Risk of Hospitalized Bacterial Infections Associated With Biologic Treatment Among US Veterans With Rheumatoid Arthritis

Authors

  • J. R. Curtis,

    Corresponding author
    1. University of Alabama at, Birmingham
    • University of Alabama at Birmingham, Faculty Office Towers 802, 510 20th Street South, Birmingham, AL 35294. E-mail: jcurtis@uab.edu

    Search for more papers by this author
    • Dr. Curtis has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Pfizer, BMS, Crescendo, UCB, and AbbVie, and (more than $10,000 each) from Roche/Genentech, Janssen, CORRONA, and Amgen.

  • S. Yang,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
  • N. M. Patkar,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
  • L. Chen,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
  • J. A. Singh,

    1. University of Alabama at Birmingham and VA Medical Center, Birmingham, Alabama
    Search for more papers by this author
    • Dr. Singh has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Savient, Regeneron, URL Pharmaceuticals, Ardea, Allergan, and Novartis, and (more than $10,000) from Takeda, and has received investigator-initiated grants from Takeda and Savient.

  • G. W. Cannon,

    1. VA Medical Center, Salt Lake City, Utah
    Search for more papers by this author
  • T. R. Mikuls,

    1. Omaha VA Medical Center, Omaha, Nebraska
    Search for more papers by this author
    • Dr. Mikuls has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Genentech and Roche.

  • E. Delzell,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
    • Dr. Delzell has received research support from Amgen.

  • K. G. Saag,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
    • Dr. Saag has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Amgen, Abbott, BMS, Roche, and AbbVie.

  • M. M. Safford,

    1. University of Alabama at, Birmingham
    Search for more papers by this author
  • S. DuVall,

    1. VA Medical Center and University of Utah, Salt Lake City
    Search for more papers by this author
    • Dr. Duvall has received research grants from Amgen, Anolinx, Genentech, Merck, Mylan Specialty, Roche, Shire, and Hoffman-La Roche.

  • K. Alexander,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Dr. Alexander owns stock and/or stock options in Genentech/Roche.

  • P. Napalkov,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Dr. Napalkov is a shareholder of Hoffmann-La Roche/Genentech.

  • Kevin L. Winthrop,

    1. Oregon Health Sciences University, Portland
    Search for more papers by this author
    • Dr. Winthrop has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from UCB, Genentech, Regeneron, BMS, and AbbVie, and (more than $10,000) from Pfizer.

  • M. J. Burton,

    1. G. V. Sonny Montgomery VA Medical Center, Jackson, Mississippi
    Search for more papers by this author
  • A. Kamauu,

    1. Anolinx, Salt Lake City, Utah
    Search for more papers by this author
    • Dr. Kamauu owns stock and/or stock options in Anolinx, has received research grants from Genentech, and has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Roche, Shire, and Dey Pharma.

  • J. W. Baddley

    1. University of Alabama at Birmingham and VA Medical Center, Birmingham, Alabama
    Search for more papers by this author
    • Dr. Baddley has received consultant fees, speaking fees, and/or honoraria (less than $10,000 each) from Pfizer, Abbott, and Merck.


Abstract

Objective

The comparative risk of infection associated with non–anti–tumor necrosis factor (anti-TNF) biologic agents is not well established. Our objective was to compare risk for hospitalized infections between anti-TNF and non–anti-TNF biologic agents in US veterans with rheumatoid arthritis (RA).

Methods

Using 1998–2011 data from the US Veterans Health Administration, we studied RA patients initiating rituximab, abatacept, or anti-TNF therapy. Exposure was based upon days supplied (injections) or usual dosing intervals (infusions). Treatment episodes were defined as new biologic agent use. Hazard ratios (HRs) and 95% confidence intervals (95% CIs) for hospitalization for a bacterial infection were estimated from Cox proportional hazards models, adjusting for potential confounders.

