Incidence of Serious Infections and Design of Utilization and Safety Studies for Biologic and Biosimilar Surveillance

BACKGROUND: There is a need for postmarketing evidence generation for novel biologics and biosimilars. OBJECTIVE: To assess the feasibility, strengths, and limitations of the Biologics and Biosimilars Collective Intelligence Consortium (BBCIC) Distributed Research Network (DRN) to examine the utilization and comparative safety of immune-modulating agents among patients with autoimmune diseases. METHODS: We conducted a retrospective cohort study among patients enrolled in health insurance plans participating in the BBCIC DRN between January 1, 2006, and September 30, 2015. Eligible patients were adult (≥18 years) new users of a disease-modifying nonbiologic and/or biologic agent with a prior diagnosis of rheumatoid arthritis (RA), other inflammatory conditions (psoriasis, psoriatic arthritis, ankylosing spondylitis), or inflammatory bowel disease (IBD). Follow-up started at treatment initiation and ended at the earliest of outcome occurrence (serious infection); treatment discontinuation; or switching, death, disenrollment, or end of study period. The study leveraged the FDA Sentinel System infrastructure for data management and analysis; descriptive statistics of patient characteristics and unadjusted incidence rates of study outcomes during follow-up were calculated. RESULTS: Eligible patient drug episodes included 111,611 with RA (75% female), 61,050 with other inflammatory conditions (51% female), and 30,628 with IBD (52% female). Across all 3 cohorts, approximately half of the patient drug episodes initiated a biologic (50% in RA; 60% in psoriasis, psoriatic arthritis, ankylosing spondylitis; and 55% in IBD). The crude incidence rate of serious infection was 9.8 (9.5-10.0) cases per 100 person-years in RA, 7.1 (6.8-7.5) in other inflammatory conditions, and 14.2 (13.6-14.8) in IBD patients. CONCLUSIONS: This study successfully identified large numbers of new users of biologics and produced results that were consistent with those from earlier published studies. The BBCIC DRN is a potential resource for surveillance of biologics.

the index date was defined as the baseline period. In addition, eligible patient episodes were required to satisfy the following inclusion criteria: (a) a diagnosis of rheumatoid arthritis (RA), psoriasis, psoriatic arthritis, ankylosing spondylitis, ulcerative colitis, or Crohn disease identified from physician encounters any time prior to treatment initiation; (b) at least 18 years of age at the time of treatment initiation; and (c) at least 12 months of continuous medical and pharmacy coverage prior to treatment initiation. Follow-up started on the index date and ended at the earliest of the following: occurrence of outcome (e.g., hospitalized infection, described in detail below), treatment discontinuation or switching (described in detail below), disenrollment (which may indicate health plan change or death), or end of the study period.
Three cohorts of patient episodes were assembled based on disease diagnosis, which included: (a) RA; (b) psoriasis, psoriatic arthritis, and ankylosing spondylitis; and (c) ulcerative colitis and Crohn disease, collectively referred to as inflammatory bowel disease (IBD). All analyses were conducted independently within each cohort. A patient diagnosed with multiple autoimmune conditions may be included in more than one cohort. Additionally, prior use of methotrexate in the baseline period was required for patients in the RA cohort to create a more homogeneous cohort of patients who failed their initial methotrexate therapy. Patients were excluded from the analyses if they (a) ever had nonmelanoma skin cancer based on all available data, (b) had an immune-compromising condition (organ transplantation, HIV, advanced kidney/liver disease) based on all available data, or (c) had a serious infection any time during the 183-day period before the index date. Patients were censored if they developed an immune-compromising condition during follow-up.

