Analysis of clinical data with breached blindness by Shein‐Chung Chow and Jun Shao, Statistics in Medicine 2004; 23 :1185–1193

In their recent paper, Chow and Shao [1] proposed a method for analysing clinical data with breached blindness, claiming that bias caused by knowledge of the identity of the treatment can seriously distort statistical inference on therapeutic eeects. Thus, they argued that adjustments to statistical analyses should be made when the integrity of blinding is doubtful. However, if blinding was a fundamentally essential requirement of the validity of studies, it would have dramatic eeects on medical research since no meaningful studies could be carried out on cigarette smoking, rare side-eeects of drugs, surgery, etc. The validity of Chow and Shao's argument is thus highly important. First, identiÿcation of treatment by subjective observation should not be considered merely a nuisance, because in many cases the unambiguous purpose of physicians is to reduce the subjective symptoms of patients. At the individual level, such eeects can be investigated using the 'N = 1' type of trial in which breaching of blindness often is an important explicit outcome [2, 3]. Second, there is no valid evidence indicating that the so-called placebo eeect is large and omnipresent. A recent meta-analysis of trials comparing placebo and no-treatment groups found no placebo eeect in studies that measured binary outcomes [4, 5]. In studies measuring continuous outcomes, only those that measured pain found evidence of a placebo eeect, but it was quite small. Chow and Shao disregarded these negative empirical ÿndings in arguing that in general any knowledge of treatment may seriously distort statistical inference on the treatment eeect. Chow and Shao brieey discussed two old trials as examples of unreliable results caused by breaching of blindness. The trial by Brownell and Stunkard focused on reducing weight in obese women using an appetite suppressant [6]. Since dosage was adjusted according to reports of side-eeects, many study participants could obviously infer their treatment correctly [6]. Chow and Shao speculated that the diierence between the trial groups in loss of weight might have been caused by the breached blindness, i.e. by the placebo eeect [1]. However, to evaluate the eeect of believing, Brownell and Stunkard compared patients who believed they were taking the drug with those who believed they were taking the placebo. These two groups did not diier, which is inconsistent with the placebo eeect explanation. Furthermore, pooling the results of eight trials comparing a placebo group to a no-treatment group found no evidence that placebo would aaect obesity …