John DiNardo 141researchers in this area routinely make no adjustment of any sort for the high rates
of attrition in studies looking at chronic headache pain.
An illustrative exception to non-severe testing involves, not a test of MOH, but
rather a study of the efficacy of sustained opioid therapy – opioids being considered
a particularly pernicious cause of MOH (Saper and Lake, 2006a, 2006b). Although
Saperet al.(2004) had no control group, the researchers treated individuals who
dropped out foranyreason, died for non-opioid related reasons, were suspected of
“cheating” (using more opioids than allowed by the doctors), and so on, astreat-
ment failures. This also included some patients who reported a substantial improve-
ment but were considered to have “failed” to satisfy theresearchers’definition of a
significant reduction in functional impairment. In defining treatment failure more
broadly, the researchers were essentially using a “worst case” bound.^59 While the
use of “worst case bounds” is infrequent (or nonexistent) in the MOH literature,
the argument for doing so has validity: it is entirely consistent with the notion
of “severe testing.” Indeed, leading researchers in MOH are aware of the potential
value of such bounds. Saper and Lake (2006b), for example, harshly criticize a meta-
analysis of RCTs on the efficacy of opioids for non-cancer pain for failure to adhere
to “intent-to-treat” principles. In this instance, this meant treating as failures those
individuals who began opioid treatment but then stopped foranyreason.^60
The severity of the tests to which opioid efficacy has been confronted is in sharp
contrast to extant studies of MOH (sometimes by the same researchers), where
a failure of a patient to reduce his medications is not treated as a failure of MOH
therapy. Indeed, where attrition rates of 40% or higher are common, were the litera-
ture to treat those who were unwilling or unable to abstain from the offending med-
ication as failures of “MOH therapy,” it would appear that few, if any, of the studies
in Zedet al.(1999) that purport to provide evidence favorable to the existence of
MOH would continue to do so. Indeed, although plagued by non-random attrition
and written by advocates of MOH, it has been observed that patients with MOH
who “lapse” and re-establish medication overuse have higher measured “quality
of life” on average than those with MOH who do not lapse (Piniet al., 2001).
It might fairly be argued that an intelligent Bayesian might not have moved
his/her posterior much in light of the foregoing discussion. Moreover, it is certainly
the case that no formal Bayesian analysis has been employed. At least superfi-
cially, the “usual” statistical analysis was employed. What this literaturedoesn’tdo,
however, is:
- test the theory in such a way that the observed result would be unlikely if the
obvious alternative (the one “favored by patients”) was true – that it is chronic
pain that causes use of pain relieving medication, not the other way around - employ the “usual” techniques to make tests more “severe” – the failure to
use worst-case bounds, for instance, to deal with the problem of non-random
attrition - react to each threat to the theory as potential reason to abandon the theory.
Instead, the reaction of researchers has been continued modification of the
theory until it is no longer capable of being refuted by evidence.