Randomized controlled trials

Navigation:  Statistical data > Statistics in Medical Research > Conduct and reporting of medical research >

Randomized controlled trials

Previous pageReturn to chapter overviewNext page

The randomized controlled trial (RCT) is generally accepted as the preferred approach to conducting a wide variety of medical trials and is a central technique in the broader field of evidence-based medicine. The advantages of the approach are many, but in particular they have been shown to be very effective in controlling for selection bias and confounding, which other procedures may fail to achieve. In order to explain the central ideas behind RCTs, we start by providing a brief description below of the steps involved in the very first RCT, which was designed to determine the effectiveness of a new treatment for tuberculosis (TB).

The first step when designing an RCT is to carefully define the problem to be studied. Ideally this definition should be easily understood and as narrow as is practical. In the British Medical Research Council's (MRC) study of tuberculosis treatments in the late 1940s, the trial (of the effectiveness of streptomycin as a treatment) was defined by restricting it to: "acute progressive bilateral pulmonary tuberculosis of presumably recent origin, bacteriologically proved, unsuitable for collapse therapy, age group 15 to 25 (later extended to 30)" [BH1; Ch.20, and [MRC1]). This particular trial was the first truly randomized clinical trial and this aspect of the trial was devised by Austin Bradford Hill - he himself had been diagnosed with TB and spent two years in hospital and a further two years convalescing from the disease in his early 20s.

The second step, in the most commonly applied form of RCT, is to consider two distinct groups, ideally of approximately equal size and make-up. In the tuberculosis trial 55 patients were given the new treatment (Streptomycin plus bed rest) whilst a separate group of 52 patients were just given bed rest. One group is the treatment group (or intervention group) whilst the other is the non-treatment group or control group. By non-treatment we mean that this group receives a placebo, or no treatment, or continues on an existing, established treatment programme - the approach chosen must be precisely defined.

The patients selected to be involved in the trial are then randomly allocated to one of the two groups and their progress monitored over a period of time (usually relatively short). Most RCTs now carried out follow this general approach. There are other variants, notably cross-over RCTs, where patients are randomly assigned to groups but are then randomly re-assigned so they receive a sequence of treatments or non-treatments.

In the case of the 1948 tuberculosis trial the results after 6 months were as follows:

 

Streptomycin

Control

Considerable improvement        

28 (51%)

4 (8%)

Lesser improvement or deterioration

23 (42%)

34 (65%)

Deaths        

4 (7%)

14 (27%)

This finding was regarded as being an important breakthrough, both in terms of the success of the treatment, which was both clear to see and statistically significant (i.e. is extremely unlikely to have occurred by chance). After 12 months a further 8 of the treated patients had died, and a further 10 of the control group, again a significant result, but also indicating a reduced efficacy of the treatment over time. None of the patients, however, could be regarded as having been cured. Interestingly enough, at the time the authors did not report why they chose the sample sizes used. In a remarkable recorded interview in 1990 Bradford Hill, then aged 93, stated that the main reason for choosing 50 or so patients was that this was as much Streptomycin as could be obtained from the USA at the time given its scarcity, high cost and the considerable problems in obtaining US currency in the UK in the immediate post-WWII period. He also stated that the patients were not informed about the treatment they were to receive. The authors also do not describe what form of statistical analysis was performed, merely that the results were statistically significant - the impression one has from Bradford Hill's book is that simple chi-square tests were carried out.

In order to carry out the random assignment of patients to groups, researchers originally used simple random selection by patient number, possibly stratified into blocks (e.g. by sex and by age grouping) but this can lead to problems. An example is given by Cancer Research UK:

"it is possible to be biased without realizing it. For example, if a new treatment has quite bad side effects, the doctors running the trial might subconsciously avoid putting sicker patients into the group having the new treatment. So as the trial went on, the control group would have more and more of the sickest patients in it. The people in the new treatment group would then do better than the control group. So, when the trial results come out, the new treatment would [incorrectly] look as if it works better than the standard treatment."

To avoid such problems a system known as blinding is applied. In blind trials one or more parties are unaware of the assignment of treatments. For example, in a so-called single-blind trial the individual receiving the treatment is not made aware whether the treatment they are being given is an existing treatment, a new treatment or a placebo (i.e. a tablet or preparation that is not a drug at all and has no effect on the patient). In double-blind trials neither the experimenter nor the patient know which treatment has been assigned to which patient (treatments are coded) thereby minimizing the risk of any influence the experimenter may have on the experiment. Unfortunately this terminology is not uniformly applied, leading to current recommendations to describe in detail the kind of blinding applied (if any), even extending to those involved in analysis and interpretation of the results.

Whilst the RCT procedure described above appears simple and straightforward, it does have numerous difficulties. Determining the appropriate sample size to use can be problematic, especially in the case of rarer conditions and/or where suitable triallists are simply not available in the location or at the period of time required (see the earlier discussion on sample size, and in particular, the Salk Polio vaccine trials of 1954). The cost of conducting such trials may be high, and if the trial is interrupted during its structured program, or there are problems with adverse effects or triallists dropping out, how are the results to be interpreted? Some treatments are not amenable to such form of trial, for example those requiring specific actions or exercises by participants. Sample size determination can be difficult and it may be impossible to obtain the desired numbers in each group and stratum or the required sample size if far larger than is achievable within time and cost constraints. There are also often ethical problems - for example, should a particular treatment that is thought to be very promising be withheld from very sick patients? (see further, the Declaration of Helsinki). Finally, recent research has raised questions over the entire notion of placebo usage, even in double-blind RCTs. There is increasing evidence to suggest that some treatments only work or are only effective if the recipient believes they are receiving an effective drug, i.e. there is a complex relationship between treatments administered and the way in which the body responds to such treatments. There is also widely reported evidence of so-called nocebo effects, in which individuals experience adverse reactions to placebos.

References

[BH1] Bradford Hill A (1937) Principles of Medical Statistics. The Lancet, London (issued in various editions until 1971. Then republished as "A Short Textbook of Medical Statistics" in 1977

[MRC1] Medical Research Council (1948) Streptomycin treatment of pulmonary tuberculosis. BMJ, 4582, 769-782

Wikipedia: Randomized controlled trials: http://en.wikipedia.org/wiki/Randomized_controlled_trial