Results

Among 3,152 unique RA patients contributing 4,158 biologic treatment episodes to rituximab (n = 596), abatacept (n = 451), and anti-TNF agents (n = 3,111), the patient mean age was 60 years and 87% were male. The most common infections were pneumonia (37%), skin/soft tissue (22%), urinary tract (9%), and bacteremia/sepsis (7%). Hospitalized infection rates per 100 person-years were 4.4 (95% CI 3.1–6.4) for rituximab, 2.8 (95% CI 1.7–4.7) for abatacept, and 3.0 (95% CI 2.5–3.5) for anti-TNF. Compared to etanercept, the adjusted rate of hospitalized infection was not different for adalimumab (HR 1.4, 95% CI 0.9–2.2), abatacept (HR 1.1, 95% CI 0.6–2.1), or rituximab (HR 1.4, 0.8–2.6), although it was increased for infliximab (HR 2.3, 95% CI 1.3–4.0). Infection risk was greater for those taking prednisone >7.5 mg/day (HR 1.8, 95% CI 1.3–2.7) and in the highest quartile of C-reactive protein (HR 2.3, 95% CI 1.4–3.8) and erythrocyte sedimentation rate (HR 4.1, 95% CI 2.3–7.2) compared to the lowest quartile.

Conclusion

In older, predominantly male US veterans with RA, the risk of hospitalized bacterial infections associated with rituximab or abatacept was similar to etanercept.

INTRODUCTION

Previous research has found an increased risk of serious infections associated with some anti–tumor necrosis factor (anti-TNF) therapies compared to other anti-TNF therapies and also compared to nonbiologic disease-modifying antirheumatic drugs (DMARDs) ([1-9]). However, the comparative risk of serious infections associated with anti-TNF therapies in rheumatoid arthritis (RA) compared with non–anti-TNF biologic DMARDs, whose mechanism of action is not via TNF inhibition, has been examined only minimally. A trial conducted for regulatory purposes comparing abatacept (ABA) and the anti-TNF infliximab (INF) suggested that ABA might be safer with respect to serious infections ([10]). Likewise, a more recent trial compared ABA plus methotrexate (MTX) to a different anti-TNF agent, adalimumab (ADA), plus MTX and found a nonsignificant risk difference of less than 1 per 100 patient-years favoring ABA ([11]). Limited observational data suggest the possibility that biologic agents with different mechanisms of action may have a more favorable safety profile with respect to serious infection compared to some anti-TNF therapies ([12, 13]). Overall, however, the evidence base directly comparing the risks of infections across biologic agents with different mechanisms of action is scant.

In addition to knowledge gaps relevant to infection risk with non–anti-TNF biologic agents, the potential impact of systemic inflammation on the risk of serious infections also has been subject to only limited investigation. Some prior reports found that elevated erythrocyte sedimentation rate (ESR) or C-reactive protein (CRP) level was associated with serious infections in RA patients ([14, 15]). However, little is known about a potential dose-response relationship between measures of systemic inflammation and the risk of serious infections.

In light of these uncertainties, we evaluated the drug-specific absolute and relative incidence of hospitalized bacterial infections among US veterans with RA. We compared the individual non-TNF biologic agents, ABA and rituximab (RTX), to anti-TNF biologic agents and, at the same time, determined the relative impact of oral glucocorticoids and various comorbidities. We also examined associations with other clinical factors, including clinical measures of systemic inflammation quantified using ESR and CRP level.

Box 1. Significance & Innovations

  • Few direct comparisons exist regarding the infection profile of various biologic agents for rheumatoid arthritis (RA).
  • Using electronic health records linked to administrative data, we examined the risk for hospitalized infections among older, predominantly male US veterans with RA.
  • In this cohort, and compared to etanercept, the risk of hospitalized infections associated with rituximab or abatacept was comparable, although it was higher for infliximab.

PATIENTS AND METHODS

Study population and inclusion/exclusion criteria (history of biologic agent use and malignancy).

