The Risk of Uveitis in Patients With JIA Receiving Etanercept

The Challenges of Analysing Real-World Data

Rebecca Davies; Diederik De Cock; Lianne Kearsley-Fleet; Taunton Southwood; Eileen Baildam; Michael W. Beresford; Helen E. Foster; Wendy Thomson; Athimalaipet V. Ramanan; Kimme L. Hyrich


Rheumatology. 2020;59(6):1391-1397. 

In This Article



Patients recruited to one of two UK JIA national prospective treatment registries were included [the British Society for Paediatric and Adolescent Rheumatology Etanercept Cohort Study (BSPAR-ETN) established in 2004 and the Biologics for Children with Rheumatic Diseases (BCRD) study established in 2010]. To be enrolled in the studies, patients were required to have a diagnosis of JIA, classified according to the ILAR criteria.[13] A detailed explanation of the study methods of both studies has been described previously.[14] Patients starting etanercept (BSPAR-ETN) or a non-etanercept biologic (BCRD) for JIA are approached to join the respective studies alongside children starting MTX, who form a comparison cohort within the studies. Recruitment is recommended but not mandatory. Data are captured in identical manners regardless of which drug is started and which study the child is enrolled in. Both studies received ethical approval from a National Health Service Ethics Committee and written informed consent from parents (and where appropriate patients) was provided in accordance with the Declaration of Helsinki.

Data Collection and Follow-up

Baseline data (defined as at the time of starting a biologic or MTX) were collected by the paediatric rheumatologist or clinical research nurse using a web-based questionnaire. Data collected include demographics (age, gender), disease status [disease duration, active joint count, limited joint count, ESR, CRP, physician global assessment, patient/parent global assessment, pain visual analogue scale, Childhood Health Assessment Questionnaire (CHAQ),[15] juvenile arthritis disease activity score-71],[16] ILAR disease classification, drug history and comorbidities. It is also recorded whether or not the patient has a history of uveitis at study registration and whether it was active at the time of registered drug start. ANA status was not collected over the period of patient recruitment included in this analysis. Follow-up data were extracted from the medical record at 6 months, 12 months and annually thereafter, and included current treatments and changes to anti-rheumatic therapy, as well as occurrence of serious and non-serious adverse events, including uveitis.

All adverse events are reported verbatim by the hospitals and coded centrally using the Medical Dictionary for Regulatory Activities.[17]

Newly diagnosed uveitis cases were defined as any reported adverse event of uveitis in patients that had no previous history of uveitis recorded at baseline. Events coded to the Medical Dictionary for Regulatory Activities preferred terms 'uveitis' or 'iridocyclitis' (known as anterior uveitis) were included in the analysis. Additional information using a standardized proforma, including location and type of uveitis, was requested in all cases to verify the event as a new case of uveitis (vs a flare). Only first diagnoses of uveitis were included in our analysis.

Statistical Analysis

The analysis included all children with non-systemic JIA registered at the point of starting MTX, etanercept, adalimumab or infliximab who did not have a history of uveitis at the start of the registered drug. For the purpose of analysis, children starting adalimumab and infliximab were combined as numbers in each individual drug cohort were small. For all patients, person-years of exposure began from date of first treatment with the respective drug and continued until first diagnosis of uveitis, most recent study follow-up recorded up to 30 June 2018, discontinuation of registered drug or death, whichever came first. Events were only included if patients were receiving their treatment of interest (biologic or MTX) at the time of or within the 90 days preceding their first diagnosis of uveitis, to allow for any lag effect.

Patients who registered on MTX and later switched to etanercept or another biologic were followed in the MTX cohort until the point of biologic start. At this point they were censored from the MTX cohort and subsequently followed in the etanercept or adalimumab/infliximab cohort as described above. Similarly, patients who switched between biologics were followed in one cohort until the point of switch, from which point they were censored from the first cohort and followed up in the second cohort.

Baseline comparisons between cohorts are shown, using non-parametric descriptive statistics. Crude rates of uveitis are presented per 100 person-years with 95% CIs. Cox proportional hazard models were used to compare rates of newly diagnosed uveitis between the MTX and etanercept cohorts across all exposure time. Due to a lack of events, only unadjusted hazard ratios (HRs) are presented comparing MTX and adalimumab/infliximab or between etanercept and adalimumab/infliximab. Sensitivity analyses included (i) patients diagnosed with JIA under the age of 12 years (all follow-up included) and (ii) patients diagnosed with JIA under the age of 12 years with follow-up censored at 12th birthday. Current UK JIA uveitis screening guidelines suggest more frequent screening in patients younger than 12 years,[3] therefore the latter two analyses were conducted to rule out any bias caused by this screening effect.

In order to reduce any effects of selection bias, a series of propensity scores stratified into deciles were used to adjust for potential confounding effects of baseline differences between the cohorts (etanercept vs MTX, etanercept-combination vs MTX, etanercept-monotherapy vs MTX and etanercept-combination vs etanercept-monotherapy) and included age, sex, disease severity (using baseline CHAQ and juvenile arthritis disease activity score-71), disease duration, baseline oral steroid use, ethnicity (white vs non-white) and ILAR category (Supplementary Table S1, available at Rheumatology online). The reported bias between the cohorts was low at between 1.5 and 5%. Two time-varying covariates were also included to estimate the probability of an etanercept-combination patient becoming an etanercept-monotherapy patient, and an etanercept-monotherapy patient becoming an etanercept-combination patient. These were included as covariates in the etanercept-combination vs etanercept-monotherapy model. Finally, a series of univariable Cox regressions were performed on baseline variables to identify possible risk factors in the development of new onset uveitis within the whole cohort.

All analyses were performed using Stata, version 14 (StataCorp, 2015, Stata Statistical Software: Release 14, College Station, TX, USA: StataCorp LP). Missing data were accounted for using multiple imputation (20 imputations), using the ice package in Stata.[18] As well as including baseline co-variates in the imputation model, uveitis incidence (quantified as whether a patient ever developed new onset uveitis) and log time to first uveitis were also included.