Estimating Treatment Effects in Clinical Trials Subject to Regression to the Mean

作者：Stephen John Senn, Richard A. Brown, Kenneth E. James · 发表于：Biometrics · 年份：1985 · DOI：10.2307/2530881 · 被引用次数：48 · 研究领域：Statistical Methods in Clinical Trials

If patients are selected for a clinical trial because they have been measured as having high or low values, as the case may be, of some variable of interest (e.g., blood pressure) and if the correlation between first and subsequent measurements is less than one, then it is to be expected that if as a group they are measured again their average measurement value will have regressed to the mean. This well-known phenomenon of regression to the mean thus presents a problem for the evaluation of uncontrolled clinical trials and, clearly, may cause difficulties in other fields of research as well. In a paper on this subject, James (1973) proposes a method of estimating the regression effect using moments and is able to obtain estimators of a fairly simple form, both for the regression effect and for various parameters of interest, by assuming that the proportion of the population selected for treatment is known. James (1973) assumes that in the absence of any treatment and in the absence of any selection procedure, first measurements X and second measurements Ywill have a bivariate normal distribution with