Medication Exposure and Covariates
The medications of interest included anti-TNF biologics (adalimumab, certolizumab, etanercept, golimumab, infliximab), non-TNF biologics (abatacept, natalizumab, rituximab, tocilizumab, ustekinumab, vedolizumab), the oral Janus kinase inhibitor (tofacitinib), and nonbiologic disease-modifying agents (azathioprine, hydroxychloroquine, leflunomide, methotrexate, sulfasalazine, and 6-mercaptopurine). Biologics (e.g., secukinumab) approved close to the end of the study period (September 30, 2015) were not included, as it was unlikely that there would be sufficient numbers of new users to conduct the intended analysis. Appendix A presents medications examined in each of the 3 cohorts, as not all the medications are approved for all the conditions. Exposure to the medications was ascertained based on Healthcare Common Procedure Coding System (HCPCS) codes for infused medications and National Drug Code (NDC) numbers for pharmacy-dispensed medications. Because HCPCS codes do not have a built-in days supply, the dosing interval postmarketing safety monitoring and evaluation of approved biologics. 7 Further, the evidence generated through postmarketing comparative utilization and safety studies of biosimilars and reference biologics may help ensure the safety and effectiveness of a biosimilar and instill health care provider and patient confidence in these products. 8,9 The Biologics and Biosimilars Collective Intelligence Consortium (BBCIC) is a nonprofit public service initiative dedicated to providing scientific evidence on the comparative safety and effectiveness of biosimilars and reference biologics, with participants from managed care organizations, integrated delivery systems, pharmacy benefit management firms, research institutions, and pharmaceutical companies. 10,11 The BBCIC's Distributed Research Network (DRN) consists of geographically diverse U.S. commercial, including Medicare Advantage, health plans that also participate in the FDA Sentinel System, thus enabling BBCIC to conduct claims-based research specific to biologics using the Sentinel common data model and analytic tools. 12,13 To evaluate the potential capability of the BBCIC DRN for future study of biosimilar use and safety, we conducted a study among patients with select autoimmune conditions that are commonly managed with anti-inflammatory biologics, such as anti-tumor necrosis factor (TNF) alpha antagonists. [14][15][16][17][18][19] These biologics reduce disease activity and systemic inflammation, improve quality of life, and, when initiated early in the disease course, can effectively improve the prognosis of these patients. 14-20 However, these agents have also been associated with increased risk of developing serious infections in some studies, but not others. [21][22][23][24] Because serious infection is a relatively uncommon adverse event (AE), an abundance of published safety data come from analyses of administrative claims from large populations. [23][24][25][26][27][28] The present study focused on determining whether the BBCIC DRN (a) provides sufficient sample size and exposed follow-up time and (b) contains key data elements needed to ascertain exposure to biologics, define disease-based cohorts, and identify study outcomes.

■■ Methods Study Design and Population
We conducted a retrospective cohort study using curated data from the BBCIC DRN held in the Sentinel common data model format. The study population included individuals enrolled in one of the participating plans during the study period between January 1, 2006, and September 30, 2015. We identified all new users of a conventional or biologic disease-modifying antiinflammatory agent (Appendix A, available in online article) during the study period. New user patient episodes were defined as patients who initiated such an agent without any previous use in the prior 365 days. The date of treatment initiation was defined as the index date, and the 365 days preceding provided on the FDA-approved label was applied to impute the days supply and determine exposure status during follow-up. For example, patient episodes were considered to be exposed for 56 days after receiving an infliximab infusion. Because a biologic may be used as monotherapy or in combination with 1 or more nonbiologic disease-modifying medications, patient episodes were classified into 1 of 3 mutually exclusive categories on the index date: (a) initiated any anti-TNF biologic (regardless of nonbiologic use); (b) initiated any non-TNF antagonist biologics (regardless of nonbiologic use, RA only [Appendix A]); and (c) initiated any nonbiologic disease-modifying agents (no biologic use). Only the first patient episode was used in the outcome ascertainment analyses. During follow-up, treatment episodes were truncated when a patient episode switched within (from one anti-TNF biologic to another) or between the categories (step up from nonbiologic to biologic).
Age, sex, and calendar year of treatment initiation were ascertained on the index date. Other covariates ascertained during the 183-day period before the index date included oral glucocorticoid use, comorbidities (Charlson Comorbidity Index and individual components of the index), and health services use (e.g., numbers of inpatient visits, distinct classes of medications prescribed).

