Randomized trials and allocation concealment

A great example of problems with a randomization scheme for a clinical trial - "Reliable evidence is the best medicine," Howard Brody, The Grand Rapids Press


More on allocation concealment:

- Schulz KF, Grimes DA. Allocation concealment in randomised trials: defending against deciphering. Lancet. 2002 Feb 16;359(9306):614-8.

- Schulz KF. Assessing allocation concealment and blinding in randomised controlled trials: why bother?, Evidence-Based Medicine 2000; 5:36-38

- CONSORT statement: allocation concealment example

-Forder PM, Gebski VJ, Keech AC. Allocation concealment and blinding: when ignorance is bliss. MJA 2005; 182 (2): 87-89

- Beller EM, Gebski V, Keech AC. Randomisation in clinical trials. MJA 2002 177 (10): 565-567.

and a study exploring how allocation concealment is reported in clinical trial reports: Hewitt C, Hahn S, Torgerson DJ, Watson J, Bland JM. Adequacy and reporting of allocation concealment: review of recent trials published in four general medical journals. BMJ. 2005 May 7;330(7499):1057-8.

Picking "random" numbers

Often in a report of a randomized clinical trial, you see description in the methods section of the paper indicating a bit more detail than just saying "patients were randomly assigned to treatment or control." Due to recognized (and unrecognized), internal (and possibly external) potential sources of bias, it's virtually impossible for a human to truly randomly pick numbers.

Dave of Cognitive Daily conducted an experiment earlier this month that illustrates this point very well -- he asked readers of his blog to pick any number between 1 and 20, and tabulated the responses. He reports the results in this post. Charting 347 responses from humans against 347 random picks made by a computer, it is clear that the distribution of the human selections are not randomly distributed -- humans were much more likely to select the number 17, far beyond what would be expected if the distribution was truly random.

(he also sums up reader comments on the experiment in this post and makes some inferences about why the number selection wasn't random)

So if we can't pick numbers randomly, how do researchers get random numbers for use in projects like randomized controlled trials?

There are a number of sources, including computer-based random number generators (e.g. the True Random Number Service, Microsoft Excel's random number generator) or a table of random numbers (e.g. the table on this page from the Partnership for Kentucky Schools -- the instructions provide a walk-through example for those who are interested in doing a little exercise about how you really use these numbers to randomly assign people to groups).

Researchers use these kinds of tools to randomize patients, and typically provide some explanation of their process for randomization in the Methods section, as noted above (e.g. the randomization section of the methodology in this article by Taha et al. in PLoS Clinical Trials).

Randomized experiments in social research

MDRC has released a guidance statement, "The core analytics of randomized experiments for social research," that may be of interest to those thinking about evaluation of programs/services in the health sciences. The abstract notes, "Its goal is to provide a compact discussion for faculty members, graduate students, and applied researchers of the design and analysis of randomized experiments for measuring the impacts of social or educational interventions."

The paper covers both design issues (e.g. selecting a study sample, randomizing particpants) and analysis issues (e.g. estimating impact in the presence of participant noncompliance, estimating impacts on randomized groups rather than individuals).

The document is part of a broader initiative by the MDRC to aid evaluation efforts: "Launched in 1999, the Initiative is exploring the use of quantitative and qualitative methods in the context of experimental, quasiexperimental and mixed research designs to rigorously study how programs are implemented, the impacts they produce, and how their implementation affects their impacts." Other papers in the series are also available.