Following institutional internal review board approval, we conducted a retrospective cohort study and examined records of patients receiving care from the US Veterans Health Administration (VHA). The data source was electronic health record (EHR) data from the VA Informatics and Computing Infrastructure ([16]), linked to administrative data from the VHA Decision Support System from 1998–2011. We defined RA based upon ≥2 diagnosis codes (International Classification of Diseases, Ninth Revision [ICD-9], code 714.xx) from rheumatologists on separate days, or a single RA diagnosis code plus pharmacy dispensing of a DMARD or biologic agent approved for use in RA ([17]). DMARDs included MTX, sulfasalazine, leflunomide, and hydroxychloroquine. Biologic agents included etanercept (ETA), INF, ADA, RTX, and intravenous ABA (subcutaneous ABA was not used in the VHA in this time frame). Newer anti-TNF agents (e.g., certolizumab, golimumab, and tocilizumab) were not used frequently enough to allow for inclusion in this study.

Because patients who fail first-line biologic agent treatment have more refractory RA and may be more predisposed to a hospitalized infection, which could introduce substantial confounding, anti-TNF exposure was limited to patients who had prior exposure to a different anti-TNF. In the primary analysis, this requirement was not implemented for ABA and RTX given that their predominant use has been for RA patients with an inadequate response to 1 or more anti-TNF agents. At least in the past, these 2 therapies were rarely used as first-line agents, and any failure to have observed prior anti-TNF therapy for the ABA and RTX patients likely resulted from left censoring of the data. Patients with active cancer, or a history of cancer, were excluded from analysis given that their risks for infection related to malignancy or its treatment could introduce substantial heterogeneity in the patient population that could not be adequately controlled for analytically. Patients were allowed to contribute exposure time to more than 1 medication exposure group (ABA, RTX, or anti-TNF exposure group) if they used multiple different biologic medications.

Observation time for this analysis began at the “index date,” defined as the date of the first use of ABA, RTX, or a new anti-TNF therapy with prior exposure to a different anti-TNF agent. In order to verify that all patients were new users of each biologic agent ([18]), patients were required to be “observable” in the 12 months prior to the index date. Patients were considered observable in each person-month if they had at least 1 visit with a primary care physician in the preceding year, plus a prescription for any medication in the preceding 6 months. In order to remain under followup, patients must have continued to be observable in each person-month using this definition.

Exposure and outcome assessment

Each drug-specific index date defined a medication treatment episode, and patients could contribute up to 1 treatment episode for each specific biologic medication, but could contribute multiple treatment episodes to the analysis. For example, a biologic-naive RA patient who initiated ETA, later switched to ADA, then to INF, and then to ABA would contribute 2 anti-TNF treatment episodes (one for ADA and another for INF) and also an ABA treatment episode. However, the ETA treatment episode would not be included, since at the ETA index date the patient was biologic naive. Exposure was characterized “as treated” (i.e., current exposure) starting with the drug-specific index date with followup for subcutaneously injected biologic agents based upon the days' supply dispensed. Current exposure after infusion was based upon the usual dosing frequency and assumed to be 56 days for INF, 30 days for ABA, and 270 days for RTX. Current exposure was extended by 90 days for all biologic medications so as to capture infectious events that might occur shortly after medication discontinuation (which might occur because of early symptoms of an infection) ([5]). The 270-day duration of exposure assumed for RTX plus the 90-day extension totaled 1 year, which was selected based on common usage as a washout period for RTX in RA clinical trials. Patients with overlapping exposure to 2 biologic agents simultaneously (e.g., an INF infusion, followed 30 days later by a switch to ABA) were handled using an indicator variable assessed for inclusion in the multivariable model.

The outcome of interest was first hospitalization for bacterial infection, defined using ICD-9 codes and listed as the principal reason for hospitalization. This method of case ascertainment has been shown to have high validity in previous studies that have compared this approach to medical record review with case adjudication by infectious disease doctors ([1, 19, 20]). Opportunistic infections were not included in this analysis as they do not always require hospitalization, are numerically uncommon, are sometimes subject to meaningful misclassification in administrative data ([21]), and are the subject of another report where detailed microbiology data were available to characterize these outcomes. Among bacterial infections, those requiring hospitalization represent more severe infections with worse outcomes and satisfy a regulatory definition for a serious adverse event.

