An analysis of 215 OA clinical trials with more than 40,000 participants indicated that 75% of pain reduction could be attributed to “contextual effects,” including patient baseline status, expectations, beliefs and compliance; clinician behaviors, such as healing rituals, verbal suggestions and therapeutic touch; attractiveness of a treatment site; and many other effects.3 Placebo responses in OA are highest for invasive procedures, such as joint injections, intermediate for pills and least for topical agents.3 One surgical trial reported similar modest responses of OA patients to arthroscopic lavage, debridement and sham surgery; differences were neither statistically nor clinically significant.9
The importance of placebo responses highlights recognition of more general limitations of clinical trials, which often are ignored in medical education and communication, based on a belief that clinical trials invariably provide the best evidence concerning a therapy compared with another therapy or placebo.10 Randomized, double-blind, controlled clinical trials remain the optimal method to analyze the efficacy and safety of a therapy, mimicking a laboratory reductionist experiment to focus on a single variable—an active treatment vs. a control—while attempting to keep all other conditions identical.11 That goal is met most effectively in short-term trials of an external pathogen, but is less achievable in longer trials involving chronic diseases.12,13
Two early clinical trials in cardiovascular disease illustrate complexities in interpretation of placebo results. A trial to analyze clofibrate (to reduce cholesterol) vs. placebo to prevent death from cardiovascular disease indicated 15% five-year mortality in greater than 80% of patients who took clofibrate vs. 25% in those who did not, but also 15% mortality in greater than 80% of patients who took placebo vs. 28% in those who did not.14 Therefore, much better (and similar) results were seen in compliant vs. non-compliant patients, regardless of allocation to active vs. placebo treatment.14
In another clinical trial over three years to prevent cardiovascular death after acute myocardial infarction by a beta blocker or placebo (BHAT study), mortality was 9% in patients randomized to a beta blocker vs. 13% in those randomized to placebo, a statistically significant difference.15 However, in secondary analyses, mortality was 5%, 7% and 13% in patients with more than 12, 10–12 and fewer than 10 years of formal education, regardless of treatment allocation.15 These results were further explained by life stress and social isolation; mortality was 15% in patients with high levels of both, 7% with either and 2% with neither. The data indicate again that patient characteristics and behaviors were far more significant to outcomes than allocation to active treatment vs. placebo.15
The Limitations
Important pragmatic limitations are seen in clinical trials involving chronic rheumatic diseases.12 One limitation is that inclusion and exclusion criteria usually restrict the number of participating patients to a small fraction of those with a given diagnosis; for example, at one site, only 5% of rheumatoid arthritis (RA) patients were eligible for the first study of infliximab in RA.16 Another limitation is that short-term surrogate measures may not predict long-term undesirable outcomes (e.g., a tender joint count is a poor predictor of long-term work disability and premature death in RA).17