Derivation of severity index for rheumatoid arthritis and its association with healthcare outcomes

Objectives: To develop a claims-based severity index for rheumatoid arthritis (RA) using the Veterans Health Administration (VHA) database. Methods: Adult patients with at least two RA diagnoses 2 months apart were identified between 10/1/200809/30/2009. Patients were required to have at least 12 months continuous health plan enrollment before and after the index date (first RA diagnosis date) for an overall study period of 10/1/200709/30/2010. A severity index for rheumatoid arthritis (SIFRA, a proprietary algorithm of SIMR, Inc. [STATinMED Research]) was developed by calculating a weighted sum of 34 RA-related indicators assessed by an expert Delphi panel of six rheumatologists, including laboratory, clinical, and functional status, extra-Articular manifestations, surgical history, and medications, during a 1-year pre-index period. Separate SIFRA versions were derived for patients with and without laboratory information. Correlations between SIFRA and previously validated claims-based indexes for RA severity (CIRAS), and other traditional comorbidity indexes were calculated during the pre-index period. The relationship between SIFRA and follow-up healthcare outcomes was also examined using histograms. Results: The Spearman's rank correlations between SIFRA and CIRAS were 0.525 for SIFRA without and 0.539 with laboratory data. The correlations between SIFRA and the Charlson Comorbidity Index (CCI) (0.1503 without, 0.1135 with laboratory data), Elixhauser Index (ELIX) (0.105 without, 0.079 with laboratory data), and Chronic Disease Score (CDS) (0.255 without, 0.239 with laboratory data) were low. Histograms showed that patients in the upper tercile of SIFRA incurred $9123 more all-cause and $1326 more RA-related healthcare costs during the 1-year post-index period than patients in the lower tercile. Using SIFRA in combination with CCI, CDS, or ELIX significantly increased the percentage of variation explained in outcomes measures. Limitations: Patients in the VHA database may not represent typical RA patients since the database generally contains older, economically disadvantaged men with a high disease burden. Validity of the score is indirectly based on disease activity score 28 (DAS28), which measures disease activity rather than severity. Conclusions: SIFRA was found to have moderate correlations with the previously validated CIRAS score, and demonstrated evidence of being a significant determinant of total and RA-related healthcare costs for RA patients. This study suggests that SIFRA could be an important methodological tool to control for severity in RA-related outcomes research. The algorithm can be applied to any claims dataset.