Covariate assessment

Covariates were selected based on content expertise and potential association with infection risk informed by review of the literature ([1, 14, 22]). These covariates included demographics, comorbidities (e.g., diabetes mellitus, chronic lung disease, and cardiovascular disease), RA-associated health care utilization, and medications used for arthritis (e.g., MTX, narcotic analgesics, and oral glucocorticoids). The 12-month period preceding each index date defined the baseline period for assessment of most covariates, except for oral glucocorticoids, for which the 6 months preceding the index date was used. Comorbid conditions were characterized using ICD-9 codes from provider diagnoses. In order to classify prior malignancy and prior anti-TNF therapy with maximal accuracy, both the 12-month baseline and all preceding data were used. Likewise, body mass index (BMI) was assessed during the 12-month baseline or earlier.

Seropositive RA was defined as having a positive test result for either rheumatoid factor (RF) or anti–cyclic citrullinated peptide (anti-CCP) antibody using the upper laboratory limit of normal at any time, given that these laboratory tests generally do not vary over time for RA patients with established disease. Only if both values were missing was the seropositive variable considered missing. In the subgroups of patients who had ESR or CRP level tested at least once (approximately 84% and 89%, respectively; 99.8% with either), these 2 laboratory values were examined in separate models (to avoid collinearity) in a time-varying fashion during the baseline period and throughout followup. The results of these laboratory tests were “lagged” by 14 days to the hospitalized infection to avoid measuring systemic inflammation as a consequence of, rather than antecedent to, an active infection. In other words, a laboratory test result occurring within 14 days of a hospitalized infection was not used and, rather, its preceding value (if available) was used. Since most but not all patients had laboratory data measured during the study period, associations with these laboratory test results were conducted as subgroup analyses.

Statistical analysis

Cox proportional hazards models quantified the risk of infections associated with ABA and RTX compared to anti-TNFs, adjusting for potential confounders. Time since the index date for each treatment episode was used as the time axis. Censoring of each treatment episode occurred at the first event, the end of the study period (September 30, 2011), the end of current exposure plus a 90-day extension, death, or the loss of VHA observability. Clustering of treatment episodes within patients was accounted for by using Huber-White sandwich estimates ([23]).

Three sensitivity analyses were performed. The first sensitivity analysis restricted all index dates to 2006 or later (corresponding to the availability of all the biologic agents examined in this study) to allow for comparability between patients eligible to receive any of the therapies of interest. The second sensitivity analysis explicitly required evidence of prior anti-TNF use (at any time and using all available data) for all ABA and RTX patients. The final sensitivity analysis restricted results to patients who lived within 30 miles of the nearest VHA hospital. The purpose of this analysis was to reduce concern that driving distance and associated non-VHA health care utilization were confounders for the main exposure–outcome relationships, in as much as patients who reside at a great distance from the nearest VHA hospital might seek non-VHA care at a local community hospital ([16]). Analyses were performed using SAS, version 9.2.

RESULTS

Among the 36,433 patients who met diagnosis and observability criteria for RA, a total of 3,152 unique patients contributed 4,158 biologic treatment episodes. There were 451 initiations of ABA, 596 initiations of RTX, and 3,111 initiations of an anti-TNF therapy (Table 1). While a majority of patients (n = 2,344, 56.3% of the cohort) contributed only a single treatment episode, 636 patients contributed exactly 2 treatment episodes, 146 patients contributed 3 episodes, and 26 contributed 4 episodes. Mean age was 60 years, and the cohort was more than 80% male. Mean BMI was 29 kg/m2. In general, the prevalence of various comorbidities was somewhat greater among ABA and RTX users compared to anti-TNF users. Most patients (61–73%) were seropositive for either RF and/or anti-CCP antibody. The median duration of followup time in each of the 3 exposure groups was slightly more than 1 year.

