Inefficiencies in Osteoarthritis and Chronic Low Back Pain Management | Page 1
Published Online: October 23, 2013
Margaret K. Pasquale, PhD; Robert Dufour, PhD; Ashish V. Joshi, PhD; Andrew T. Reiners, MD; David Schaaf, MD; Jack Mardekian, PhD; George A. Andrews, MD, MBA, CPE; Nick C. Patel, PharmD, PhD, BCPP; and James Harnett, PharmD, MS
Two forms of chronic pain, osteoarthritis (OA) and chronic low back pain (CLBP), have exceptionally high prevalence and associated healthcare costs in the United States. An estimated 13.9% of adults 25 years and older and 33.6% of adults 65 years and older are affected by OA1; and roughly a quarter of all adults in the United States suffer from CLBP during their lifetime.2 Associated healthcare costs for these diseases have been estimated at $48 billion for OA and $40 billion for back problems in 2005; collectively, musculoskeletal conditions ranked as the third-highest spending category among medical conditions in the United States.3 Given the high prevalence of and healthcare spending on musculoskeletal conditions, it is crucial for providers and payers to provide appropriate and adequate pain management for these conditions. Exposure of potentially inefficient provider practices or patient behavior can aid providers and payers in providing appropriate care and reducing costs.
Sources of inefficiencies in pain management include underdiagnosis or inappropriate diagnosis, use of unnecessary procedures and tests,4-6 and improper use of medications.7 Although examples of such inefficiencies are numerous, very few studies have indicated how to identify patients experiencing suboptimal pain management and to quantify their associated healthcare costs. Goldberg and colleagues8 published measures of inefficiencies developed by an expert clinical panel (Patient Population Assessment to Identify Need [PAIN]) concerning pain management in patients with OA and/or low back pain (LBP). These indicators were then overlaid on a managed care database to identify members who were likely in need of better pain management.8 Additional analysis by Goldberg and colleagues9 demonstrated that total per member per month pain-related costs of members identified with any PAIN indicator of inefficiency ($843) were significantly higher than those for members not identified with an indicator of inefficiency ($121).
The objective of the current study was to use Goldberg and colleagues’ PAIN indicators as a guide to develop inefficiency measures applicable to a specific health insurance provider for Medicare members with OA and/or CLBP. With the information provided in this study, providers and payers will be able to focus on identified members to determine whether their painful conditions are being managed adequately and efficiently.
This study utilized data from Humana’s SAS database, containing enrollment, medical, and pharmacy claims data for Humana’s Medicare membership. All data sources were merged using de-identified member identification. The finalized protocol was approved by an independent institutional review board.
This was a retrospective cohort study using a claims database to evaluate OA and CLBP patients identified as having suboptimal management of pain based on inefficiency measures of prescription drug and medical service use (Table 1). Internal experts from Humana’s clinical and drug utilization review programs were consulted to review the Goldberg PAIN indicators8 and revise them based on clinical judgment. Patients with OA and/or CLBP were identified from January 1, 2008, to June 30, 2010. Each member’s date of first OA or LBP diagnosis was considered to be the index date, and members were required to have 365 days of continuous enrollment preindex and 365 days postindex.
Members 18 years and older in Humana’s Medicare Advantage plans were included if they were identified with 2 or more claims for OA on different days and/or 2 or more claims for LBP in the primary diagnosis position that were 90 or more days apart to ensure the “chronic” nature of pain. The following International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code was used to identify OA: 715.xx. ICD-9-CM codes for LBP were 721.3, 721.42, 721.5-721.9x, 722.1x-722.2, 722.30, 722.32, 722.52, 722.6, 722.70, 722.73, 722.80, 722.83, 722.90, 722.93, 724.00, 724.02, 724.09, 724.2-724.6, 724.8, 724.9, 737.10-737.19, 737.2x, 737.3, 737.30, 738.4, 738.5, 793.3, 793.4, 756.10-756.19, 805.4, 805.6, 805.8, 846.x, 847.2, 847.3, 847.9, 739.3, 739.4, and 996.4.
