Objectives To determine among patients with autoimmune diseases in the USA whether the risk of non-viral opportunistic infections (OI) was increased among new users of tumour necrosis factor α inhibitors (TNFI), when compared to users of non-biological agents used for active disease.
Methods We identified new users of TNFI among cohorts of rheumatoid arthritis (RA), inflammatory bowel disease and psoriasis-psoriatic arthritis-ankylosing spondylitis patients during 1998–2007 using combined data from Kaiser Permanente Northern California, two pharmaceutical assistance programmes for the elderly, Tennessee Medicaid and US Medicaid/Medicare programmes. We compared incidence of non-viral OI among new TNFI users and patients initiating non-biological disease-modifying antirheumatic drugs (DMARD) overall and within each disease cohort. Cox regression models were used to compare propensity-score and steroid- adjusted OI incidence between new TNFI and non-biological DMARD users.
Results Within a cohort of 33 324 new TNFI users we identified 80 non-viral OI, the most common of which was pneumocystosis (n=16). In the combined cohort, crude rates of non-viral OI among new users of TNFI compared to those initiating non-biological DMARD was 2.7 versus 1.7 per 1000-person-years (aHR 1.6, 95% CI 1.0 to 2.6). Baseline corticosteroid use was associated with non-viral OI (aHR 2.5, 95% CI 1.5 to 4.0). In the RA cohort, rates of non-viral OI among new users of infliximab were higher when compared to patients newly starting non-biological DMARD (aHR 2.6, 95% CI 1.2 to 5.6) or new etanercept users (aHR 2.9, 95% CI 1.5 to 5.4).
Conclusions In the USA, the rate of non-viral OI was higher among new users of TNFI with autoimmune diseases compared to non-biological DMARD users.
- DMARDs (biologic)
- Rheumatoid Arthritis
Statistics from Altmetric.com
Biological immunosuppressive therapies such as tumour necrosis factor (TNF) α inhibitors (TNFI) represent important treatment advances for patients with rheumatoid arthritis (RA) and a number of other inflammatory conditions. Although these drugs have revolutionised the treatment of inflammatory and rheumatological disorders, there is an important safety issue: a potential increased risk of infection caused by a broad spectrum of organisms.1–5
Infectious complications of biological medical therapies are often subgrouped into severe or opportunistic. We and others have demonstrated that at least some TNFI are associated with an increased risk of serious infections compared to non-biological therapies.1 ,2 ,4 ,6 Additional reports suggest a possible increased risk of opportunistic infections (OI), including diseases such as tuberculosis and systemic mycoses, legionellosis and progressive multifocal encephalopathy, among patients treated with TNFI.7–12 It remains difficult to ascertain the magnitude and significance of the risk because of the rarity of infections and variability in groups under study. For example, a recent report from the French registry estimated the annual sex and age-adjusted incidence rates of all non-mycobacterial OI to be 1.52 per 1000 person-years.5 In contrast, a rate of OI (herpes zoster and tuberculosis included) of 30 per 1000 person-years among patients using TNFI was estimated with data from the CORRONA registry.13 These studies differed in methodologies, grouping of infectious outcomes, comparator groups and patient populations (RA vs all TNF indications). In addition, most analyses have focused on prevalent, not new users. Therefore, questions remain about the impact of specific TNF agents versus non-biological therapy and the risk of infection associated with TNFI for patients with autoimmune diseases other than RA.
As part of a multi-institutional US initiative, the Safety Assessment of Biologic Therapy (SABER) project, we investigated among patients with RA and other autoimmune or inflammatory diseases whether the risk of non-viral OI was increased among new users of TNFI, when compared to new users of non-biological agents.
This retrospective cohort study combined data from 1998 to 2007 from four large US data systems.14 Exposure to TNFI and other disease-modifying antirheumatic drugs (DMARD) was determined using pharmacy files and procedure codes (for infusions). OI were identified using hospital and outpatient diagnoses. The incidence of OI between exposure groups was compared using Cox proportional hazard regression models.
Data from four data systems (National Medicaid and dual Medicaid-Medicare databases, TennCare, New Jersey's Pharmaceutical Assistance to the Aged and Disabled and Pennsylvania's Pharmaceutical Assistance Contract for the Elderly programmes linked to Medicare data, and Kaiser Permanente Northern California) and a common data model were used to assemble the cohort of patients who were new users of selected biological and non-biological DMARD.14 Within each data system, patients with autoimmune diseases were identified as those with an International Classification of Diseases, 9th edition, clinical modification (ICD9-CM)-coded healthcare encounter for an autoimmune disease followed by a prescription filled for, or infusion of, a study DMARD. We required availability of a baseline period of 365 days with continuous enrolment in the respective data system preceding the first qualifying new drug prescription fill or infusion (described below), for ascertainment of study covariates.
