Reply

Abstract

We believe there are at least 3 reasons for this overestimation of OA-attributable medical expenditures. First, Kotlarz and colleagues appear to have identified the sample based on self-reported OA. The sensitivity of self-reported OA in the MEPS is very low (6). That, combined with evidence from a national population-based survey that the accuracy of self-reported arthritis was almost twice as high among people who had been hospitalized for this condition (7), suggests that the study cohort as defined is likely biased toward those with more severe and costly OA, which would elevate both the individual and aggregate estimates.Second, the population prevalence of OA used to calculate the aggregate estimate was based on data for AORC (46.4 million adults, or 21%) (8), deflated by 5% to remove the subpopulation with rheumatoid arthritis and then further adjusted for insurance status. The specific estimate used was not presented in Kotlarz and colleagues' report but is likely much higher than the best available estimate for OA (27 million) provided in the companion article (9) to the one they cite, thus elevating the aggregate estimate.Third, the modeling of the OA-attributable costs included adjustment for only 5 chronic conditions: hypertension, hyperlipidemia, diabetes mellitus, anxiety disorders, and asthma. This is a short list, considering the frequent co-occurrence of OA with other expensive chronic conditions (e.g., heart disease). In effect, these OA-attributable expenditure estimates include the costs of other expensive medical conditions that occur more frequently among people with OA. Lee et al describe how inadequate adjustment for comorbid conditions can falsely increase the magnitude of OA-attributable costs (10).Population-based cost-of-illness studies are an important source of information for policymakers and society. Cost-ofillness research is rapidly evolving with wider use of econometric modeling like that used in Kotlarz and colleagues' study. As the field evolves, we need to understand and share the methodological complexities inherent in the design of such studies, clearly communicate the methodology of our studies, and continually question whether results (our own and those of others) seem reasonable, especially in light of existing literature.