Outcomes
The study included 2 safety outcomes: serious infection and pneumonia. Serious infection was defined as incident cases of infection that required an emergency department (ED) visit or hospitalization. Cases were identified based on the presence of a discharge diagnosis code for infection of the respiratory tract, skin and soft tissue, genitourinary tract, gastrointestinal tract, or central nervous system or septicemia/sepsis. Pneumonia was defined as incident cases of pneumonia that required an ED visit or hospitalization and were identified based on a discharge diagnosis code for pneumonia from an ED visit or hospitalization.

Statistical Analysis
This study leveraged the FDA Sentinel System infrastructure, including the common data model, Sentinel-curated data (common data model v6.0.2), and modular programs for data a Exclusion episodes based on claim in query period, episode incident, and pre-index enrollment criterion. b Exclusions based on nonmelanoma cancer, immunocompromising conditions, and serious infection code lists. AS = ankylosing spondylitis; CONSORT = Consolidated Standards of Reporting Trials; IBD = inflammatory bowel disease; PsA = psoriatic arthritis; PsO = psoriasis; RA = rheumatoid arthritis.  Study Flow for Each Drug Cohort (continued) management and analysis. 13 Specifically, cohort identification and descriptive analysis (Cohort ID and Descriptive Analysis v5.0.3) Level-1 queries (created using SAS version 9.4 [SAS Institute, Cary, NC]) were executed to identify the cohorts, ascertain exposures and outcomes, and produce unadjusted incidence rates. 29 Descriptive statistics (mean [standard deviation], number [percentage]) of patient characteristics ascertained during baseline or on the index date were calculated (before implementation of the exclusion due to nonmelanoma cancer). Crude incidence rates and 95% confidence intervals of serious infection and pneumonia during follow-up were calculated, stratified by categories of medications initiated (defined above: nonbiologic disease-modifying agents, anti-TNF biologics, and non-TNF biologics). Crude incidence rates are reported as observed, without adjustment for potential confounding variables such as oral glucocorticoid use. Data analyses were performed separately for each of the 3 cohorts. Codes used to identify exposure and outcome and to

Baseline Patient Characteristics, by Disease Cohort and by Therapy Class a
continued on next page RA; 7.1 (CI = 6.8-7.5) in psoriasis, psoriatic arthritis, and ankylosing spondylitis; and 14.2 (CI = 13.6-14.8) in IBD patients. When stratified by age, numerically higher incidence rates were observed in older age groups, and a gradient was observed across 3 cohorts and across all treatment categories (Table 2). Different from serious infection, the highest incidence rate of pneumonia was observed in RA patients (1.6 cases per 100 PYs; CI = 1.5-1.7); followed by patients with IBD (1.3 cases per 100 PYs; CI = 1.1-1.5); and lowest in patients with psoriasis, psoriatic arthritis, or ankylosing spondylitis (0.8 cases per 100 PYs; CI = 0.7-0.9; Table 3). Again, consistently across diagnosis and treatment, numerically higher incidence rates were observed among older patients compared with younger ones.

