Data Source and Study Sample
The data source for this exploratory, retrospective, longitudinal study was the Medical Expenditure Panel Survey (MEPS) for 2010–2015. MEPS is sponsored by the Agency for Healthcare Research and Quality and is a nationally representative annual survey of the civilian noninstitutionalized population in the USA [16, 17]. Data collected from families and individuals during the survey included demographic characteristics; socioeconomic status; medical conditions; and health status, such as functional and activity limitations; as well as health-related quality of life; use of medical services; medical expenditures; access to care; and health insurance coverage. Because this study used secondary data from MEPS, which is a publicly available database, all ethics approvals and consent to participate were waived by the institutional review board at West Virginia University.
These data are collected in interview panels, with each panel following the same individuals for two consecutive years. A total of five rounds of data (interviews) are collected over the course of each 2-year panel (Fig. 1). This panel design enables longitudinal assessment of outcomes for each participant over a 2-year period. The current analysis was based on panels for 2010–2011, 2011–2012, 2012–2013, 2013–2014, and 2014–2015, corresponding to panels 15, 16, 17, 18, and 19, respectively, in the MEPS database.
The criteria used for identification of the evaluated population was the same as that in a previously published study on the association of opioid use and PIA with economic outcomes . For inclusion in the analysis, adults (aged at least 18 years old) with OA were identified in the MEPS database for panels 15–19 from queries related to whether the respondent had “OA” in the full-year consolidated files or medical care encounters recorded in the medical condition files using the Clinical Classification Code (CCC) identifier 203. This CCC identifier matches to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes 715.xx (osteoarthrosis and allied disorders). The CCC code of 203 also includes the ICD-9 code V134 (personal history of arthritis). This code was considered relevant because including this code could help identify patients with OA who did not seek healthcare services for OA in the year surveyed. Individuals reporting any cancer (CCC 11–44) were excluded.
The outcomes were PIA and functional limitations during the follow-up period (Fig. 1). PIA, which was collected in rounds 2 (baseline) and 4 (follow-up) in each panel, was derived from the 12-item Short Form Health Survey version 2 (SF-12 v2) health survey : “During the past 4 weeks, how much did pain interfere with your normal work (including both work outside the home and housework)?” with response options of “not at all,” “a little bit,” “moderately,” “quite a bit,” and “extremely.” In the current analysis, these responses were recategorized as no/mild (responses of “not at all/a little bit”), moderate (response of “moderately”), and severe (responses of “quite a bit/extremely”).
Changes in PIA groups were considered “better” if an individual moved from a higher to a lower PIA level during follow-up, “worse” if the movement was from a lower to a higher PIA level during follow-up, and “no change” if PIA levels were the same at both baseline and follow-up.
Functional limitations, dichotomized as “Yes” or “No” and collected in rounds 3 (baseline) and 5 (follow-up) in each panel, included any limitations in activities of daily living (ADL), instrumental ADL (IADL), physical activities, social function, work, and cognitive function. Although ADL is defined as the basic activities required for daily function (e.g., bathing, eating, dressing, transferring, toileting, and walking) [19, 20], IADL are not required for daily functioning but contribute to maintaining an individual’s independence (e.g., housework, preparing meals, taking medications, shopping, telephoning, and managing money) [21, 22]. Changes in functional limitations were defined by the following four categories: “better” if an individual with a limitation at baseline was not limited at follow-up, “worse” if there was a limitation at follow-up in an individual with no limitation at baseline, “same” if there were limitations at baseline and follow-up, and those with no limitations at both baseline and follow-up.
Study participants reported their opioid use at every round and thus definitions of opioid use patterns needed to account for the manner in which data were collected. Thus, persistent use was defined when opioid use was reported in at least two consecutive rounds before the last round of follow-up (rounds 1–3 for PIA analysis and rounds 2–4 for functional limitation analysis), and intermittent opioid use defined when reported in any one of the rounds; no opioid use was defined if such use was not reported in any of the rounds.
Unadjusted differences across the opioid use patterns for observed longitudinal changes in PIA and functional limitations were evaluated using Rao–Scott chi-square tests. Multinomial logistic regressions were conducted to examine the association between opioid use patterns and PIA during follow-up. Stratified multivariable logistic regressions explored the association of persistent or intermittent use of opioids versus no use of opioids with follow-up functional status; the stratification variable was the baseline functional status (i.e., patients with or without functional limitations at baseline).
Adjustment for covariates in the regression models was based on Andersen’s Behavioral Model of Health Services Use that included demographics, comorbidities, and health characteristics . Specific variables included in this model were predisposing factors (age group, sex, race/ethnicity), enabling factors (marital status, educational level, employment status, economic status based on family income relative to the federal poverty line, health insurance coverage, prescription drug coverage), external environment (geographic region), need factors (number of comorbidities, depression and anxiety), and personal health practices (smoking status, body mass index, and physical activity [at least five times a week]). In the adjusted analyses, no opioid use with no/mild PIA was used as the reference group. The regression analyses also adjusted for the use of prescription NSAIDs, which was determined in every round and defined in a manner similar to that for opioid use (i.e., persistent, intermittent, and no use). Regression results are expressed as adjusted odds ratios (AOR) and their 95% confidence intervals (CIs).
All analyses account for the complex survey design of MEPS and were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC). Weights specifically assigned for longitudinal data files (LONGWT) were used to account for the longitudinal design. All reported results were weighted to be nationally representative.