Nonserious Infections in Patients With Rheumatoid Arthritis: Results From the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis

To describe the frequency and predictors of nonserious infections (NSI) and compare incidence across biologic agents within the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis (BSRBR‐RA).


INTRODUCTION
Patients with rheumatoid arthritis (RA) experience a greater number of infections compared to the background population. These infections are frequent and contribute to substantial morbidity and mortality (1,2). Infection susceptibility is a combination of disease-related immunologic dysfunction, immunocompromising comorbidities, and the use of immunomodulatory drugs. It is also determined by patient lifestyle and other factors beyond the RA disease.
The risk of serious infections, defined as life-threatening infections or those requiring hospitalization or intravenous antibiotics, has been an important focus of long-term clinical trial extension studies and observational drug registries. Conventional synthetic disease-modifying antirheumatic drugs (csDMARDs) have relatively little impact (3,4), glucocorticoids consistently demonstrate a dose-dependent risk (5,6), and biologics are associated with a small but significant risk of serious infection (7)(8)(9)(10). Differences in risk observed between biologic agents have particular clinical relevance for patients considered to be "high risk" (8,10,11).
Serious infections are the tip of the iceberg. Nonserious infections (NSI), defined as those events managed outside of a hospital admission, have been reported in 20-30% of RA patients each year (1,12) and are the most common adverse events in large clinical trials. In elderly RA patients, rates of NSI are estimated at 47.5 per 100 patient-years (13). Although these events are not life-threatening, their burden is high (14), and recurrent NSI may lead to variable periods of treatment discontinuation (15). Metaanalyses of data on immune-mediated inflammatory diseases have suggested differences in the risk of NSI between tumor necrosis factor inhibitor (TNFi) agents (14), but the impact of other biologics and the predictors of such risk are less well understood.
Despite extensive literature on infection in RA, data on NSI are limited. To our knowledge, there has been little research on variables that predict NSI in patients with RA and the extent to which immunomodulatory drugs influence this risk. The primary objective of this study was to describe the frequency and pattern of NSI and to compare the incidence of NSI between biologic drugs within a large national registry.

PATIENTS AND METHODS
Patient population. Study subjects were participants in the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis (BSRBR-RA) (Appendix A), a national prospective observational cohort study established in 2001 to monitor long-term safety of biologic therapy. Initial biologic cohorts were for patients receiving etanercept and infliximab. The csDMARD cohort was recruited in parallel between 2002 and 2009. Subjects had moderate-to-severe disease activity but were not eligible for biologic treatment. Adalimumab, rituximab, tocilizumab, and certolizumab pegol cohorts were recruited beginning in 2004, 2008, and 2010, respectively, while JAK inhibitor (tofacitinib and baricitinib) and sarilumab cohorts were recruited beginning in 2017/2018. Abatacept and golimumab cohorts were not recruited. The BSRBR-RA methodology has been previously described in detail (16). Ethics approval was granted in 2000 (MREC no. 00/8/053 [IRAS no. 64202]). The data cutoff date for this study was January 2019.
Baseline assessment. Data collected at registration included demographic information, disease duration, smoking status, DMARD and glucocorticoid exposure, Disease Activity Score in 28 joints using the erythrocyte sedimentation rate (DAS28-ESR) (17), Health Assessment Questionnaire (HAQ) (18) scores, and comorbidities (yes/no) from a list. For analysis, comorbidity burden was scored using the Rheumatic Disease Comorbidity Index (19).

Follow-up.
Follow-up data were collected every 6 months for the first 3 years via questionnaires sent to patients and their supervising rheumatology teams, and annually thereafter via questionnaires sent only to the supervising rheumatology team. Data on adverse events were captured from clinician questionnaires: from patient diaries every 6 months and by linkage to NHS Digital, which provides mortality data. Patient diaries were provided for the first 3 years, in which patients were asked to record details of all new prescriptions (including antibiotics) and hospital attendances. Patient-reported serious adverse events required verification by the supervising rheumatology team. No additional verification of nonserious adverse events occurred, but all reported NSI were recorded in the database and coded.
