Opioid Treatment for Pain and Work Outcomes: Evidence from Physicians’ Prescribing Patterns

Medical conditions that cause pain are an important and growing reason why workers exit the labor force: the share of Social Security Disability Insurance (SSDI) beneficiaries with a musculoskeletal disorder (often back or joint pain), for example, has increased from 13 percent in 1983 to 31.2 percent in 2014 and is the leading reason for which disabled workers receive benefits (SSA 2015). The fall in the labor supply of individuals with pain has occurred despite the increased availability of medications to treat pain – such as opioids – over the same time period (Sites et al. 2014). The impact of opioid use on the labor supply of individuals with pain remains poorly understood: while more effective pain control could in principal increase labor supply, the numerous adverse medical consequences of sustained opioid use – including but not limited to dependence and addiction – might offset these gains. Moreover, increased opioid supply in a given area could have negative spillovers to the labor supply of individuals without pain, if it increases the risk of recreational drug use and addiction. Understanding the economic trade-offs of opioid treatment for pain is therefore critical to informing public policy. A key empirical challenge in estimating the causal impact of opioid use on employment is that it is correlated with the prevalence of pain, and perhaps other unobserved measures of labor force attachment. We therefore use an instrumental variables strategy exploiting exogenous variation in physicians’ opioid prescribing preferences. This approach draws on a growing literature demonstrating that, under certain conditions, physicians’ characteristics can predict their propensity to use specific treatments, even after controlling for characteristics of the patients they serve (Perry 2008; Barnett et al. 2017). We use claims data from multiple large commercial insurers operating across the US to obtain a high-resolution view of individual physicians’ opioid prescribing behaviors within their patient panel, and develop measures of propensity to prescribe opioids, and propensity to prescribe with higher intensity and duration, that are not fully explained by patient characteristics. Because our data offer broad geographic coverage, we examine whether physicians with higher propensities to use opioids are concentrated within certain areas, and relate this to area-level labor force participation.