Response: Prescription Opioid Use and the Risk of Disability
We would like to take this opportunity to respond to a recent Letter to the Editor submitted by Franklin and colleagues that referenced our systematic review, “Early prescription opioid use for musculoskeletal disorders and work outcomes: a systematic review of the literature” by Carnide et al.1
First, we wish to thank Franklin et al for bringing to our attention the error we made mislabelling the design of their study in our manuscript.2 In our efforts to be concise and reduce text, we inadvertently grouped all studies and labeled them as historical cohort studies. However, when conducting the review, we recognized that the Franklin and colleagues paper was a prospective cohort study. We sincerely apologize for this error and support the publishing of an erratum to reflect the fact that all studies in our systematic review were observational cohort studies.
We agree with their assessment that, given the limitations of conducting an randomized controlled trial on this topic, a prospective cohort study is certainly a strong observational design. We would like to clarify that our assessment of risk of bias was not dependent on whether the cohort study was historical or prospective in nature. That is, there was no penalization for a study being a historical cohort rather than a prospective cohort.
We also agree with Franklin and colleagues that one of the strengths of their study compared with others is that primary data were collected on a number of measures of relevance to both opioids and work disability, including pain intensity, function, injury severity, recovery expectations, and fear avoidance. However, as noted in our review, for a large number of workers, interviews were conducted following initiation of opioid prescribing and adjustment for these modifiable factors in the analyses may have led to an underestimation of the total effect of opioids on work disability. Further, as described in our review, other potential confounders have not yet been considered in this literature, including prior history of musculoskeletal symptoms, other comorbidities, preinjury opioid experience, and co-occurring health care. A number of other “indications” for opioids that could influence recovery have also not been addressed. This includes workplace and system-level factors, such as insurer preferences and employer support of the worker, as well as the role of the physician, who could influence not only treatment but also provide recommendations for return-to-work. Accordingly, our team considered the risk of confounding bias to be moderately high in this study.
We certainly concur with Franklin and colleagues that there is an inherent trade-off to primary data collection, where researchers have greater control over the information collected at the expense of potential selection bias due to nonresponse. Unfortunately, this does not negate the possibility of selection bias in this study. In their comparison of participants and nonparticipants, participants were found to be more likely to be on wage replacement benefits at 1 year and, thus, may have represented more severe cases.
Finally, in their letter, Franklin and colleagues expressed concern about our assessment of exposure measurement bias in relation to incomplete prescription capture. In (Franche R-L, Carnide N, Lee H, Hogg-Johnson S; 2007) unpublished data, they found that only 3% of workers with new workers’ compensation claims had a recent opioid prescription (previous 3 mo) and 1.5% demonstrated chronic preinjury opioid use. When completing our review, we did not have access to the data cited in their letter. Still, while preinjury patterns described suggest few prescriptions in the early postinjury period would likely result from preexisting conditions, the possibility exists that the workers’ compensation system did not reimburse for all injury-related prescriptions, as seen in some prior studies.