Outcome measure. The primary outcome measure was an NSI reported to the BSRBR-RA by either the clinical team or the patient. Infections were coded using terminology from the Medical Dictionary for Regulatory Activities (MedDRA), and their severity was recorded according to the established MedDRA definition as an infectious episode that did not require hospitalization or intravenous therapy or lead to severe disability or death.
Exposure. Individuals were considered "at risk" from the date of beginning their first registered biologic treatment for up to 3 years, or until the date of treatment discontinuation, last received follow-up, or death, whichever came first. Censorship at 3 years was aligned to the time frame when diaries where collected, which was a key source of NSI. Patients could discontinue or switch therapies during the 3-year period, and all biologic exposure during this 3-year window was included. A switch to another biologic during this time would not extend the total follow-up window past 3 years, as diary collection terminated 3 years after registration. For example, if a patient started a subsequent biologic treatment after 2 years, they would only contribute a maximum of 1 year of exposure to this second biologic.
Due to the BSRBR-RA study design, hospitals had the option of re-registering existing study patients with the BSRBR-RA at the point of a patient switching to a therapy for which a cohort was actively being recruited. For example, a patient recruited in 2003 at the point of starting etanercept could then re-register in 2012 with a new study ID number when starting a new biologic treatment. All subsequent follow-up time would be transferred to the new study ID, but the 2 IDs would be linkable in the data set. This increased the frequency of follow-up and restarted diary capture for a further 3 years.
To account for ongoing exposure risk from the biologic's halflife after stopping therapy, an additional 90 days of exposure time was considered for all biologics. For rituximab, an additional 180 days of exposure time was considered, although in all cases it was censored at the maximum 3-year cutoff.
Statistical analysis. Crude incidence rates per 100 patient-years with 95% confidence intervals (95% CIs) were calculated. A multiple-failure Cox proportional hazards model was used to compare risk of NSI across groups, since many patients  * Except where indicated otherwise, values are the number (%) of patients. Due to study design, hospitals had the option of re-registering existing study patients with the British Society for Rheumatology Biologics Register for Rheumatoid Arthritis (BSRBR-RA) at the point of them switching to a therapy cohort that was actively recruiting patients. This occurred with 1,174 patients, 5% of the total cohort. Where this occurs, patients are included each time in the table. TNFi = tumor necrosis factor inhibitor; IL-6 = interleukin-6; RDCI = Rheumatic Disease Comorbidity Index; IQR = interquartile range; DMARD = disease-modifying antirheumatic drug; MTX = methotrexate; HCQ = hydroxychloroquine; DAS28-ESR = Disease Activity Score in 28 joints using the erythrocyte sedimentation rate; TJC = tender joint count; SJC = swollen joint count; PtGA = patient global assessment; CRP = C-reactive protein; HAQ DI = Health Assessment Questionnaire disability index.
† The total BSRBR-RA population includes patients commencing receiving JAK inhibitor therapy and anakinra. As these drug classes were excluded from the analysis, their individual baseline data are not presented here. ‡ Included chronic obstructive pulmonary disease and asthma. § Included ischemic heart disease and cerebrovascular accidents. ¶ The Cox proportional hazards model allowed patients to stop or switch therapies during the 3-year period. The data presented in this table refer to the patients in each drug cohort at BSRBR-RA registration and do not reflect the characteristics of patients who may have switched into a new drug cohort during the analysis window.
| 1803 experienced multiple events. A traditional (single-failure) model examining time to first event would ignore any additional infections, overlooking important information to enable us to understand risk. We therefore used a multiple-failure model, allowing patients to contribute more than 1 event and in which dependency in the hazard function was modeled as a shared frailty (i.e., random effect). Cluster-robust estimates for CIs were calculated. The risk of NSI were compared across biologic cohorts and reported as hazard ratios (HRs). The TNFi class was chosen as the referent for comparison, as it was the most widely used class of drug in the register. For analyses within the TNFi class, etanercept was used as the referent for comparison. Biosimilar treatment was not considered different from originator treatment, and all continuous exposure to the "same" drug was combined. Golimumab, abatacept, tofacitinib, and baricitinib were excluded from the analyses, as the number of patients receiving these medications was low or absent.