Table 1. Baseline characteristics of individuals initiating abatacept, rituximab, or anti-TNF therapy after prior exposure to a different anti-TNF agent*
 Anti-TNFs (n = 3,111)Abatacept (n = 451)Rituximab (n = 596)ADA (n = 1,885)ETA (n = 844)INF (n = 382)
  1. Values are the number (percentage) unless otherwise indicated. Number (n) refers to treatment episodes as defined in the text. Patients were allowed up to 1 treatment episode for each specific biologic agent. Anti-TNF = anti–tumor necrosis factor; ADA = adalimumab; ETA = etanercept; INF = infliximab; COPD = chronic obstructive pulmonary disease; NSAIDs = nonsteroidal antiinflammatory drugs; RF = rheumatoid factor; anti-CCP = anti–cyclic citrullinated peptide.
  2. aAll factors measured in the 12-month baseline period prior to the index date (first use of abatacept, rituximab or each specific anti-TNF therapy), except for glucocorticoid dose, which was assessed in the 6 months prior to the index date. Comorbidities were defined by International Classification of Diseases, Ninth Revision, codes from physician diagnoses.
  3. bProportions given for nonmissing values, approximately 12% (rituximab patients) to 20% (anti-TNF patients) were missing both RF and anti-CCP antibody laboratory test results.
Male2,729 (87.7)377 (83.6)522 (87.6)1,658 (88.0)747 (88.5)324 (84.8)
Age, mean ± SD years60.1 ± 10.660.3 ± 10.660.8 ± 10.660.1 ± 10.859.9 ± 10.757.9 ± 10.5
Body mass index, mean ± SD kg/m229.2 ± 5.829.7 ± 6.129.1 ± 5.929.1 ± 5.729.3 ± 5.929.5 ± 5.8
Comorbiditiesa      
COPD379 (12.2)54 (12)110 (18.5)234 (12.4)94 (11.1)51 (13.4)
Heart failure44 (1.4)18 (4)22 (3.7)26 (1.4)14 (1.7)4 (1.1)
Myocardial infarction81 (2)8 (1.8)12 (2)40 (2.1)15 (1.8)6 (1.6)
Diabetes mellitus695 (22.3)136 (30.2)154 (25.8)421 (22.3)184 (21.8)90 (23.6)
Hypertension1,551 (49.9)249 (55.2)317 (53.2)969 (51.4)411 (48.7)171 (44.8)
Kidney disease76 (2.4)13 (2.9)23 (3.9)50 (2.7)15 (1.8)11 (2.9)
Any joint surgery124 (4)28 (6.2)25 (4.2)73 (3.9)35 (4.2)16 (4.2)
At least 1 hospitalized infection335 (10.8)72 (16)112 (18.8)199 (10.6)89 (10.6)47 (12.3)
At least 1 outpatient infection1,492 (48)244 (54.1)319 (53.5)917 (48.7)386 (54.3)189 (49.5)
Prescription medication use      
Methotrexate1,697 (54.6)278 (61.6)347 (58.2)976 (51.8)498 (59.0)223 (58.4)
NSAIDs1,521 (48.9)221 (49)287 (48.2)909 (48.2)418 (49.5)194 (50.8)
Glucocorticoid dose      
None1,912 (61.5)236 (52.3)308 (51.7)1,187 (63.0)506 (60.0)219 (57.3)
1–7.5 mg/day695 (22.3)118 (26.2)128 (21.5)413 (21.9)194 (23.0)88 (23.0)
>7.5 mg/day504 (16.2)97 (21.5)160 (26.9)285 (15.1)144 (17.1)75 (19.6)
Any intraarticular injection727 (23.4)145 (32.2)183 (30.7)415 (22.0)188 (22.3)124 (32.5)
Any overlap between previous biologic and new biologic892 (28.7)119 (26.4)175 (29.4)585 (31.0)193 (22.9)114 (29.8)
RF or anti-CCP antibody positiveb1,885 (60.6)299 (66.3)433 (72.7)1,122 (59.5)527 (62.4)236 (61.8)
Calendar year of index date      
≤20061,442 (45.7)46 (10.2)95 (15.9)867 (46.0)364 (43.1)191 (50.0)
2007–2008710 (22.8)131 (29)198 (33.2)478 (25.4)163 (19.3)69 (18.1)
≥2009979 (31.5)274 (60.8)303 (50.8)540 (28.7)317 (37.6)122 (31.9)
Duration of exposure, mean ± SD days487 ± 571403 ± 387385 ± 337491 ± 562490 ± 590459 ± 573

