A review of several pain clinical trials did not support claims of a bimodal distribution of changes in pain intensity from baseline in response to efficacious medications, and found that the inaccurate appearance of bimodality was due to the use of inappropriate statistical methods in the original studies, according to a report published in Pain.
An assumption of bimodality regarding pain patients’ response to therapy stems from research that appears to demonstrate dual distributions of “responders” and “non-responders” in the patient population, with relatively few individuals falling between the two in terms of change in pain from baseline. As a consequence, many clinicians and researchers have called for changes in the design and analysis of such trials, including evaluation of treatment effects via “responder” analyses – which are often less efficient and require larger samples – rather than commonly utilized between-group mean comparisons. The investigators reviewed 4 clinical trials, focusing on the statistical methodologies used, to determine whether the claim regarding distribution bimodality and the push for changes were legitimate.
Data from 4 randomized pain clinical trials were analyzed, 2 each investigating pregabalin and duloxetine for the treatment of chronic neuropathic and musculoskeletal pain. The pregabalin studies concerned treatment of diabetic peripheral neuropathy or postherpetic neuralgia, while the duloxetine studies involved patients with chronic low back pain or osteoarthritis. After review and replication of the original statistical methodologies used by the principal researchers in each study, investigators re-analyzed the data while making several procedural adjustments that they hoped would clarify whether the patients’ reported changes from baseline were truly bimodally distributed.
The overarching issue that appeared to be most responsible for the incorrect bimodality claim was that the original histograms were constructed improperly, with unequal bin widths that were wider at the extremes. In addition, 2 other flaws were deemed contributory: the use of percent change as the primary metric and performance of missing data imputation via the maligned baseline observation carried forward method.
After correction of the apparent methodologic errors, including the use of absolute change instead of percent change, missing data entry via control-based multiple imputation and construction of proper histograms with equal bin widths, the distribution of pain intensity change from baseline was considerably more unimodal.
Study strengths included the ability to closely replicate the original bimodal findings using similar statistical strategies. Study limitations included using data from only 4 trials focusing on 2 interventions.
“While our findings neither support nor refute the hypothesis that distinct populations of “responders” and “non-responders” to pain interventions exist, the analyses presented in earlier work do not provide support for this hypothesis, nor for the recommendation that pain clinical trials prioritize “responder” analyses,” noted the investigators.
Disclosure: Several study authors declared affiliations with the pharmaceutical industry. Please see the original reference for a full list of authors’ disclosures.
Mbowe OB, Gewandter JS, Turk DC, Dworkin RH, Mcdermott MP. Are there really only two kinds of people in the world? Evaluating the distribution of change from baseline in pain clinical trials [published online September 27, 2019]. Pain. doi:10.1097/j.pain.0000000000001708