THE SEQUENTIAL PROBABILITY RATIO TEST APPLIED TO THE DESIGN OF CLINICAL EXPERIMENTS*

Journal/Book: SPECIAL ARTICLE Reprinted from the New England Journal of Medicine 256:498-505 (March 14) 1957. 1957;

Abstract: BOSTON *From the Biophysics Research Laboratory of the Department of Medicine Harvard Medical School and Peter Bent Brigham Hospital and the Department of Biostatistics Harvard School of Public Health. Supported by Contract No. NR-119-227 between Harvard University and the Office of Naval Research and by a grant-in-aid from the Lasdon Foundation. Â† Fellow of the Rockefeller Foundation in Mathematical Biology Biophysics Research Laboratory of the Department of Medicine Harvard Medical School and Peter Bent Brigham Hospital; associate an the Associate Staff Peter Bent Brigham Hospital. I DEALLY any experiment should be terminated as soon as sufficient data are at hand to warrant the acceptance or rejection of the proposed hypothesis at an acceptable level of statistical significance. Unfortunately the most commonly employed statistical tests fail to predict the expected number of observations before the experiment is begun or to indicate the point of termination once the experiment is in progress. The advantage of a statistical design that simultaneously accomplishes both ends and also gives this information for a specified degree of statistical acceptability is obvious. Such a procedure for certain classes of experiments clinical and otherwise may be obtained from the theory of "sequential analysis." In a recent investigation1 a new experimental design based an this type of analysis was applied to a study of the difference in serum zinc concentrations between "normal" persons and patients with "severe" and "mild" Laënnec cirrhosis. After only 9 observations it was possible to conclude that patients with severe cirrhosis have a mean serum zinc level at least two standard deviations less than the normal. Classic statistical methods confirmed this result after a total of 25 observations. This new technic gives promise of wide application to clinical studies especially in cases in which a single observable is to be compared with a specified normal population. In this paper an exposition of this new design is given. The sequential aspects of the procedure are based an Wald's2 "sequential probability ratio test" (S. P. R. T.). ... ___MH

Zurück | Weiter