Overall, there were 165 hospitalizations where the primary reason for hospitalization was a bacterial infection (Table 2). The most common types of infections were pneumonia, skin/soft tissue, genitourinary infections, and sepsis/bacteremia. The median time to infection from the index date was 250 days for anti-TNF, 251 days for ABA, and 195 days for RTX. The crude rate of hospitalized infections was 2.8 per 100 patient-years for ABA and 4.4 per 100 patient-years for RTX compared to a pooled estimate of 3.0 per 100 patient-years for all anti-TNF therapy treatment episodes (Table 3). Rates of hospitalized infection were lowest for ETA (2.2 per 100 patient-years) and highest for INF (4.8 per 100 patient-years).

Table 2. Sites and types of hospitalized bacterial infections among RA patients initiating abatacept, rituximab, or anti-TNF therapy*
InfectionsCohort, no. (%)
  1. RA = rheumatoid arthritis; anti-TNF = anti–tumor necrosis factor.
Pneumonia61 (37)
Cellulitis/soft tissue37 (22)
Kidney/urinary tract14 (9)
Bacteremia/sepsis11 (7)
Device-associated10 (6)
Gastroenteritis10 (6)
Septic arthritis7 (4)
Upper respiratory tract8 (5)
Abdominal abscess5 (3)
Osteomyelitis2 (1)
Total165 (100)
Table 3. Crude incidence rates of hospitalized bacterial infection among RA patients initiating abatacept, rituximab, or anti-TNF therapy*
Biologic agentsEvents, no.Person-years, no.Infection rate by biologic agent per 100 patient-years (95% CI)
  1. RA = rheumatoid arthritis; anti-TNF = anti–tumor necrosis factor; 95% CI = 95% confidence interval.
Abatacept14498.02.8 (1.7–4.7)
Rituximab28630.24.4 (3.1–6.4)
Anti-TNF agents1234,147.03.0 (2.5–3.5)
Etanercept251,133.42.2 (1.5–3.3)
Infliximab23480.24.8 (3.2–7.2)
Adalimumab752,534.03.0 (2.4–3.7)

Table 4 shows the multivariable-adjusted factors associated with hospitalized bacterial infection. Compared to ETA, ABA and RTX were not associated with a statistically higher increase in risk for hospitalized infections, although INF was. Older age, chronic obstructive pulmonary disease (COPD), and higher prednisone dosage (>7.5 mg/day) were associated with a significantly increased risk for infection.

Table 4. Multivariable-adjusted HRs for risks of hospitalized bacterial infections initiating abatacept, rituximab, or anti-TNF therapy*
Infection-related risk factorHR (95% CI)
  1. All factors in the table were adjusted and included in the multivariable model. HRs = hazard ratios; anti-TNF = anti–tumor necrosis factor; 95% CI = 95% confidence interval; COPD = chronic obstructive pulmonary disease; ESR = erythrocyte sedimentation rate; RA = rheumatoid arthritis; anti-CCP = anti–cyclic citrullinated peptide.
  2. aSubgroup analysis where this laboratory test measured at least once (16% excluded who never had an ESR laboratory test).
Medication exposure (referent to etanercept) 
Adalimumab1.4 (0.9–2.2)
Infliximab2.3 (1.3–4.0)
Abatacept1.1 (0.6–2.1)
Rituximab1.4 (0.8–2.6)
Age group (referent to <50 years) 
50 ≤ 70 years2.8 (1.4–5.4)
≥70 years2.7 (1.3–5.9)
COPD1.8 (1.2–2.7)
Body mass index (referent to 20–25 kg/m2) 
<20 kg/m21.9 (0.9–3.8)
>25 kg/m20.8 (0.6–1.1)
Prednisone-equivalent steroid dose (referent to no use) 
1–7.5 mg/day1.3 (0.9–1.9)
>7.5 mg/day1.8 (1.3–2.6)
Previous exposure to ≥3 biologic agents1.4 (0.7–2.6)
Hospitalizations during baseline period, no. 
None1.0 (referent)
11.4 (0.9–2.2)
≥24.1 (2.5–6.5)
ESR, mm/houra 
Quartile 1: ≤101.0 (referent)
Quartile 2: >10–221.1 (0.6–2.2)
Quartile 3: >22–421.9 (1.1–3.5)
Quartile 4: >424.1 (2.3–7.2)
Seropositive RA (rheumatoid factor and/or anti-CCP antibody)a0.8 (0.5–1.2)