Patients were categorised in three mutually exclusive autoimmune disease groups: RA, inflammatory bowel disease (IBD), and psoriasis, psoriatic arthritis (PsA) and ankylosing spondylitis (AS), PsO-PsA-AS. Within the pool of potential cohort members, we identified new users of study medications, defined as having filled one prescription or infusion for a study DMARD after 365 baseline days without prescriptions filled for the specific study medication(s). This ‘first’ filling or infusion date was termed time zero and marked the beginning of follow-up. Potential RA cohort members were further required to be aged 16 years or older.
Within each cohort, an episode of new medication use began on time zero and follow-up continued through the earliest of the following: death, loss of enrolment, study outcome, switch to a new DMARD regimen, or discontinuation of current regimen (defined as 30 days without medication). Patients could contribute additional episodes of new medication use for a different medication (in the same or alternative exposure group) if they fulfilled the eligibility criteria again.
Claims data on pharmacy prescription fills and infusions were used to determine medication exposure following a new user design. Study DMARD were classified in two groups: TNFI (including infliximab, adalimumab and etanercept (not included for IBD)) and alternative non-biological DMARD regimens. For RA, alternative regimens were initiation of leflunomide, sulfasalazine or hydroxychloroquine after the use of methotrexate in the previous year (ie, methotrexate failures), whereas for IBD, the comparison group was initiation of azathioprine or 6-mercaptopurine (AZA/6-MP). For PsO-PsA-AS the comparison was initiation of non-biological DMARD (methotrexate, hydroxychloroquine, sulfasalazine and leflunomide).
For all groups, exposed person-time encompassed all follow-up person-time covered by prescription fills (and using 56 days for infliximab) and an additional 30 person-days without subsequent medication available. This 30-day grace period was allowed because some residual effects of study medications could extend beyond the last day of use and to account for imperfect adherence. Therefore, this approach allowed a short gap in which outcomes identified after drug supply exhaustion could be related to the most recent exposure. Both TNFI and the non-biological DMARD regimens allowed the concurrent use (continuation or addition) of methotrexate. Analyses of IBD allowed for continuation of or simultaneous initiation of AZA/6-MP in the TNFI group.
The primary outcome was non-viral OI. Additional analyses were performed for the subgroups of tuberculosis and non-tuberculous mycobacterial (NTM) disease patients. For fungal infections, we used primary or non-primary discharge diagnoses or an outpatient diagnosis for histoplasmosis, coccidioidomycosis, cryptococcosis, blastomycosis, or aspergillosis plus an outpatient prescription for at least 30 days of any active systemic antifungal drug (itraconazole, fluconazole, voriconazole).15 Tuberculosis required an inpatient or outpatient ICD-9 diagnosis code (018.x) plus pharmacy records indicating prescription of pyrazinamide prescribed within 90 days of the diagnosis code. Diagnosis of other non-viral OI (pneumocystosis, nocardiosis/actinomycosis, non-tuberculous mycobacteria, salmonellosis, listeriosis, legionellosis) required an inpatient or outpatient physician ICD-9 diagnosis code without concomitant anti-infective medication.
Baseline covariates were measured, including demographics: age, gender, race, residence (urban and rural), nursing home/community dwelling, median area income, calendar year; generic markers of co-morbidity: number of hospitalisations, outpatient and emergency room visits, medication classes filled during baseline; markers of disease severity (extra-articular manifestations of disease, number of intra-articular and orthopaedic procedures), number of laboratory tests ordered for inflammatory markers, baseline use of DMARD; previous hospitalisation due to infection, chronic obstructive pulmonary disease, diabetes and previous use of antibiotics.
Baseline use of oral glucocorticoids was categorised according to the estimated average daily dose of prednisone equivalents (none, >0–<5 (low dose), 5–10 (medium dose) and >10 mg (high dose) averaged in the 6 months before time zero. This covariate was not included in the propensity score (PS) so as to be able to estimate its association with infection in outcome models.
