Randomized controlled trials (RCTs) are justified by many considerationsâboth methodological and ethical. From a methodological standpoint, the assignment of subjects to treatment arms based on random criteria is designed to prevent systematic differences from influencing the outcomes; potential differences between the arms of the trial will be âwashed outâ via the random procedure. Similarly, the double-blind element is designed to prevent the subjective biases of the researcher from influencing the outcome.1 From an ethical standpoint, the use of the random element is tied up with broader ideas of fairness; just as in sporting events and political debates, a chance element (e.g., a flip of a coin) is used to ensure that no side gets an unfair advantage. Also, since the results from a randomized study are more credible than observational studies without randomization, the determination of safety and efficacy might be accomplished more quickly than through unsystematic observation; when interventions pose potentially significant health risks, this is clearly a socially beneficial outcome.
But problems still exist when researchers seek to operationalize these procedures in practice. In particular, if researchers can deduce the randomization procedure, then there is the possibility of still assigning subjects in ways that will bias the final outcome. Over the years, strategies for addressing this problem have evolved; however, the underlying rationale of preventing bias remains. In this chapter, I will examine the history of how bias and randomization have been dealt with to assess the âlessons learnedâ from past experience.
1.1Early Twentieth-Century Discussions of Bias and âRandomizationâ in Clinical Research
As early as the first decade of the twentieth century, there was overt discussion of bias and how it could be preventedâas illustrated by published discussions about anti-typhoid inoculation by the bacteriologist Sir Almroth Wright and the statistician Karl Pearson. In his 1904 book A Short Treatise on Anti-Typhoid Inoculation, Wright outlined how he believed the statistical arguments on this issue should be assessed. He claimed that there should be a control group âwhich ought to correspond with the inoculated group in all points save only in the circumstance of inoculation.â Also, he claimed that small errors in the data could be overlooked because a large number of observations would mean âall chance errors⌠are spontaneously eliminated.â However, he warned that, if an error always favored one conclusion over another, it would not necessarily be eliminated by the accumulation of a large number of observations.2 Although he did not use the term âbias,â he clearly acknowledged the possibility of systematic distortion when reporting statistical findings. However, he still claimed that his medical judgment could trump statistical rules: âThe plain everyday man will find it possible to reconcile the demands of his statistical conscience with the demands of his practical life. He will neglect the mint and anise and cummin of statistical criticism while holding fast to weightier principles of the statistical law.â2
Wrightâs methodological views were questioned by Karl Pearson. In a letter to an official in the British War Office, Pearson noted how the voluntary nature of inoculation could lead to distorted results. He commented that âthe regiment being divided for special duties, companies might very easily run different risks⌠May not⌠the more careful men have been inoculated, just as the more cautious are vaccinated? Thus the average correlation between inoculation and immunity might only mean a correlation between greater caution and immunity.â To prevent this, Pearson advocated an experimental protocol in which every alternative man in a regiment of 800 be inoculated so that âwe can exhibit your results in correlative form, showing a distinct relation between inoculation and immunity.â3 Even though the War Office recommended against such as alternation procedure (since inoculation was voluntary),4 Pearson did publish an article in the British Medical Journal in which he computed the correlation coefficient for this procedure so that it could be compared to other widely used procedures such as smallpox vaccination.5 What this exchange shows is that both Wright and Pearson acknowledged that chance and bias could play a role when data are collected; however, neither advocated for a method of completely ârandomâ assignment as a way to address the bias issue.
In the following decades, Pearsonâs recommendationâalternate allocation of patients to the experimental and control groupsâwould become a prominent feature of clinical trial design. As Iain Chalmers has noted, the approach was used in studies on plague and cholera in India in the first decade of the century. In the second and third decades of the century, alternate allocation was used in controlled trials of serum treatment for pneumonia in the United States.6
In later studies, the word ârandomâ was explicitly used to discuss subject assignment. However, it usually still referred to alternative allocation. Illustrative of this semantic ambiguity is the 1938 study reported by Diehl et al., which involved testing the efficacy of a vaccine for the common cold among college students at the University of Minnesota. The assignment procedure was described in the following manner:
At the beginning of each year of the study students were assigned at random and without selection to a control or to an experimental group. The students in the control groups were treated in exactly the same manner as those in the experimental groups but received placebos instead of vaccine. All students thought that they were receiving vaccine and so had an unprejudiced attitude toward the study. Even the physicians who saw the students at the health service when they contracted colds during the period of the study had no information as to which group they represented.7
While this description has a very âmodernâ sound to it (aside from the unethical practice of leaving all students under the impression that they were getting the vaccine), Armitage has documented that Diehl actually used alternate allocation when assigning the subjects. He had used alternation in earlier common cold trials, and in a speech from 1941, Diehl declared that âat the beginning of the [1938] study, students who volunteered to take these treatments were assigned alternatively and without selection to control groups and experimental groups.â8
1.2Austin Bradford Hill and the Streptomycin Trial in Treating Tuberculosis (1948)
By general consensus, the British statistician Austin Bradford Hill (1897â1991) is regarded as the âfatherâ of the truly randomized controlled trials. Before World War II, Hill supported alternation as a method of patient assignment. In the first edition of his 1937 textbook Principles of Medical Statistics, Hill wrote that alternative allocation was âoften satisfactoryâ because âwe can fairly rely upon this random allotment of patient to equalize in the two groups the distribution of other characteristics.â6 Reflecting on this recommendation over half a century later, Bradford Hill noted that he âwas trying to persuade the doctors to come into controlled trials in the very simplest form and I might have scared them off⌠I thought it would be better to get doctors to walk first, before I tried to get them to run.â9
In World War II, Hill would be provided with an opportunity to develop a new patient assignment mechanism when he was appointed to a committee that was going to administer a clinical trial of the drug streptomycin to treat tuberculosis. Streptomycin had been developed by researchers in the United States, and the preliminary evidence appeared to suggest that patients might improve when given the drug. To determine whether this improvement was related to receiving streptomycin, a clinical trial was launched in the United Kingdom by the Medical Research Council. Because of wartime shortages, many tuberculosis patients would still have had to go untreatedâwhether a controlled clinical trial was executed or not. This empirical fact allayed the ethical qualms of those conducting the trial on the issue of having a placebo control group. As D.D. Reid observed in 1950, âOur genteel poverty thus paid a scientific dividend by quieting any doubts we might have about the ethics of controlled trials of streptomycin in pulmonary tuberculosis.â10
Instead of alternate allocation, Hill used a randomization procedure. Each participating hospital was given a series of numbered envelopes (with a separate list for each gender), and each env...