THE SO-CALLED PLACEBO-EFFECT IN BENIGN PROSTATIC HYPERPLASIA TREATMENT TRIALS REPRESENTS PARTIALLY A CONDITIONAL REGRESSION TO THE MEAN INDUCED BY CENSORING

Objectives. To study the variability of assessment instruments (sympto m questionnaires and flow rate recordings) in healthy volunteers durin g repeat administration in short intervals. To study the effect of inc lusion criteria-based censoring of patients during screening for benig n prostatic hyperplasia (BPH) treatment trials on the outcome of subse quent tests. Methods. One hundred forty-five male volunteers without k nown prostatic diseases with a mean age of 52 years (range 23 to 83) w ere given the American Urological Association (AUA) Symptom Index (SI) , BPH Impact Index (II). Quality of Life (QOL) score, and a flow rate recording twice 10 to 20 days apart without any healthcare interventio n. Data were collected and analyzed after typical BPH trial criteria w ere applied to the first test, and patients who did not ''qualify'' we re censored. Results. Good correlation exists between two closely spac ed administrations of the AUA SI, BPH II, QOL score, and flow rate rec ordings in healthy male volunteers with correlation coefficients betwe en 0.73 and 0.89. Censoring patients and excluding them from the analy sis of the second test based on typical BPH trial criteria induces a r egression to the mean phenomenon, which results in an artificial impro vement in these outcome parameters. The magnitude of the improvement i ncreases as the selection and censoring criteria tighten. The mean dif ferences between the first and second test range from 1.4 to 1.7 mL/s for the peak flow rate, from -1.0 to -1.4 for the AUA SI, and from -0. 4 to -0.8 for the BPH II. All these differences induced solely by cens oring with resulting regression to the mean are statistically signific ant. Conclusions. Censoring of patients based on inclusion and exclusi on criteria is typical for BPH treatment trials. One of the under-reco gnized effects of censoring is a regression to the mean, which leads t o an apparent improvement in the outcome parameters assessed. In both placebo or sham-controlled trials, as well as in clinical series witho ut controls, one must keep this effect and its relative magnitude in m ind, and interpret the data from such trials with appropriate caution. (C) 1998, Elsevier Science Inc. All rights reserved.