■■ Discussion
This study describes the utilization of biologic and nonbiologic anti-inflammatory agents and unadjusted rates of serious infection and pneumonia in 3 cohorts of patients with autoimmune diseases. This study also identifies strengths and limitations of the BBCIC DRN, along with methodologic challenges and recommendations for improving design and conduct of future comparative safety/effectiveness studies of biologics.
First, this descriptive analysis demonstrates that the BBCIC DRN provides a sufficiently large sample size and follow-up time to examine relatively uncommon AEs in patient populations and subgroups of interest. For instance, pneumonia is an implement inclusion and exclusion criteria are available in Appendix B (available in online article). Figure 1 shows a study flow for each drug cohort of interest within each clinical condition of interest. We identified 3 large cohorts of patient drug episodes, including 111,611 patient episodes with RA (75% female; Figure 1A); 61,050 with psoriasis, psoriatic arthritis, ankylosing spondylitis (51% female; Figure 1B); and 30,628 with IBD (52% female; Figure 1C). Across all 3 cohorts, patients between 18 and 64 years of age accounted for more than 85% of the total population. Approximately half of the patients initiated a biologic (50% in RA; 60% in psoriasis, psoriatic arthritis, ankylosing spondylitis; and 55% in IBD patients). Among RA patients, we identified a total of 1,512 tofacitinib new patient episodes. Because we expected very low numbers of patients with either study outcome, tofacitinib new users were not included in the analyses to assess crude incidence rates. Details of patient baseline characteristics are presented in Table 1.

■■ Results
Patients in the 3 cohorts (RA; psoriasis, psoriatic arthritis, ankylosing spondylitis; IBD) contributed a total of 47,580, 22,076, and 15,251 person-years (PYs) of exposed follow-up time, respectively, to the analysis of serious infection. The overall estimated incidence rate of serious infection was 9.8 (95% confidence interval [CI] = 9.5-10.0) cases per 100 PYs in   (continued) outcomes are defined: the present study identified serious infections from both ED visits and hospitalizations and included both principal and secondary diagnoses, whereas the Grijalva study only included the principal diagnosis from hospitalizations. 23 Likewise, the unadjusted crude incidence rate of serious infection among RA patients aged 65 years or older in the present study was 18.7 per 100 PYs (1,354 cases; 7,223 PYs), which is qualitatively higher than the risk estimates reported among elderly RA patients (e.g., 13.1-18.7 per 100 PYs) with Medicare coverage. 21 Another key strength of the BBCIC DRN is the ability to leverage the Sentinel infrastructure for data management and analysis, including the common data model and analytic tools, which were developed to support active postmarketing drug safety surveillance. 13 This approach offered significant computational efficiency to conduct multidatabase studies, as it allowed rapid local execution of centrally developed modular programs concurrently by multiple data partners using the curated data in the common data model format. 30 This study infrequent AE, which occurred at crude incidence rates of 0.8-1.6 cases per 100 PYs across the 3 cohorts in this study. Still, the large size of the BBCIC DRN enabled the identification of 50,106, 22,819, and 16,377 PYs of exposed follow-up time and 792, 187, and 214 incident cases of pneumonia among patients with RA; psoriasis, psoriatic arthritis, or ankylosing spondylitis; and IBD, respectively. The BBCIC DRN also contains adequately detailed data elements to characterize the study populations that included comorbidities and use of prescription medications and other health services. These measures have the potential to serve as proxies of RA severity, health status, and health-seeking behavior that may be included as potential confounders to be adjusted for in future studies.
The crude incidence rates of serious infection among patients in the RA and IBD cohorts initiating an anti-TNF agent in the present study were numerically higher (8.9 and 14.9 per 100 PYs, respectively) than the crude rates reported in a similar study by Grijalva   Incidence Rates of Serious Infection, Overall and Stratified also leveraged the Sentinel Operation Center query-specifications document to communicate study design and operational details in a standardized format that enables reliable replication. Thus, it is critical that the specifications are developed thoughtfully and appropriately tailored for the specific needs of each project.
Because of the complexity and importance of these tools, we recommend that research teams be trained on the Sentinel infrastructure to obtain an in-depth understanding of the common data model structure, the capability of the analytic tools, and the specifications document. As previously reported, we recommend a phased approach in which results from a test database or one data partner are reviewed by the investigators before executing the analytic programs across all data partners. 31 This approach allows the investigators to revise the study design, if necessary, and thus avoid executing analytic programs multiple times across all data partners, which requires significant effort and resources. For example, our research team worked closely to determine the allowable gap to bridge self-injectable biologics given that HCPCS codes in a medical procedure table do not contain an adjudicated days supply field found in pharmacy-dispensing tables. Finally, an analyst with detailed knowledge of the Sentinel data model and analytic tools is an invaluable resource who can also offer insights from experience that further ensure the successful implementation of a study.