The effects of potential confounders were controlled for using PS quintiles and baseline corticosteroid use within the past year before drug exposure. Covariates included in the PS derivation can be found in supplementary table S1 (available online only). Within each of the four data systems, logistic regression models estimated the site-specific PS for each episode of use within each study disease. A single value summarised covariate information for each medication episode.14 Visual inspection of the distribution of predicted probabilities across exposure groups showed substantial overlap of PS distributions, indicating that identification of patients with similar covariate distributions for each comparison was feasible. Non-overlapping regions (approximately 1%, 4% and 8% of RA, IBD and PsO-PsA-AS patients, respectively) of the PS were trimmed within each data system.
Cox-proportional hazard regression models assessed the association between exposure groups and study outcomes, with stratification by study site to allow the baseline hazard to vary. Within the RA cohort we evaluated the risk of OI associated with individual biological therapies, specifically infliximab and adalimumab (using etanercept as referent). Because patients could contribute more than one treatment episode, SE were adjusted using the Huber–White sandwich estimator. The final disease-specific outcome models for the overall cohort and the RA cohort evaluating specific TNFI included exposure groups, adjustment for PS quintile and baseline glucocorticoid use 1 year before time zero. All analyses were performed in SAS V.9.13. This study was approved by the institutional review boards of Vanderbilt University, Kaiser Permanente, Brigham and Women hospital, the University of Pennsylvania, and the University of Alabama at Birmingham.
Cohort assembly and baseline characteristics
We identified 407 319 potentially eligible patients with autoimmune diseases in the respective study databases, of which 170 788 (42%) patients were excluded due to having more than one autoimmune disease or autoimmune diseases other than RA, IBD, psoriasis, PsA or AS. We identified 36 212 (RA), 10 717 (IBD) and 12 137 (PsO-PsA-AS) patients who were either new users of TNFI therapy or a comparator non-biological DMARD. Within each disease group, baseline demographics and covariates were relatively similar between TNFI and non-biological DMARD users (table 1). The median (IQR) follow-up time in the TNFI and non-biological groups was 170 (299) and 104 (166) days, respectively.
Across all disease indications, we identified 107 OI (80 in new TNFI users; table 2). The most common were pneumocystosis (n=18), nocardiosis/actinomycosis (n=12) and tuberculosis (n=10; table 2). Of these cases, 74 (69.1%) patients used corticosteroids at the time of the OI event, and 27 (25.2%) patients were receiving methotrexate. Among TNFI users, the median time to infection post-initiation of TNFI was 131.5 days (range 9–1503 days); 56% of patients who developed an OI did so within 6 months of TNFI initiation. In the combined disease cohort (including RA, IBD and PsO-PsA-AS) and for each specific disease cohort, crude OI incidence rates were higher among those starting TNFI therapy versus the comparator group. For the combined disease cohort, rates among TNFI versus comparator patients were 2.7 versus 1.7 per 1000 person-years (adjusted HR (aHR) 1.6, 95% CI 1.0 to 2.6; table 3). Baseline glucocorticoid use was also associated with non-viral OI (aHR 2.5, 95% CI 1.5 to 4.0).
Compared with non-biological DMARD patients, the adjusted risk of non-viral OI in RA patients was not increased significantly among new users of any TNFI (aHR 1.6, 95% CI 0.9 to 3.1; table 3). Glucocorticoids had a borderline significant association with non-viral OI (aHR 1.8, 95% CI 1.0 to 2.8). There were differences among specific TNFI: infliximab (aHR 2.6, 95% CI 1.2 to 5.6) when compared with non-biological DMARD was associated with an increased risk of non-viral OI (table 4). In comparisons between specific TNFI, infliximab initiation was associated with an increased risk of non-viral OI when compared to etanercept (aHR 2.9, 95% CI 1.5 to 5.4).
In the mycobacterial analysis (table 5) most cases occurred in TNFI users with a crude rate four times greater for tuberculosis in this stratum, although the difference did not reach statistical significance (aHR 4.2, 95% CI 0.5 to 33.5). Rates of NTM were similar between exposure groups (table 5).
Within SABER, a USA multi-institutional research initiative, we studied rates and frequency of non-viral OI among a large cohort of patients with selected autoimmune diseases. Pneumocystis and mycobacterial infections accounted for almost half of the OI occurring among new users of TNFI therapy, and the majority of OI occurred within 6 months of TNFI initiation. Across disease indications, non-viral OI occurred more frequently among patients initiating TNFI agents when compared to those starting non-biological DMARD. Glucocorticoid use in the baseline period was associated with an increased risk of non-viral OI.
