Risk of Biologics and Glucocorticoids in Patients With Rheumatoid Arthritis Undergoing Arthroplasty

Abstract

Biologic disease-modifying antirheumatic drugs (bDMARDs) selectively target the immune system and are increasingly used in the treatment of rheumatoid arthritis (RA). These medications increase risk for serious infection compared with conventional synthetic DMARDs, such as methotrexate (14). Although bDMARDs are often discussed as a single class, adverse effect profiles (including infection risk) may differ between medications because of mechanisms of action or dosing. Randomized trials have infrequently compared different biologics and have not been powered to assess serious infections (5). In some observational studies, infliximab and tocilizumab have been associated with greater infection risk and abatacept and etanercept with lower risk (610). Studies in surgical patients are lacking. Infection risk is particularly important in patients undergoing major surgery. Patients with RA frequently have orthopedic surgery, especially total hip or knee arthroplasty, and are at increased risk for postoperative complications (11, 12). Joint arthroplasty can be complicated not only by immediate postoperative infections but also by prosthetic joint infection (PJI), a serious complication that can occur later after surgery (13). Understanding risk for postoperative infection with different immunosuppressive therapies is important to optimizing perioperative management. We used 2 large administrative data sets to compare risk for postoperative infections and readmissions in patients with RA exposed to different biologic therapies before total hip or knee arthroplasty. In addition, we evaluated the risk for postoperative infection associated with glucocorticoids. Methods This retrospective cohort study used Medicare claims and MarketScan databases (Truven Health Analytics) from 1 January 2006 to 30 September 2015 to evaluate patients with RA who had hip or knee arthroplasty. Medicare is a public health plan that covers more than 90% of U.S. adults aged 65 years or older (14). Younger persons with disabilities (such as RA) may also be covered. MarketScan is a U.S. database that includes inpatient, outpatient, and pharmacy data from large employers, health plans, and government and public organizations for more than 143 million persons (15). This study was approved by the University of Pennsylvania and University of Alabama at Birmingham institutional review boards. Cohort Identification Included patients were aged 18 years or older; had RA based on 2 International Classification of Diseases, 9th Revision (ICD-9), codes (714.xx) assigned by a physician at least 7 days apart and use of a DMARD (16); and had inpatient elective hip or knee arthroplasty, either primary or revision, between 1 January 2007 and 31 August 2015 on the basis of ICD-9 and CPT (Current Procedural Terminology) codes for primary surgical procedures and CPT codes for revisions, according to validated algorithms (Supplement Table 1) (1720). The index date was the day of surgery and occurred within the index hospitalization (Supplement Figure 1, shows the study design). Supplement. Supplementary Material Included patients had documented infusions of or filled prescriptions for tumor necrosis factor (TNF) antagonists (infliximab, adalimumab, or etanercept), abatacept, or tocilizumab within 8 weeks or received rituximab within 16 weeks before surgery. We required at least 3 infusions or prescriptions in the past year (2 for rituximab) to identify patients receiving long-term therapy. We also required a 1-year baseline period before the index date with continuous enrollment in Medicare Parts A, B, and D or MarketScan. We excluded patients with the following evidence of preexisting infection or nonelective surgery: diagnosis or treatment of native joint infection or PJI in the year before surgery, diagnosis of femur fracture or bone or metastatic cancer from the index hospitalization, admission through the emergency department or transfer from another acute care hospital, surgery after hospital day 3, or major surgery in the previous 6 months (Supplement Table 1). Patients with an admission status of emergent (available only in Medicare) were also excluded. Those with a diagnosis of inflammatory bowel disease, psoriatic arthritis, ankylosing spondylitis, HIV, or active cancer in the past year were excluded because these could affect bDMARD use. A patient could contribute multiple procedures if they were more than 6 months apart. To avoid double counting patients, we excluded those in MarketScan whose derived birthdate was within 31 days of that of any patient in Medicare with the same admission date and biologic exposure (Supplement Table 2). Outcomes Primary outcomes were hospitalized infection within 30 days and PJI within 1 year after surgery. We identified hospitalized infections on the basis of ICD-9 codes from any discharge diagnosis, not just the primary diagnosis (positive predictive value >80%) (1, 