Limitations
We identified study limitations in our ability to conduct biologic and biosimilar surveillance. In the present study, new users are defined as those new to the entire class of study medications (e.g., anti-TNF biologic) and not just the specific cohort-defining medication (e.g., etanercept). As this was a feasibility assessment of the BBCIC DRN to address biologic and biosimilar surveillance, future work will need to be conducted at the medication and manufacturer level. The research team made the decision to avoid selection bias related to medication effects common to the class. For instance, a patient may be  Given that biosimilar initiators are likely to include patients who have switched from the reference biologic, how to handle switching and compare switchers to those who opt to remain on the reference biologic is of critical importance to the biosimilar research study design. 32 For example, patients who switched to a biosimilar from a reference biologic to comply with a formulary decision may be different from patients who switched from a structurally unrelated biologic with regard to their baseline risk of developing an AE. These 2 scenarios of switching to a biosimilar may yield very different safety results, and it would be useful if these different scenarios are captured.
Another methodologic challenge that affects all biologics (including biosimilars) is that a specific HCPCS code that uniquely identifies the medication is usually not assigned until the beginning of the following calendar year after FDA approval. Our study did not utilize these nonspecific codes. Additionally, unique to biosimilars, since 2016, the Centers for Medicare and Medicaid Services (CMS) assigned the same HCPCS code to biosimilars of the same reference biologic and used modifiers to identify the manufacturers of the biosimilars. While the policy was reversed in November 2017 and CMS now assigns an individual HCPCS code to each biosimilar, 33 there is a need to empirically assess the utilization of the modifiers for biosimilars that are affected.
To address the issues identified early in this study, BBCIC commissioned 2 additional work groups to address the challenges of drug switching and exposure identification. First, specific methodological recommendations are needed to handle the complexity of switching in the context of comparative effectiveness or safety studies that may involve several classes of biologics with potentially multiple biosimilars for each reference biologic. Second, an empirical examination of the BBCIC DRN databases has been conducted to assess the utilization of HCPCS modifiers and NDC numbers in medical claims. 34 The goal was to determine the potential of using these data fields to improve the identification of biologic and biosimilar exposure that is billed using nonspecific HCPCS codes, biosimilars that do not have a modifier to specify the manufacturer under the older CMS policy, and administrations that occurred at inpatient and ED settings that would have otherwise not been captured. While the incorporation of HCPCS modifiers and medical claim NDC numbers may complicate exposure ascertainment, it has the potential to meaningfully improve the capture of biologics and biosimilars. While the FDA has issued guidance on nonproprietary naming of biologicals products, it will take time for widespread adoption into our electronic medical records and routine utilization in administrative claims processing. 35 The 2 BBCIC work groups are currently ongoing, and the findings will inform future biosimilar surveillance activities.
Finally, as the existing BBCIC DRN is based on administrative claims data, there are challenges with identifying outcome measures related to clinical effectiveness. This study helped confirm whether effectiveness could be quantified either directly or indirectly (e.g., surrogate measures) and provided a means of identifying opportunities where data enrichment of the existing DRN will be necessary to better answer some effectiveness outcomes of interest. Work is ongoing to develop more robust capabilities to support BBCIC research, and this study provided a necessary baseline from which to build.

■■ Conclusions
This descriptive study demonstrates that the BBCIC DRN is a useful source for biologic surveillance. Efforts are currently in progress to address the methodological challenges identified in the study, such as the ascertainment of biosimilars in the DRN. The lessons learned are translatable to other future studies, including the importance of a thorough appraisal of methods used to assess and classify exposure, an in-depth understanding of the Sentinel distributed system and tools, and a phased analytic approach.