://doi.org/10.1016/S0889-8588(05)70311-7Get rights and content

In designing and analyzing any clinical study, two issues related to patient heterogeneity must be remembered: the effect of chance, and the effect of bias. These issues are addressed by (1) having adequate numbers of patients in the study, and (2) randomizing treatment assignment. Adequate sample size allows the control of the probability of a false-positive result and the probability of a false-negative result. Randomization avoids bias (whether conscious or unconscious); with randomized allocation, predictive factors (known and unknown) tend on average to be balanced between treatment groups.3, 8, 11 Also, randomization provides a valid basis for statistical tests of significance.3 These concepts are discussed further later.

Section snippets

HYPOTHESIS TESTING IN A COMPARATIVE CLINICAL TRIAL

In a clinical trial, the patients (who form a sample) are selected from a conceptual population of all patients who could have entered the trial (sometimes called the source population). In a comparative trial, with two or more treatment groups, each treatment group represents such a sample. The goal is to compare the groups with respect to some outcome measure. Ideally, investigators want to determine what the difference in outcome would be if all patients in the source population were treated

SAMPLE SIZE AND STATISTICAL POWER

In designing a clinical trial, one must plan for an adequate number of patients (sample size). Increasing the sample size will increase the precision of the estimate of treatment effect. Expressed another way, increasing the sample size decreases the variability (decreases the standard error) of the estimate, thus producing narrower confidence intervals.

In hypothesis-testing framework, investigators are interested in selecting the sample size to achieve adequate statistical power.7 This

SUBGROUP ANALYSIS AND INTERACTIONS

Following an analysis of all randomly allocated patients in a clinical trial, there is often interest in investigating whether treatment effects differ by patient subgroup. Concern must be raised, however, about multiple subset-specific analyses, particularly when these are data-derived. Investigators should consider whether it is reasonable to expect that the actual treatment effect may differ in a meaningful way between different subgroups. The danger is finding spurious subset differences

FACTORIAL DESIGNS

Factorial designs in clinical trials are sometimes appropriate and can lead to efficiencies by answering more than one question (addressing more than one comparison of interventions) in a single trial.6 The simplest design is the balanced 2 x 2 factorial, addressing two treatment comparisons: A versus not-A, and B versus not-B. Conceptually, patients are first randomized to A or not-A, then randomized also to B or not-B. In effect, equal numbers of patients are randomly allocated to one of four

GROUP-RANDOMIZED TRIALS

Group-randomized trials (sometimes also called cluster randomization trials) randomly allocate intact groups (clusters), rather than individuals, to intervention. Units of group randomization include communities, small towns or villages, factories (workplaces), schools or classrooms, religious institutions, chapters of social organizations, families, and clinical practices.9

Sample size calculations for group-randomized trials need to consider the extra source of variation resulting from the

References (20)

There are more references available in the full text version of this article.

Address reprint requests to Sylvan B. Green, MD, Department of Epidemiology and Biostatistics, Case Western Reserve University, School of Medicine, W-G57, 10900 Euclid Avenue, Cleveland, OH 44106–4945

*

Department of Epidemiology and Biostatistics, School of Medicine, Case Western Reserve University, Cleveland, Ohio

View full text
Copyright © 2000 W. B. Saunders Company. Published by Elsevier Inc. All rights reserved.