Use of administrative claims data for comparative effectiveness research of rheumatoid arthritis treatments

Observational studies, particularly those using large administrative claims databases, have become increasingly popular sources of comparative effectiveness or comparative safety research. Studies using claims data often face challenges and criticisms due to the lack of certain clinical information, such as lifestyle risk factors, disease severity, and questionable accuracy of disease diagnoses. A novel, claims-based algorithm to evaluate the clinical effectiveness of rheumatoid arthritis medications has been developed and its performance seems promising, although further validation is needed.

Since the fi rst approval of biologic therapy for rheumatoid arthritis (RA) in 1998, the treatment of the disease has changed substantially. A number of diff erent biologic agents targeting various cytokines are currently available, but few data exist comparing the eff ectiveness of one biologic agent with another, highlighting the importance and need for comparative eff ectiveness research (CER) in RA [1]. In fact, comparative eff ectiveness of biologic therapy in RA was one of the top 25 priority research topics recom mended by the Institute of Medicine in 2009 [2].
A number of diff erent study designs such as randomized clinical trials, meta-analyses of randomized clinical trials, and observational studies using various data sources including patient registries, electronic medical records, and administrative claims databases can be considered for CER. Th e use of randomized clinical trials is limited in CER because of intrinsic weaknesses such as lack of generalizability, insuffi cient sample size, inadequate follow-up time, and high cost. Observational study designs include prospective registries and retrospective analysis of administrative healthcare data, often collected for insurance payment.
Prospective RA patient registries have a number of benefi ts, including detailed information on RA diagnosis, disease activity, and treatment, but often have limited generalizability and sample size, and incomplete data on comorbidities and other medications [3]. Observational studies, particularly those using large administrative claims databases, have therefore become increasingly popular sources of CER or comparative safety research, because they have several important strengths such as large size and effi ciency, generalizability, high validity and completeness of prescription drug data, and low cost [4]. Furthermore, a previous validation study showed that RA patients can be accurately identifi ed using a combination of diagnosis codes and disease-modifying antirheumatic drug (DMARD) prescriptions in claims data [5]. However, pharmacoepidemiologic studies using claims data face challenges and criticisms due to the lack of certain clinical information, such as lifestyle risk factors, disease severity, and questionable accuracy of disease diagnoses. A number of previous studies successfully used claims data to assess comparative eff ectiveness of DMARDs on specifi c outcomes [6,7], but not so much research has been done to compare the eff ectiveness of DMARDs in RA activity.
In the current issue of Arthritis Research & Th erapy, Curtis and colleagues present the development and validation of a novel, claims-based algorithm to evaluate the clinical eff ectiveness of RA medications [1]. Th is study has an important implication in CER of RA and shows the potential for using the claims data to compare the clinical eff ectiveness of multiple biologic or nonbiologic DMARDs in large real-world populations. RA patients initiating one of the biologic agents (abatacept,

Abstract
Observational studies, particularly those using large administrative claims databases, have become increasingly popular sources of comparative eff ectiveness or comparative safety research. Studies using claims data often face challenges and criticisms due to the lack of certain clinical information, such as lifestyle risk factors, disease severity, and questionable accuracy of disease diagnoses. A novel, claims-based algorithm to evaluate the clinical eff ectiveness of rheumatoid arthritis medications has been developed and its performance seems promising, although further validation is needed.

E D I TO R I A L
*Correspondence: skim62@partners.org 1 Division of Rheumatology, Immunology and Allergy, Brigham and Women's Hospital, 1620 Tremont Street, Suite 3030, Boston MA 02120, USA Full list of author information is available at the end of the article adalimumab, etanercept, infl iximab and rituximab) were identifi ed based on the data from the longitudinal Veterans Aff airs RA registry linked to the Veterans Health Administration medical and pharmacy claims [1]. Th e eff ectiveness algorithm consists of strict, a priori defi ned criteria: high drug adherence, an increase in biologic dose compared with the starting dose, switching to a diff erent biologic or adding a new nonbiologic DMARD, initiation of chronic glucocorticoids, an increase in glucocorticoid dose during the follow-up period, and more than one parenteral or intra-articular injection on a given day after the patient had been on biologic treatment for longer than 3 months. Th e gold standard for eff ectiveness was defi ned as 28-joint Disease Activity Score <3.2 (low disease activity) or improvement in 28-joint Disease Activity Score >1.2 units at the 1-year follow-up visit following the index visit. In the authors' preliminary assessment, the algorithm seems promising with good performance characteristics, ranging from 75 to 90% [1].
While this study represents an important eff ort, several potential pitfalls in this claims-based eff ectiveness algorithm should be noted. First, performance of the algorithm may be database dependent. In other words, the algorithm may perform diff erently in a commercially insured or Medicare population versus the Veterans Aff airs population in which it was developed. Whether the algorithm will have a similar performance in other claims databases therefore needs to be further examined. Second, as the algorithm required patients to have high adherence to DMARDs (over 80%), it may not perform well in non adherent patients. One cannot therefore assume the algorithm represents good disease control since it was developed in a population who were medi cation adherent. Th ird, the performance of the eff ectiveness algorithm was assessed at 1-year follow-up. As the authors suggested, the validity of the algorithm should be confi rmed at diff erent time points.
A claims-based eff ectiveness algorithm with acceptable performance characteristics across diff erent data settings will be a powerful and desired tool for CER of RA. Such an algorithm will enable large-scale, population-based studies comparing the eff ectiveness of diff erent DMARD regimens. Such studies will facilitate head-to-head comparisons, supplementing typical randomized controlled trials and prospective registries that usually include disease activity.

Competing interests
SYK has received research support from Takeda Pharmaceuticals North America and Pfi zer. DHS has received research support from Abbott Immunology, Amgen, Lilly, and an educational grant from Bristol-Myers Squibb. He serves as an unpaid member of an Executive Committee and a Data Safety Monitoring Board for two analgesic trials sponsored by Pfi zer.