Although the topic of biological therapy and OI risk is an important safety issue, little epidemiological work to document frequency and relative risk, particularly with regard to specific organisms other than tuberculosis, has been completed.5 ,13 ,16 ,17 Our study identified rates of non-viral OI similar to those described in European studies, but our estimates are lower than another North American study that has evaluated this question, perhaps explained by the definition of OI used.13 The British registry identified rates of ‘intracellular infection’ of 2.2 per 1000 person-years.6 The French registry recently published rates of ‘non-tuberculosis OI’ of 1.56 per 1000 person-years, and previously identified a rate for tuberculosis of 1.2 per 1000 person-years among users of anti-TNF agents.5 ,18 Our overall rate of 2.7 per 1000 person-years is comparable for new users of TNFI therapy, although the choice of OI included within the analyses and methods differed slightly. While the French study included herpes zoster but excluded tuberculosis among its outcomes, the British registry included all such infections. We, however, excluded zoster but included mycobacterial infections. Data from the CORRONA collaboration in North America, which included zoster and mycobacterial infections, revealed 10-fold higher rates of OI than our study, which is probably accounted for by the inclusion of zoster.13
Although tuberculosis has been reported previously as the most frequent OI in this setting, our data suggest that pneumocystosis occurs more frequently than other OI in the USA among TNFI users. Relatively high rates of pneumocystosis have been reported previously from Japan among TNFI users, and sometimes more commonly than tuberculosis among infliximab-treated patients.19–21 However, it is possible that differences in case definition and more sensitive diagnostic procedures typically used in clinical practice (eg, PCR) may influence higher observed rates in that region when compared to our findings. In addition, we may have underestimated the number of tuberculosis cases due to suboptimal sensitivity of diagnostic codes. Moreover, screening practices for latent tuberculosis in the USA may have led to a smaller number of cases. In our overall population, rates for Pneumocystis infection among TNFI users (0.56 per 1000 person-years) and non-biological DMARD users (0.51 per 1000 person-years) were similar, and were probably influenced by concomitant use of glucocorticoids. On the basis of the low rates of pneumocystosis identified in our study, it is doubtful that Pneumocystis prophylaxis in US TNFI users would be cost effective.
Before our study, Winthrop and colleagues22 surveyed infectious disease specialists across the USA, the results of which suggested histoplasmosis to be more common than mycobacterial disease in the setting of biological therapy. Our study identified histoplasmosis as the fourth most frequent OI. Several of the cohorts we evaluated included patients who did not live within traditional histoplasmosis-endemic areas, which may explain a lower than expected rate within our study.23 Furthermore, for histoplasmosis and other fungal infections, we required both a diagnostic code and antifungal use to define a case. It is likely that the expected improvement in specificity diminished the sensitivity of our case-finding algorithms; there were patients with diagnostic codes that lacked anti-fungal usage.
While French and British studies have reported rates of OI to be higher in patients using TNFI therapy, it is likely that only a subset of OI are influenced by TNF blockade. Those infections by which the host relies on granulomatous response occur more commonly in those using TNFI, and biological mechanisms from animal models and in-vitro studies support this association.24 ,25 It is not clear that other OI, for which the host immune response might be different, are similarly affected. Furthermore, even within Mycobacterium species, in which tuberculosis and NTM both trigger granulomatous responses, there might be a differential risk associated with TNF blockade, as observed in our study. Patients can be effectively screened for tuberculosis and disease prevented, while no such screening exists for NTM. We suspect there exists a potentially strong confounding by indication bias in our evaluation of NTM risk with TNF blockade, in that most patients with pulmonary NTM are those who have pre-existing lung disease.26 ,27 It is possible that rheumatologists and other specialists are more likely to avoid biological therapies in those patients with baseline severe underlying lung disease.
Use of the monoclonal antibodies infliximab or adalimumab, when compared to etanercept, have been associated with an increased risk of bacterial infections, tuberculosis and other granulomatous diseases in some but not all studies.4 ,10 ,25 ,28–31 In addition, a variety of biological mechanisms has been identified that potentially explain this risk differential.24 ,28 When evaluating our RA population, we found a similar increased risk for non-viral OI, strongest among new infliximab users. Of note, we adjusted for baseline glucocorticoid use but did not control for time-varying risks such as differences in prednisone use or methotrexate after drug initiation. This avoided inappropriate adjustment for downstream factors that could be causally related to TNFI initiation.
