 |
 |
Mark Newman writes: >I am about to conduct a (small) RCT of a continuing professional >educational intervention with nurses comparing problem based >learning with standard small group learning. > >Our proposed method of randomization was to use a computerized >table of random numbers. However the practicalities of the >situation mean we won't know the final list of participants really >until the course starts and those participants wo register prior to >the beginning of the course need to know which days they need to >obtain as study leave (it is a part time course) as early as possible >in order to make the necessary cover arrangements (the different >groups are 'taught' on different days to avoid contamination >between groups) . So basically we need to randomly allocate >people as soon as they register. > >q.1 - I am thinking about the best way to do this - is it by just >tossing a coin for each individual ? Assign a pseudo-code to each participant: 101, 102, 103, etc. When you recruit a subject, then match their employee id or some other identifier to the pseudo-code. Make sure that you check out all of the inclusion and exclusion criteria before you link an employee id to the pseudo-code, and this will insure concealment of the assignment. If you or the person doing the recruiting might be tempted to peek at the pseudo-code before deciding whether to include the subject, then you might need to use sealed envelopes, or dial in access to a research co-ordinator. Will the pre-registered subjects know which course they are getting, and will they care? If so, I would watch drop-outs very carefully. By the way, the actual mechanics of using random numbers is not obvious to everyone, because there are several ways to do this. The way I recommend is to list your treatments in a systematic order (ABAB...) and then attach a random number to each. Then sort by the random number. This effectively shuffles the treatments into a random order. If you are comfortable with any basic computer spreadsheet, you can do this in a matter of minutes. >q.2 When critically appraising an RCT we ask whether the control >and intervention groups were similar at the start of the trial. Now I >am looking at the issue from the other end I am wondering how I >can 'make sure' that the groups are similar at the start of the trial. >It is after all possible that one group could end up much larger than > the other if the coin tossing method is used. You are confusing two issues here. Randomization, by itself, will insure that the two groups are comparable, unless the sample size is terribly small. A group of sixty subjects can be comparable to another group of forty subjects, as long as they have the same mix of ages, gender, race/ethnicity, and so forth. If the two groups do not have exactly the same number of subjects, then you might see a small loss in power (and in your situation may also raise some logistical issues). Unless your coin is loaded, the loss in power is likely to be so small that it is not worth worrying about. If you use the random shuffling mentioned above, you are guaranteed to have a perfect balance as long as you recruit exactly the number of subjects you think you will. If there is a chance that you will recruit fewer subjects than what your randomization table has listed, then you might see a small imbalance in sample sizes. Still, there is a simple technique called block randomization that guarantees almost a perfect balance, under all scenarios. You randomize within each block of 4, 6, or 10 patients. So if you recruit all the way to the end of a block, you are guaranteed perfect balance. If you recruit partway through a block, then you might have a small imbalance, but it is tightly constrained by the size of the block. >(Hmmm now I have written this I think I have a better answer - but I >would like to hear your suggestions anyway) Some of this is just common sense, but the logistics of research is often tedious and tricky. I have a web page that gives more details and examples on how to randomize. Go to my main page and look for the section on planning a research study. Steve Simon, [log in to unmask], Standard Disclaimer. STATS - Steve's Attempt to Teach Statistics: http://www.cmh.edu/stats %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%