As placebo is the best comparator but is generally limited to use within the double-blind phase of an RCT, this study provides context for infections requiring hospitalization in patients with an inadequate response to methotrexate or an anti-TNF agent who were treated with long-term exposure to abatacept in the clinical development program.
Bongartz and Saillot has shown an increase in serious infections in the double-blind phase of biologic studies when compared to placebo [3, 23]. Previously stated, results from the double-blind randomized trials suggest an increase in serious infections: 3.47/100 patient-years (py) and 2.41/100 py for abatacept and placebo, respectively  (rate ratio 1.44 (0.86 to 2.42)). This serious infection rate ratio of 1.44 (0.86 to 2.42) is within the lower limit of the 95% CI for the pooled odds ratio reported in the meta-analysis of serious infections in anti-TNF controlled trials by Bongartz (pooled Mantel-Haenszel OR 2.0 (1.3 to 3.1))  and similar to the fixed combined OR of 1.35 (0.79 to 2.32) reported by Saillot .
The incidence rate for hospitalized infections (a subset of serious) in the DB periods of the RCTs was also increased however again not significant: 3.05/100 py for abatacept and 2.15/100 py for placebo (rate ratio 1.42; 95% CI: 0.82 to 2.45).
The cumulative data presented here suggest that the observed incidence rates of infections requiring hospitalization (hospitalized infections) hospitalized pneumonia, and TB in patients ever treated with abatacept in the cumulative CDP (both double-blind and open-label) is in the same order of magnitude as those expected from cohorts of RA patients treated with non-biologic DMARDs.
As patients who are enrolled in RCTs move from the double-blind to open label phases, the lack of a control group to monitor the occurrence of adverse events in the long term becomes challenging. It is difficult to assess whether the risk of infection in patients with RA is due to underlying RA, as suggested by Dobloug, Fox and Koetz [24–26], or by the treatments administered. Some studies have demonstrated that the use of corticosteroids may increase infection risk [19, 27–29] as well as anti-TNF agents given to patients with RA [30–32]. With potential confounding by disease activity, disease severity, duration of disease, presence of comorbid conditions, and concomitant treatment, it becomes challenging to determine a true association between the risk of infections and a new treatment.
To date, there have not been any similar such studies evaluating cumulative exposure of a biologic agent to an external cohort. There have been a number of observational studies reporting rates of infection by treatments and there are cumulative and open-label study periods that have been reported. However none have offered a combined evaluation as the analysis reported here.
When conducting pharmacoepidemiologic studies, it has been recommended that at least two cohorts be evaluated to establish the extent of reproducibility . The resulting variation in IRs among the cohorts provides a useful range of estimates for comparison, which is preferable to a single cohort IR. Also, published literature suggests that RA patients are at higher risk of infection compared with the general population, rendering the RA DMARD observational cohorts an appropriate reference group and not the general population [2, 29].
There have been a number of published papers evaluating the risk of infections in RA patients where the exposure, outcomes, and risk measurement were not consistent across studies. These definitions include, but are not limited to, the regulatory definition of serious, hospitalized infections, and the possible inclusion of administration of IV antibiotics . For transparency, the authors defined the outcomes and present the number of events, incidence rates, and percentages (proportions) for the RA cohorts and abatacept CDP.
This study has several limitations, including those inherent to its design. The data collected and analyzed from the RA cohorts were not primarily collected for this type of study. The limitations associated with the use of external control groups include, but are not limited to, differences in physician management and diagnoses of RA, the inclusion of both prevalent and new users of DMARD agents, differences in the ascertainment and verification of outcomes, length of follow-up, validity of RA diagnosis, disease duration, and severity of disease. We acknowledge that these variables may be different among the RA cohorts. Although the abatacept population appeared to be demographically similar to the cohorts, clinical trial patients are inherently different. The abatacept CDP enrolled mainly prevalent, stable non-biologic DMARD users who had an inadequate response to their current therapy; thereby implying more severe disease in these patients. The RA cohort populations were a diverse group of both prevalent and new non-biologic DMARD users. It is difficult to know how comparable the RA cohorts are to the abatacept trial population.
The RA cohort populations are potentially more stable in that for this analysis the population had to be on a non-biologic DMARD throughout follow-up without addition of a biologic therapy. However, the non biologic DMARD treatment in these groups could be altered during follow-up such that non biologic DMARD therapy could be increased, decreased, or another non-biologic treatment could be added to the current regimen all together. As with any population (trial or observational), it is difficult to know if a patient was previously exposed to a biologic agent for RA or had any other indicated comorbid condition (psoriasis, transplant) prior to entry into the trial or cohort. Trial patients may be monitored more closely for adverse events, which might overestimate our calculated SIRs. However, some of the cohorts have pre-specified observation times for patients enrolled in the registry (for example, NOAR). TB screening prior to entry in the trials may have resulted in a lower incidence of TB in the abatacept CDP. The exclusion of patients with a recent infection, suggesting that enrolled patients are potentially healthier, may have resulted in a lower overall incidence of infections, however the authors did apply this exclusion to the data to be more reflective of the trial population. We were not able to adjust for potential confounders such as severity of disease (rheumatoid factor (RF) or anticitrullinated protein/peptide antibody (ACPA) status), ethnicity, co-morbidities, and use of non-DMARD medications (for example, NSAIDs and corticosteroids) due to unavailability of data on these variables in most external cohorts. Finally, each of the databases used in this study may be associated with specific limitations, such as uncertainty surrounding diagnostic accuracy in administrative claims databases (for example, BC), small cohort size (for example, NOAR), use of self-reporting (for example, NDB), and so on. However, these individual limitations were minimized by the use of multiple cohorts, resulting in a range of references that provided relatively consistent results.
This study has several strengths. The authors predefined exposure and outcome criteria in an attempt to harmonize the methods applied to this analysis including exposure to non-biologic DMARDs, definition of events (hospitalized infections), and the computation of person-time in the evaluation of infections in the abatacept CDP and the RA cohorts.
In randomized trials, serious infections are defined as infection resulting in death, life threatening, requiring inpatient hospitalization or prolongation of a hospitalization, resulting in disability, a congenital anomaly or a medically important event based on medical and scientific judgment. This definition is difficult to apply to observational cohort data. Hence, 'infections resulting in hospitalization' was used since this outcome can be easily identified in all external observational RA data sources.
The effort to harmonize the methods used in these cohorts may seem trivial; however, Solomon and colleagues outlined the importance and complexities of defining exposure risk windows, comparators and endpoints, and when different, the challenge in interpreting data from epidemiological studies published independently . The analyses presented here from six RA cohorts were done in collaboration with registry and observational data holders in an attempt to control these key methodologic issues.
Despite geographic differences and differences in ascertainment methods, the ranges of age- and sex-adjusted IRs were relatively similar among the various RA cohorts. The Early RA cohorts provided IRs on less severe patients with shorter duration of disease where NDB and BC have patients with longer disease duration. Additional strengths include the choice of comparison group. We were able to include in this study, as a reference group, only those patients treated with non-biologic DMARDs from the respective RA cohorts. Given that non-biologic DMARDs comprised the background therapy in all abatacept-treated patients in the clinical trials, this constituted an appropriate reference group for comparing risk of infections.
There have been few opportunistic infections observed throughout the abatacept CDP. TB was the most common, with a total of three events reported in 4,134 patients over a total of 8,392 person-years, none of which resulted in hospitalization.