Random assignment to the treatment groups is more likely to achieve equalization if the size of the sample is adequate, the level of impairment or disability and the natural history of prognosis are similar, dropout is a random phenomenon, and sequential entry over a long time does not alter the prognosis as additional interventions evolve. One of potentially confounding problems of using a group-comparison approach is that facilities may not have a sufficient number of homogeneous patients who meet entry criteria and who are willing to participate. Also, sampling or selection bias may be present when subjects are recruited at a single hospital or clinic. This error may limit the generalizability of the results to other patients. Such bias is especially likely to profoundly affect a descriptive and quasi-experimental design.
In rehabilitation trials of physical and cognitive interventions, disease pathologies with a resulting range of impairments are associated with disabilities and handicaps that are equally heterogeneous. All four variables are subject to potentially unforseen interactions. In addition, the focus of a rehabilitation trial could be at the level of pathologic process, impairment, disability, or handicap. Outcome measures may fall within any one or more of these levels. If these 4 foci of a treament intervention are the rows and the same 4 foci of outcome assessment are the columns, a total of 16 possible strategies for rehabilitation research are apparent.245 For example, an intervention at the level of an impairment such as leg weakness in a hemiparetic patient could alter an outcome at the level of impairment (improved strength), disability (walks faster), or handicap (new ability to walk up stairs, so no further need for an elevator). If an intervention alters an impairment by improving leg strength only, but does not lessen disability (walking is still assisted or very slow), the intervention will hold little interest for rehabilitationists. On the other hand, if the outcome measure is a decrease in the number of falls related to strengthening, the outcome may be important, if that outcome measure is chosen a priori. Of course, an in tervention can simultaneously be at more than one level or can affect more than one level of outcome measurement. Taking into account all of these potential relationships can produce enough wobble in a trial to flaw all but the best designs.
The control intervention, when a placebo, is often said to have an effect on up to one-third of patients in a drug, physical, or psychological trial. With any hands-on intervention, the control group should receive some defined level of attention from the therapists that is similar to the attention given to the experimental group. The notion of a large placebo-induced degree of change may lead to the mistaken concern that no experimental intervention other than a very robust one can improve outcomes in a clinical trial. The placebo effect, however, is insignificant in studies with binary outcomes and has only a small benefit in studies with continuous subjective outcomes, especially in pain studies.246
Was this article helpful?