We obtained local institutional review board approval. We conducted this study using Medicaid Analytic eXtract (MAX) files from all 50 U.S. states and the District of Columbia from 2000 through 2010, inclusive. These were the most recent data available to us at the time of the study. MAX files contain medical and pharmacy administrative claims records for low-income children enrolled in Medicaid (U.S. government medical assistance). We have previously published studies of JIA using this data source [4, 17].
Using pharmacy and infusion claims, we identified patients who were new users of MTX or biologic agents, including TNFi (etanercept, adalimumab, infliximab, golimumab, and certolizumab), abatacept, anakinra, canakinumab, and tocilizumab. New use was defined by a 6-month baseline clean period of nonuse during which the patient had full medical and pharmacy benefits. Additionally, new users of any TNFi had no use of any TNFi or any other biologic agent during the baseline period (i.e., those switching biologic agents were excluded). Only the first observed use of each medication was considered for new use. Individual patients could have more than one episode of new use during the study period (e.g., MTX and then TNFi). New users who had at least one physician diagnosis code consistent with JIA (International Classification of Diseases, Ninth Revision [ICD-9], codes 714, 696.0, 720) before age 16 years and prior to the new-use prescription fill date (index date) were included. We excluded patients who had any of the following at any time prior to the index date: (1) a physician diagnosis code for JIA when less than 6 months of age; (2) any physician diagnosis code or hospital discharge diagnosis code for inflammatory bowel disease (ICD-9 codes 555, 556); (3) any physician diagnosis code or hospital discharge diagnosis code for malignancy, organ transplant, or HIV infection; and (4) two or more physician diagnosis codes or hospital discharge diagnosis codes for other rheumatologic diseases (systemic lupus erythematosus and other diffuse connective tissue diseases, vasculitis, sarcoidosis) that were at least 7 days but more than 183 days apart. New MTX users were excluded if they used biologic agents at any time prior to the index date.
In order to better evaluate relative infection rates associated with the biologic agents used to treat SJIA, we attempted to identify patients within the MTX new-user cohort who were very likely to have SJIA. However, within ICD-9 there is no specific, reliable physician diagnosis code for SJIA. Therefore, we considered new users of MTX who met any of the following at any time in the data to have SJIA: (1) any physician diagnosis code or hospital discharge diagnosis code for macrophage activation syndrome (ICD-9 code 288.4); (2) any receipt of anakinra, canakinumab, or rilonacept following receipt of MTX; (3) any receipt of cyclosporine in the absence of any physician diagnosis code for uveitis; and (4) any receipt of thalidomide or lenalidomide. Among children with JIA, the development of macrophage activation syndrome  and use of the selected medications mentioned above [1, 13] occur nearly exclusively in children with SJIA.
The baseline mean daily GC dose was determined by summing the total cumulative oral GC dose dispensed during the 6-month baseline period in milligrams of prednisone equivalents and dividing by 183 days. Baseline mean daily GC dose was categorized as none, low dose (<10 mg of prednisone equivalents per day), and high dose (≥10 mg of prednisone equivalents per day). Among new TNFi users, concurrent MTX use was determined on a time-varying basis during study follow-up. The resultant medication exposure categories were TNFi monotherapy and TNFi + MTX combination therapy, and patients changed exposure categories during follow-up on the basis of their treatment received. Owing to the small number of observed outcomes, we did not adjust or restrict for concurrent MTX use among new anakinra users. The concurrent use of cyclosporine was determined on a time-varying basis during follow-up.
Follow-up began on the index date. Follow-up was censored at the first date that any of the following occurred: (1) prescription for corresponding medication cohort (MTX or specific biologic agent) not filled within 90 days of the days supplied by the previous filled prescription or typical infusion interval (e.g., 56 days for infliximab), (2) meeting any of the study cohort exclusion criteria, (3) experiencing the primary study outcome of hospitalized infection, or (4) end of the study period. Patients in the new MTX user cohort were immediately censored if they received a biologic agent.
Sex, race, and age at index date were determined. The presence of physician diagnosis codes or hospital discharge diagnosis codes for the comorbid conditions of psoriasis, diabetes mellitus, and asthma during the 6-month baseline period was assessed. The occurrence of infection during the 6-month baseline period was assessed and assigned to one of three hierarchical levels: none, nonprimary hospital discharge diagnosis or outpatient physician diagnosis code, primary hospital discharge diagnosis (i.e., a nonprimary discharge or outpatient diagnosis was not considered in the presence of a primary hospital discharge diagnosis). Baseline GC use and time-varying concurrent cyclosporine use during follow-up were modeled as covariates.
The study outcome was a primary hospital discharge diagnosis of infection. The outcome included ICD-9 codes reflecting infections by all organisms (e.g., bacteria, viruses, fungi). As part of sensitivity analyses, we included any hospital discharge diagnosis of infection (i.e., primary and nonprimary hospital discharge diagnoses). The list of infection ICD-9 codes was adapted from previously validated lists [19–21].
We determined the baseline characteristics of the patients in each new medication use cohort. We did not perform statistical tests of the differences between the cohorts, because individual patients could contribute to multiple cohorts and the MTX with SJIA cohort was a subset of the MTX cohort.
We determined crude infection rates for each new medication user cohort if there were at least ten observed infection outcomes. We considered all TNFi inhibitors together as a group compared with MTX in further analyses.
We used Cox proportional hazards regression models to compare the incidence of infections between new biologic agent users and new MTX users. For new users of TNFi, we compared TNFi monotherapy and TNFi + MTX combination therapy versus MTX within the same model. For the new users of anakinra, we separately compared ANA versus MTX and ANA versus MTX with SJIA. We calculated adjusted hazard ratios (aHRs) by using multivariable models to adjust for covariates that were potential confounders of the association between the medication exposures and the outcome. Because individual patients could contribute to more than one new-user cohort, a sandwich variance estimator was applied to account for additional correlations in the data . To evaluate the possibility of statistical interaction between oral GC and biologic agent use and the association with hospitalized infection, we evaluated separate multivariable hazard models for children with and without use of oral GCs during the 6-month baseline period.
We analyzed infection rates among individual TNFi with etanercept as the referent when there were at least ten infection outcomes present. We compared patients’ characteristics among the individual TNFi using chi-square tests. In the multivariable Cox regression models for the comparison of individual TNFi, we additionally adjusted for the number of prior biologic agents used (more than 6 months prior to the index date), and we ignored the concurrent use of MTX.
To evaluate the robustness of our primary results, we performed the following sensitivity analyses: expanded the outcome definition to include hospital discharge diagnosis for infection in any position (i.e., included nonprimary discharge diagnoses), decreased the infection risk window following the days supplied by the last prescription from 90 days to 30 days, decreased the time period used to calculate the baseline mean daily oral GC dose from 183 days to 60 days, and censored all follow-up at 6 months following the index date.