According to the author, “this book is intended not only to make the method of sequential trials understood by all researchers who would utilize it but also to make the method as convenient as possible” (p. 23). The primary audience is physicians and pharmacologists who carry out screening trials and clinical evaluations of new treatments. The book teaches them what sequential trials are and how easy it is to use them. To make it even easier, BASIC programs are listed for each of the 16 different designs of sequential trials discussed in the book. Since the programs contain between 70 and 140 lines each, it is not difficult to choose one and type it into a PC. It is also possible to order a diskette containing all of the programs from the author.
A secondary audience is programmers, epidemiologists, and statisticians who work on medical trials. In fact, the book could be used as a text for a university course on sequential trials.
Sequential trials are performed as follows. A pair of subjects is selected and one of them is randomly assigned to treatment A; the other one receives treatment B. The response of each subject to the treatment is recorded. If the response is dichotomous (for example, “success” or “failure”), it is called “qualitative.” If it is a number measured on some scale (such as blood pressure or the amount of cholesterol in blood), it is called “quantitative.” Then another pair is selected and treated and their responses are recorded. The selection, treatment, and recording of the response from each pair is called a trial. After each trial, a single number (called a “statistic”) is computed based on all of the responses recorded so far. If this statistic is in a certain region, the experiment stops and the conclusion is that treatment A is better than treatment B. If the statistic is in another region, the experiment stops and the conclusion is that treatment A is not better than treatment B. If the statistic does not fall in either of these two regions, another pair is selected and the experiment continues. This approach is called a one-sided experiment. A two-sided experiment has three regions for stopping the experiment: “A is better than B,” “B is better than A,” and “there is no difference between A and B.” If the possibility exists (even if it is small) that the experiment could continue for an indefinite number of trials, the experiment is called “open.” If an upper limit to the number of trials is built into the experiment, it is called “closed.”
The programs in the book draw a graph containing the accumulative statistic and the boundaries of each region as a function of the trial number. In most cases, the boundaries are linear and the statistic is a step function.
For each of the 16 trial designs discussed in the book, the author gives the following information: when to use it (and when not to use it); the statistic to be used; the development of the mathematical formulae for the boundaries of the regions; the average number of trials required; and at least one example taken from the medical literature. Each example is run on the appropriate program, and the output appears after the listing of the program.
The book is well written. The illustrations are of high quality. Few typographical errors appear. It is a pleasure to recommend this book for all those who perform screening or clinical trials.