21), including from the index hospitalization and any subsequent acute care hospitalizations with an admission date within 30 days of surgery. We identified PJI on the basis of an inpatient or outpatient physician diagnosis (ICD-9 code 996.66) within 1 year after the index hospitalization (excluding patients with PJI diagnoses from the index hospitalization because preexisting infection may have been the indication for surgery) (22, 23). Sensitivity analyses assessed more stringent definitions of PJI that required either an inpatient diagnosis or an accompanying procedure code within 30 days of PJI diagnosis (that is, arthrotomy, prosthesis removal, central venous catheter insertion, spacer insertion, or revision surgery) (Supplement Table 1) (18). Secondary outcomes were an alternate 30-day outcome for hospitalized infection that excluded urinary tract infections (which may be more minor or incidentally discovered), as well as 30-day readmissions among patients discharged to home, home health care, short-term rehabilitation, or a skilled-nursing facility (excluding hospitalizations within 1 day of discharge or with a primary diagnosis indicating rehabilitation) (24). Exploratory outcomes included prolonged length of stay (a surrogate for postoperative complications) (25, 26) and time to revision surgery among patients who had primary knee or hip arthroplasty (27). Length of stay above the 90th percentile by year and procedure type was considered prolonged (empirically derived as >4 days for primary procedures between 2011 and 2015 and >5 days for all other procedures) (26). We also examined wound complications (28), specific infections (pneumonia, septicemia or bacteremia, and urinary tract infection) (Supplement Table 1), and 30-day mortality (Medicare only). Covariates Covariates measured during the baseline period included demographic characteristics; comorbid conditions; an adaptation of the Charlson comorbidity index (29); health care use (outpatient visits, emergency department visits, and hospitalizations); previous hospitalized infection; and medication use within 90 days, including nonsteroidal anti-inflammatory drugs, opioids, methotrexate, other conventional synthetic DMARDs, glucocorticoids, and antibiotics. Average glucocorticoid dosage in the 90 days before surgery was calculated on the basis of oral prescriptions for prednisone, prednisolone, and methylprednisolone; we used prescribed dosage in prednisone equivalents and days supply to determine each daily dose and truncated prescriptions if a new prescription was filled before the preceding end date. Medicare data included disability status, skilled-nursing facility residence, and quintiles of median household income based on ZIP code from the American Community Survey, 2009 to 2013 (30). Surgeon and hospital volume (Medicare only) were estimated among 55812 hip or knee arthroplasties in patients with any bDMARD or methotrexate use within 6 months before surgery. Statistical Analysis Primary Analysis Associations between preoperative exposure to biologics or glucocorticoids and postoperative outcomes were assessed using logistic regression for binary outcomes and competing risk regression (FineGray models) for PJI; data were censored at 1 year after surgery, end of enrollment in Medicare or MarketScan, subsequent hip or knee arthroplasty, or 30 September 2015, with death as a competing risk (31). Inverse probability weights derived from propensity scores were used to balance confounders across treatment groups (see Propensity Scores section). Predicted risk and risk differences were estimated from logistic regression models at the means of all covariates. Predicted 1-year cumulative incidence and differences in 1-year cumulative incidence were calculated using the reference cumulative incidence function and linear predictions from FineGray models (32). Abatacept and no glucocorticoid use were the reference groups for the analyses of biologic and glucocorticoid use, respectively. Cause-specific hazard models with death treated as censoring provided nearly identical results. Medicare analyses were clustered by hospital (not available in MarketScan), whereas MarketScan analyses were clustered by patient (33). The large number of hospitals (n= 2228), of which 94% contributing 10 or fewer procedures, prevented analyses within hospitals (Supplement Table 3). Analyses were done separately in Medicare and MarketScan, odds ratios (ORs) and subdistribution hazard ratios (HRs) were combined using inverse varianceweighted fixed-effects meta-analysis, and heterogeneity was assessed visually and using I 2 values (34). Combined ORs and HRs were used to calculate combined predicted risk and 1-year cumulative incidence for each exposure group; reference risk and reference 1-year cumulative incidence were based on crude results in the reference group from pooled Medicare and MarketScan data. Small sample size prevented covariate balance in the rituximab and tocilizumab exposure groups in MarketScan, so these