Among the majority of patients who had laboratory tests for ESR, the third and fourth quartiles of ESR (22–42 mm/hour, and >42 mm/hour) were associated with a significantly increased risk for hospitalized infection compared to the lowest quartile (Table 4). Similarly, in a separate multivariable-adjusted model that did not include ESR and included patients who had CRP level tested at least once, the highest quartile of CRP (>1.8 mg/dl) was associated with a higher risk for hospitalized infection (hazard ratio [HR] 2.3, 95% confidence interval [95% CI] 1.4–3.8, referent to the first quartile of CRP level, <0.3 mg/dl [data not shown]). In the subgroup of patients with RF and/or anti-CCP antibody laboratory test results available (approximately 75% of patients), there was no significant association between seropositivity and hospitalized infection (HR 0.8, 95% CI 0.5–1.2 for seropositivity).

Results from the 3 sensitivity analyses were consistent with the main findings. Restriction of treatment episodes to 2006 and later, and analyses that explicitly required evidence of prior anti-TNF therapy for the ABA and RTX users (fewer than 5% of ABA/RTX treatment episodes excluded by this restriction), yielded similar HRs with slightly wider CIs than the main results (not shown). Additionally, the third sensitivity analysis that limited eligible patients to those who lived within 30 miles of the nearest VHA hospital likewise yielded consistent findings.

DISCUSSION

In this study of US veterans with RA, we found that the overall rates of hospitalized bacterial infections were comparable between patients initiating ABA or RTX compared to those initiating anti-TNF therapies with prior exposure to a different anti-TNF agent. Consistent with prior reports, older age, COPD, and concomitant higher steroid dose were significantly associated with hospitalized bacterial infections ([1, 3, 14]). We also found that higher levels of systemic inflammation, quantified by CRP level and ESR, were associated with hospitalized bacterial infections, although we recognize that elevations in these laboratory tests may or may not be directly related to RA disease activity.

Previous research has suggested that the safety profile of biologic agents that do not act on TNF may differ from anti-TNF therapies. A small, randomized controlled trial conducted for regulatory purposes found the rate of serious infections with ABA to be lower than with INF ([10]). Consistent with our findings in this study, an observational cohort study of younger, commercially insured RA patients found that ABA had a significantly lower rate of serious infections compared to INF, and ABA was similar to ETA and ADA ([12]). More recent trial results from a head-to-head comparison of ABA versus ADA in biologic-naive patients found rates of serious infections (predominantly requiring hospitalization) to be numerically lower, but not significantly different, for ABA compared to ADA ([11]). While the present analysis focused on an anti-TNF–experienced RA population rather than biologic-naive patients, we found crude incidence rates of serious infections that were numerically lowest for ABA and highest for INF. After multivariable adjustment, the rate of serious infections associated with ABA was not significantly different compared to ETA. Likewise, crude rates of serious infections were numerically higher for RTX users compared to those who switched to another anti-TNF agent, but were not significantly different after multivariable adjustment.

We found that systemic inflammation, measured by CRP level and ESR, was associated with higher rates of hospitalized bacterial infections. Previous studies have similarly observed that elevated acute-phase response (measured by ESR) was associated with serious infections in RA patients ([14, 15]). As ESR and CRP level may be a proxy for more active RA, this observation is consistent with findings from a study, published using clinical data from the Consortium of Rheumatology Researchers Registry, which observed that higher RA disease activity (measured by the Clinical Disease Activity Index [CDAI] [24]) was associated with serious infections ([25]), although this finding was confined to the lower end of the RA disease activity range. As the CDAI does not include an acute-phase reactant, the results observed in our analysis for CRP level and ESR are complementary to that result. Additionally, we observed a trend for underweight patients with lower BMI to have higher rates of serious infection. Underweight RA patients previously have been shown to have a high mortality burden ([26]). Seropositive RA patients did not have a significantly different rate of infection compared to seronegative patients. Seropositive RA has had an inconsistent association with serious infections in the published literature ([14, 15]).

