Safety of Abatacept Compared With Other Biologic and Conventional Synthetic Disease-Modifying Antirheumatic Drugs in Patients With Rheumatoid Arthritis

Data From an Observational Study

Gulsen Ozen; Sofia Pedro; Rebecca Schumacher; Teresa A. Simon; Kaleb Michaud

Disclosures

Arthritis Res Ther. 2019;21(141) 

In This Article

Methods

This analysis utilized data from FORWARD, an ongoing, longitudinal, prospective, observational study in the USA, which has been described in detail previously.[29] Patient-recorded details collected by bi-annual questionnaires were analysed and included all medications taken in the previous 6 months (including doses, months taken, start and stop dates, reasons for discontinuation and side effects). Other information collected included patient demographics, socioeconomic data, co-morbidities, medical events, health-related quality of life, health symptoms and RA-specific outcome measures. The last questionnaire used for this analysis was administrated between January and June 2016 and reflected the events of the preceding 6-month period (August to December 2015).[29,30] The analysis population comprised all those who completed at least one full questionnaire and initiated a new course (incident users) of either abatacept, other bDMARDs (adalimumab, anakinra, certolizumab, etanercept, golimumab, infliximab, rituximab and tocilizumab) or csDMARDs (methotrexate, hydroxychloroquine, leflunomide and sulfasalazine) during the study period. The design of the analysis and inclusion of comparison arms helped mitigate complications with data capture. Each initiator of abatacept was matched with initiators of other RA treatments by date of cohort entry in a maximum of a 1:3 ratio. Patients in the abatacept group were defined as those who initiated abatacept or switched to abatacept after starting another bDMARD or a csDMARD. Patients in the comparator treatment groups were defined as those with no history of abatacept treatment. Patients in the comparator groups who switched to abatacept following treatment with another bDMARD or a csDMARD were included in the abatacept group and contributed to both the comparator (up to point of switch) and abatacept (following switch) data sets. Baseline was defined as the treatment start date; patients who switched treatment during the study could be allocated to more than one group. The database was locked on 4 June 2016.

Patients

Patients enrolled in FORWARD and eligible for inclusion in this analysis were aged ≥ 18 years, had a confirmed diagnosis of RA, and initiated abatacept, another bDMARD or csDMARD between 1 July 2005 and 31 December 2015 in the USA. For each endpoint, except hospitalized infections, patients with the endpoint of interest at baseline were excluded from the analysis. For the analysis of malignancies, patients with a prior history of cancer at baseline were excluded. All patients provided their written informed consent for participation in the analysis.

Analysis Outcomes

The primary outcomes were malignancies (overall malignancies [including NMSC], lung cancer, lymphoma, breast cancer and NMSC), infections (hospitalized infections, pneumonia, opportunistic infections and tuberculosis) and autoimmune diseases (lupus, multiple sclerosis and psoriasis). Malignancies, infections and autoimmune diseases were identified from study questionnaires and were validated by patient and/or physician interview and medical record review, using the International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes and hospitalization codes (see Additional file 1: Tables S1, Additional file 2: Tables S2, Additional file 3: Tables S3). Secondary outcomes included hospitalizations for any reason, any or serious infusion and/or injection reaction and death (identified and validated using the National Center for Health Statistics National Death Index). Reactions to infusions and injections were patient-reported only and were classified as 'severe' if they caused severe redness and/or pain, changes in blood pressure, difficulty breathing, feeling ill, chills, feeling faint or any severe symptoms that required medical care.

Follow-up

Study follow-up started from initiation of abatacept, other bDMARDs or csDMARDs and continued until the first development of any study outcome, treatment discontinuation, death, loss to follow-up or end of study period, whichever came first. Hospitalized infections were attributed to the corresponding treatment group when the treatment was ongoing or discontinued ≤ 3 months previously. This risk window was extended to ≤ 12 months for rituximab due to its long-term effects on B cells. For the assessment of hospitalized infections, this risk window after treatment discontinuation was included in the follow-up period for patients who discontinued therapy for any other reason.

Statistical Analysis

Overall IRs for each outcome with 95% confidence intervals (CIs) were calculated by dividing the number of events for each endpoint by the total patient-time at risk. Only the occurrence of the first event of interest was considered, and the corresponding rates per 100 patient-years were determined. The risks of malignancies, infections and selected autoimmune disease infections were quantified in patients receiving abatacept versus other bDMARDs and abatacept versus csDMARDs using marginal structural models (MSMs). Results were presented as hazard ratios (HRs) with 95% CIs.

MSMs allow proper adjustment of time-varying confounders that are also affected by prior treatment.[31] For this analysis, the inverse probability of treatment weights (IPTWs) were derived from each patient's treatment history and time-varying confounders. These IPTWs were then used in a logistic regression model to control for the time-dependent confounding in the outcome-treatment association, allowing less biased estimates to be obtained.[32] The IPTWs created a pseudo-population in which patients receiving treatment and those not receiving treatment were balanced over the time-varying confounders, but the relationship between treatment and outcome was not altered. The HR was obtained by using a weighted pooled logistic regression (which is equivalent to a Cox model when the hazard of treatment is small) for the probability of receiving abatacept at a given time using the following time-varying covariates measured at baseline and point of measurement: age, sex, employment status (yes/no), annual income (≤ 45 K USD, > 45 K USD), education level (≤ 12 years, 13–15 years, ≥ 16 years), smoking status (never, current and past), disease duration (≤ 3 years, 4–10 years, ≥ 11 years), Health Assessment Questionnaire-Disability Index (HAQ-DI), pain and patient global scores by visual analogue scale (0–10), body mass index, Rheumatic Diseases Comorbidity Index score, any chronic lung disease (yes/no), diabetes (yes/no), number of prior bDMARDs (0, 1, 2, ≥ 3), glucocorticoid (GC) use (yes/no, duration, daily dose [< 7.5 mg/day, 7.5–< 15 mg/day, ≥ 15 mg/day]), year of study entry (2005–2007, 2008–2010, 2011–2013, 2014–2015) and follow-up time using a three-knot spline. Considering the effects of increasing age, co-morbidities, disease duration, and severity measures on the outcomes, time-varying age, disease duration, HAQ-DI, pain and patient global scores, Rheumatic Disease Comorbidity Index (RDCI) and GC treatment duration were also included in the model. The dataset was discretized into one observation per month, and the hazard of treatment in any month was considered small.[33] Due to the low number of events for subtypes of malignancies, infections and autoimmune diseases, MSMs were created only for overall malignancies, NMSC and hospitalized infections.

In order to prevent bias from removing observations due to missing data, all missing covariates of the completed questionnaires were replaced by using multiple imputation by chained equations to create multiple imputed datasets for analyses. Since the odds ratios from MSMs are equivalent to the HRs that would be obtained from the Cox models, the results were presented as HR (95% CI). An intention-to-treat analysis was performed for all outcomes except hospitalized infections where an on-treatment plus 90 days was preferred. The weights were corrected for loss of follow-up and, for infection outcomes, induced selection bias due to artificial censoring (i.e. treatment noncompliance). All p values were two-sided and were conducted at a significance level of 0.05. All statistical analyses were performed using Stata/MP version 14.2 (StataCorp, College Station, TX, USA).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....