Potential confounders were selected a priori based on clinical knowledge and available variables. Adjustments included age, sex, disease duration, smoking, baseline DAS28-ESR, HAQ disability index (HAQ DI), steroid treatment, and year recruited to the BSRBR-RA. When a patient switched drugs, baseline characteristics were updated and reflected in the multivariate model. The number of biologic agents prescribed since registration was included as a time-varying covariate to adjust for the effect of switching treatments. A patient who switched biologics due to an infection had an increased risk of recurrent infection with their next drug (20). To account for competing risks and to adjust for clustering of events within individuals, the number of cumulative serious infections and NSI were also included as time-varying covariates. Assumptions of the Cox model were tested using Nelson-Aalen plots. Missing data were addressed using multiple imputation with chained equations for 20 cycles (Supplementary Methods and Supplementary  Tables 1 and 2, available on the Arthritis & Rheumatology website at http://onlin elibr ary.wiley.com/doi/10.1002/art.41754/ abstract). Results between the unimputed and imputed models were compared. Analyses were undertaken using Stata 15. Sensitivity analysis. Analyses using different drug exposure windows, limited to "on-drug time only" (excluding the 3-or 6-month half-life exposure risk) and also extended to an "everexposed" model until point of switch, were compared. Risk of NSI by method of ascertainment was also examined. To account for patients who registered a second time within the BSRBR-RA and contributed to more than 1 drug cohort, we recalculated SEs using the cluster-robust sandwich estimator, accounting for the within-person correlation across these different observations. To account for the effect of serious infection, sensitivity analyses using a single-failure model were performed incorporating serious infection as a competing risk using the Fine and Gray method (21).

Patient characteristics.
A total of 23,584 patients were registered in the BSRBR-RA until January 2019. The baseline characteristics are listed in Table 1. The mean age was 57 years, and the median disease duration was 10 years. The median baseline DAS28-ESR was 6.10 (interquartile range [IQR] 5.29-6.91), which is reflective of a biologic initiation cohort.
Patients were asked to return a diary every 6 months during follow-up. Diaries were received from 15,205 of 23,584 patients (64.5%). Of the patients who returned a diary during the first 3 years (the exposure window for the Cox models), 63% returned more than two-thirds of the required diaries, while 16% returned fewer than one-third. Diary return was slightly lower among the IL-6 cohort and among smokers (Supplementary Table 3 NSI. There were 17,304 nonserious infective episodes in 8,145 patients during the 3 year follow up period ( Table 2). The median number of infections per patient was 1 (IQR 1-3). Respiratory infections accounted for 36% of all NSI. Urinary, ear, nose, and throat, and skin infections were the next most frequently reported. Nonserious opportunistic infections were reported, with herpes zoster (n = 224) and candidiasis (n = 373) being the most frequent.
Limited to the on-drug time during the first 3 years of follow-up (the exposure window for the Cox models), there were 27.0 NSI events per 100 patient-years of follow-up (95% CI 26.6-27.4) in the multiple-failure model (Table 3). Increasing age, female sex, comorbidity burden, glucocorticoid therapy, higher RA disease activity (defined by the DAS28-ESR), and greater disability (recorded by the HAQ DI) were associated with an increased risk of NSI. Compared to never smokers, current smokers had a lower risk of NSI. Patients recruited into the BSRBR-RA in more recent years also had a lower NSI risk (Table 4). Using a single-failure model, there were 12.7 events * The Cox proportional hazards model allowed patients to stop or switch therapies during the 3-year period. The follow-up time (in person-years) reflects the amount of time exposed to each drug during the analysis window. NSI = nonserious infection; 95% CI = 95% confidence interval; csDMARD = conventional synthetic disease-modifying antirheumatic drug; TNFi = tumor necrosis factor inhibitor; anti-IL-6R = anti-interleukin-6 receptor.  Table 3 for other definitions). † P < 0.001. ‡ P < 0.01. § P < 0.05.