Strengths of our study include examination of a cohort of RA patients using data captured from the largest and clinically richest integrated EHR system in the US linked to administrative data. Laboratory results, physician notes, pharmacy prescriptions, and coded data (e.g., diagnoses and procedures) are available in this data source and allow for examination of potential risk factors (e.g., CRP level, BMI, and seropositivity) that are not typically available when using only administrative claims data. Our findings also were robust across a range of important sensitivity analyses.

We were somewhat limited in our statistical power to compare specific biologic agents, given that approximately two-thirds of anti-TNF exposure in this cohort was ADA. This was likely due to our study design requirement that the anti-TNF users be previously exposed to another anti-TNF agent, that ADA was approved by the Food and Drug Administration after ETA and INF, and that ADA is the preferred anti-TNF agent in some VHA centers. Moreover, we recognize that potential differences in the comparative safety profile of various biologic agents may be difficult to detect given the relatively low absolute incidence rates of serious infections seen in most trials, registries, and the present observational study. It is possible that such a comparison among higher risk patients (e.g., those with a recent hospitalized infection) may yield different findings with better statistical power to detect true differences than observed in this and previous analyses.

Our results must be considered in light of some additional limitations. We intentionally restricted our outcome to infections occurring as the primary reason for hospitalization to improve specificity, although this limited the data available for analysis. This decision was made in order to improve the specificity of our main exposure–outcome associations ([1, 27]), but yielded a lower absolute incidence rate for infections. In addition, the RA population that we studied consisted predominantly of men and these results may not be generalizable to other RA populations. Patterns of discontinuation and switching in light of the duration of followup may differ in the VHA setting than in other countries or health care systems, which could affect results ([22]). Finally, we recognize the potential for under-ascertainment of infections occurring outside of the VHA system, although this seems unlikely to be differential by drug exposure; it did likely account for lower than expected absolute incidence rates of hospitalized infection observed in our study, which typically are in the range of 3–5 per 100 patient-years in younger RA cohorts and healthier RA clinical trial populations ([28]). The sensitivity analysis that restricted the population by distance to the nearest VHA center partially addressed this concern and supported our main findings.

In conclusion, among older, predominantly male US veterans with RA, the risks of hospitalized bacterial infections for patients treated with RTX or ABA were comparable to the risk of infection among veterans who were treated with ETA following prior exposure to a different anti-TNF therapy. Systematic inflammation measured by both ESR and CRP level was associated with hospitalized bacterial infections. The risk of hospitalized infection associated with biologic agents needs to be considered in conjunction with other infection-predisposing characteristics (e.g., glucocorticoid use and dose, and comorbidities) when communicating the safety profile of biologic agents to RA patients who have failed treatment with an anti-TNF agent and when selecting a new therapy.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Curtis had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Curtis, Patkar, Chen, Singh, Cannon, Delzell, Safford, Alexander, Napalkov, Kamauu, Baddley.

Acquisition of data. Curtis, Patkar, DuVall, Baddley.

Analysis and interpretation of data. Curtis, Yang, Patkar, Chen, Singh, Cannon, Mikuls, Delzell, Saag, Safford, DuVall, Alexander, Napalkov, Winthrop, Burton, Kamauu, Baddley.

ROLE OF THE STUDY SPONSOR

Authors from Anolinx and Genentech were involved in study design and interpretation of the data but had no role in the collection or acquisition of the data. Authors from Anolinx and Genentech were involved in the review and approval of the final manuscript submitted for publication. Publication of this article was not contingent upon approval by Anolinx and Genentech.

Ancillary