Our findings must be interpreted in the light of several limitations. We relied on administrative data to identify diseases and outcomes, potentially resulting in misclassification of events. Some measures of disease activity (ie, the disease activity score in 28 joints) were not available in the dataset. We did not include Candida infections due to the poor specificity of diagnostic codes,15 and we did not include viral infections such as herpes zoster, as it was the topic of another report.32
With the use of these administrative diagnostic codes, it is also likely we underestimated the incidence of several OI. From a related project conducted within Kaiser Permanente that utilised microbiology data to find cases of tuberculosis and NTM, we found NTM diagnostic codes to be only 50% sensitive in detecting cases of disease.33 For tuberculosis diagnosis in this study we required pyrazinamide along with a code for tuberculosis; however, analyses conducted using the Kaiser Permanente and TennCare data suggested that such approaches have suboptimal sensitivity and specificity for tuberculosis.33 ,34 The ICD-9 code for tuberculosis is used frequently when patients are diagnosed with latent tuberculosis infection, making identification of active tuberculosis difficult. Despite the potential for outcome misclassification, it is unlikely that this would be differential by drug exposure and the incidence rates would probably be unbiased.
In conclusion, using a US multi-institutional cohort, non-viral OI among new TNFI users were increased compared with initiators of alternative non-biological DMARD in patients with autoimmune diseases. Baseline corticosteroid use was associated with the risk of non-viral OI. Among patients with RA, infliximab was associated with an increased risk of OI compared with etanercept. Pneumocystis and mycobacterial infections remain important OI among patients with autoimmune diseases receiving TNFI.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online table
Handling editor Tore K Kvien
Previous presentation This work was presented in part at the 49th Infectious Diseases Society of America Annual Meeting, October 2011, Boston, Massachusetts.
Contributors LC had full access to all of the study data and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: JWB, KLW, CGG, LC, ED, LL, FX, LJH, DHS, JDL, KGS and JRC. Acquisition of data: KGS, JRC, LJH and DHS. Analysis and interpretation of data: JWB, KLW, LC, ED, LL, FX, TB, KLW, LJH, NMP, DHS, JDL, KGS and JRC. Drafting of the manuscript: JWB and KLW. Critical revision of the manuscript for important intellectual content: JWB, KLW, LC, ED, LL, CGG, TB, LJH, DHS, JDL, FX, KGS and JRC. Statistical analysis: LC. Obtaining funding: KGS, JRC and LJH. Administrative, technical, or material support: NMP. Study supervision: NMP, KGS and JRC. JWB and KLW are co-primary first authors.
Funding This work was supported by the US Food and Drug Administration (FDA), US Department of Health and Human Services (DHHS) and the Agency for Healthcare Research and Quality, grant U18 HS17919. KLW's work on this manuscript was funded by an Agency for Healthcare Research and Quality (AHRQ) grant (1K08HS017552–01). JRC receives support from the National Institutes of Health (AR053351) and AHRQ (R01HS018517). TB was supported by NIH grant 5KL2 RR025776-03 via the University of Alabama at Birmingham Center for Clinical and Translational Science. CGG received support from the National Institute of Arthritis and Musculoskeletal and Skin Diseases, grant 5P60AR56116. The authors are indebted to the Tennessee Bureau of TennCare of the Department of Finance and Administration, which provided the Tennessee Medicare data.
Disclaimer The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the FDA or AHRQ.
Competing interests JWB reported consulting for Pfizer, Merck, Astellas and Mayne Pharma and received research support from Bristol-Myers Squibb. KLW reported grant/research support from Pfizer and consulting for UCB, Pfizer and Genentech. ED received grant/research support from Amgen. TB received grant/research support from Pfizer and consulting for Novartis, Roche and Genentech. LJH had research support from Genentech, Centocor, and Procter and Gamble. DHS received research support from Abbott, Amgen and Bristol-Myers Squibb. KGS received grant support from ACAR, Ardea, Savient, Takeda, Regeneron and was a consultant for Ardea, Regeneron, Takea, Savient. JRC received grant/research support from Roche/Genentech, UCB, Janssen, CORRONA, Amgen, Pfizer, BMS, Crescendo, AbbVie and was a consultant for Roche/Genentech, UCB, Janssen, CORRONA, Amgen, Pfizer, BMS, Crescendo, AbbVie. Other authors declare no competing interests.
Ethics approval This study was approved by the institutional review boards of Vanderbilt University, Kaiser Permanente, Brigham and Women hospital, the University of Pennsylvania, and the University of Alabama at Birmingham.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The data from this study are the property of the FDA and AHRQ and are not currently available.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.