NSI risk according to biologic treatment. The incidence rates of NSI according to biologic treatment class and within the TNFi class are shown in Table 3. Anti-IL-6 receptor (28.3 cases per 100 patient-years) and rituximab (33.6 cases per 100 patient-years) treatments were associated with a higher risk of NSI compared to TNFi (adjusted hazard ratio  (Table 4 and Figure 1). Each biologic agent was associated with a greater risk of NSI when compared to the biologics-naive cohort receiving csDMARDs alone (Supplementary Tables 4 and 5  The follow-up time (in person-years) reflects the amount of time exposed to each drug during the analysis window. See Table 3 for definitions. Adalimumab treatment had a higher risk of NSI compared to etanercept (adjusted HR 1.11 [95% CI 1.05-1.18]). In the unadjusted model, compared to etanercept, infliximab had a higher risk of NSI while certolizumab had a lower risk, although this did not remain significant in the multivariable analysis (Table 6 and Figure 1).

Sensitivity analyses.
Further analyses were performed by examining different exposures, including an on-drug timeonly model and an ever-exposed model (Supplementary Table 6, http://onlin elibr ary.wiley.com/doi/10.1002/art.41754/ abstract) and by examining NSI risk by method of ascertainment (patient-reported, n = 8,991; consultant-reported, n = 7,375; and patient and consultant-reported, n = 930) (Supplementary Table 7, http://onlin elibr ary.wiley.com/doi/10.1002/art.41754/ abstract). These analyses demonstrate estimates consistent with those of the primary analysis. To account for patients who were registered a second time and contributed to more than 1 drug cohort, SEs were recalculated using the cluster-robust sandwich estimator. This made no difference to the estimated confidence intervals or P values, and thus the interpretation appears robust (Supplementary Table 8

DISCUSSION
To date, NSI have received little attention in the research literature and are an underappreciated component of disease burden in RA. In this large cohort, we have demonstrated a high frequency of NSI, affecting more than 1 in 10 patients annually. For every 100 patients, we report a rate of 27 nonserious events per year. This rate is comparable to that observed in other observational studies (12). Patients experience multiple infectious episodes, with re spiratory infections being the most frequent.
The risk factors for developing an NSI are comparable to those observed in patients with serious infections (4,12,22). This includes increasing age, comorbidities, and RA disease severity. By contrast, the impact of smoking on NSI risk is distinct from what is seen with serious infections. Interestingly, being a current smoker is associated with a lower risk of NSI. It is possible that a smoker with an infection is less likely to be managed as an outpatient compared to a nonsmoker. Indeed, cigarette smoking is a significant risk factor for severe viral and bacterial infection (23) and for inpatient admission when presenting with infective symptoms (24). Smokers are susceptible to developing chronic lung disease, which is also associated with increased hospitalization, especially in the presence of infective respiratory symptoms (6,25). It is also possible that smokers underreport their infections, perhaps attributing an NSI to a chronic cough. Finally, this may be due to reporting bias as current smokers had a lower diary return rate, and we assumed that non-return indicated no infection.
There was a 5% reduction in risk of NSI for each subsequent year patients were recruited into the BSRBR-RA. The rate of infections in RA patients appears to be changing over time. This has been described with serious infectious events (26) and likely reflects shorter RA disease duration and a lower disease burden. This could be artefactual, as diary return rates have reduced in recent years.
Our findings demonstrate that biologics are likely to be associated with an increased risk of NSI. The csDMARD cohort had the lowest infection rates. There was a 40% decrease in risk of NSI with csDMARDs compared to TNFi. This is consistent with findings from the Corrona registry, in which TNFi was associated with an increased rate of outpatient infections (12). It also mirrors observations from studies examining serious infection in the BSRBR-RA (7,26) and other observational cohorts (8,9,27,28), although the magnitude of NSI risk is far greater.