Members were excluded if they had pregnancy (ICD-9- CM 630.xx-679.xx, V22.xx, and V23.xx), cancer (ICD-9- CM 140.xx-172.xx and 174.xx-208.xx), organ transplant (ICD-9-CM V42.xx), rheumatoid arthritis (ICD-9-CM 714. xx), ankylosing spondylitis (ICD-9-CM 720.xx), human immunodeficiency virus infection (ICD-9-CM 042.xx), or sickle cell anemia (ICD-9-CM 282.6x) in the first or second diagnosis position. Members were also excluded if they resided in skilled nursing homes.
Claims data of Humana Medicare members with OA, CLBP, or both conditions (hereafter OA/CLBP) were analyzed to identify individuals who met criteria for any inefficiency measures in Table 1. The count and percentage of members identified with each inefficiency measure during the postindex period were determined. Demographic and clinical characteristics ascertained during the preindex period were compared between members with and without inefficiencies identified postindex. Variables included age, sex, geographic region, top 10 comorbidities by 4-digit ICD-9-CM codes, psychiatric comorbidities, and the RxRisk-V comorbidity score.The RxRisk-V score10-14 is derived from drug claims data and thus can be applied to data from a narrow window of claims rather than the broader window typically necessary for medical–claims-based comorbidity scores.15 Accordingly, means were compared using 2-sample t tests, and count variables were compared using x2 tests.
Pain-related costs were defined for all provider, facility, and pharmacy claims categories and summed over 365 days postindex for each of the 22 inefficiencies identified during the postindex period. For professional and facility claims, 100% of costs were counted if the OA or LBP diagnosis was documented in the primary position. For claims with OA or LBP in any secondary position, costs were apportioned based on number of listed diagnostic conditions that were consistent with the study by Goldberg and colleagues.8 For pharmacy claims, all postindex costs were summed for opioids, nonsteroidal anti-inflammatory drugs, antimigraine agents, antidepressants, antiepileptics/anticonvulsants, muscle relaxants, and steroids.
Once members with inefficiencies and associated costs were identified, the inefficiencies were rank-ordered from costliest to least costly. A member was allowed to have 1 or more inefficiencies; hence, mean costs of inefficiencies were not independent of each another. In order to determine whether the rankings were statistically significant, pairwise Wilcoxon signed rank tests were performed first within each inefficiency category and then for high-cost inefficiencies overall. The alpha value to achieve significance was adjusted via the Bonferroni method for the number of comparisons within each inefficiency category (15 comparisons for opioid use, 21 for miscellaneous drug use, and 28 for medical service use). In addition, members with and without inefficiencies were compared with respect to their postindex total pain-related healthcare costs. Generalized linear modeling of the log of the mean total 365-day postindex cost (adjusted to 2010 dollars) was performed on the following covariates: age, sex, geographic region, baseline RxRisk-V score, presence of an inefficiency, disease state (OA/CLBP and CLBP versus OA), presence of any of the top 10 medical comorbidities, and psychiatric comorbidity. The parameter estimates, Wald 95% confidence limits, exponentiated estimates, and Wald x2 test results were computed using the SAS PROC GENMOD procedure with log link function and gamma distribution (SAS Institute Inc, Cary, North Carolina). This approach to linear modeling is appropriate when modeling claim costs as both the gamma and claim costs distributions tend to approach a normal distribution.
Final sample sizes of 51,773 OA, 11,510 CLBP, and 5170 OA/CLBP members were available for analysis. We identified 46.9% of OA, 80.1% of CLBP, and 84.8% of OA/CLBP members as having at least 1 inefficiency measure postindex (Table 2). The mean number of inefficiencies per member with at least 1 inefficiency was highest for members with OA/ CLBP (2.65), followed by CLBP (2.23) and OA (1.73). Overall, 71% of the members with inefficiencies were identified as having an inefficiency from 1 category rather than multiple categories; only 4% of members with inefficiencies were identified with an inefficiency from all 3 categories.
PDF is available on the last page.