Comparisons of the risk of NSI between different biologic drugs reveal similar patterns to those seen with serious infection (11). The risk was greater with rituximab compared to TNFi. IL-6 inhibition with tocilizumab therapy was also associated with a greater risk of NSI after adjusting for both patient and disease factors. It is biologically plausible that IL-6 inhibition would be associated with infection risk. This pleomorphic cytokine has a vital role in the defense against numerous pathogens, especially bacteria and fungi, as demonstrated in primary immunodeficiency diseases linked to IL-6 or its signaling pathways (29). Studies analyzing serious infections have demonstrated an increased risk with tocilizumab compared to TNFi in the BSRBR-RA (compared to etanercept, tocilizumab demonstrated an HR of 1.22) (11) and in the German biologics registry (30). While this finding was not We have also demonstrated that the rates of NSI differ within the TNFi class. The highest rates were reported with infliximab and adalimumab. Compared to etanercept, adalimumab was associated with a greater risk of NSI. This differential NSI risk with the monoclonal TNFi (infliximab and adalimumab) compared to the soluble TNF receptor antagonist (etanercept) has been demonstrated previously. A meta-analysis of placebo-controlled RCTs in the treatment of immune-mediated inflammatory diseases showed the lowest number of NSI events with etanercept. The authors estimated a 20% higher risk with infliximab and adalimumab, compared to placebo, than what was observed with etanercept (14). This differential finding was also reported with herpes zoster in the German registry (33) but not in the BSRBR-RA analysis (34).
Our study has several strengths. The first is attributable to the size and quality of real-world data that the BSRBR-RA provides. There are limited missing data on baseline covariates and accurate coding of biologics. Adverse event capture data is robust, obtained from multiple sources permitting the evaluation of nonserious events. The use of TNFi rather than csDMARDs as the comparator arm allows for the comparison across biologic agents. This is more clinically relevant for physicians who are considering therapeutic options in patients who have not responded to csDMARDs. Last, the use of particular statistical models has built on decades of registry analyses, learning how to handle complex data sets with time-varying components and significant confounding.
We acknowledge several important limitations. We are unable to comment on the risk of NSI with certain agents, as few patients were registered having received these medications. This includes golimumab and abatacept, as these cohorts were never recruited to the BSRBR-RA, as well as the JAK inhibitors tofacitinib and baricitinib, which have only been recruited since 2017/2018. We cannot account for national guidelines, drug costs, and local treatment pathways, which influence decisions on medication choice.
We describe NSI as reported to the BSRBR-RA but must acknowledge that the mode of data capture for such events is inevitably incomplete and prone to misclassification bias and reporting bias. The rates of infection are likely to be underestimated, but the HRs should be unbiased as there was no differential reporting by drug. The definitions of NSI are less robust than for serious infections. As we did not require a documented antibiotic prescription, a proportion of the events may not have been of infectious etiology. Similarly, only NSI requiring antibiotics were reported by patients in their diaries, and some infectious events, such as viral infections, may not have been captured at all. It is unlikely that misclassification or missed events differs significantly across the treatment groups, as identical capturing mechanisms were employed, although there is still a risk of reporting bias between biologic agents and csDMARDs. The proportion of patients returning diaries has reduced over time, which may also introduce bias. However, a high rate of NSI was seen with IL-6 inhibition, a drug cohort recruited to the BSRBR-RA in more recent years, with a lower rate of diary return. If anything, we may be underestimating the risk of NSI with IL-6 and biasing toward the null hypothesis. Finally, despite adjusting for baseline variables that predict NSI, there is a possibility that some degree of confounding persists.
In conclusion, NSI events are common in patients with RA, with similar predictors to those observed with serious infections. An NSI history should be routinely captured in clinical practice. Biologics are associated with a greater risk of NSI, with differences in incidence and risk between treatments. These results provide clinicians with information on how to identify patients at a greater risk of NSI and guide them on the best possible treatment strategies.