Infliximab Is Associated With an Increased Risk of Serious Infection in Patients With Psoriasis in the U.K. and Republic of Ireland

Results From the British Association of Dermatologists Biologic Interventions Register (BADBIR)

Z.Z.N. Yiu; ; D.M. Ashcroft; I. Evans; K. McElhone; M. Lunt; C.H. Smith; S. Walton; R. Murphy; N.J. Reynolds; A.D. Ormerod; C.E.M. Griffiths; R.B. Warren; BADBIR Study Group

Disclosures

The British Journal of Dermatology. 2019;180(2):329-337. 

In This Article

Patients and Methods

BADBIR is a national prospective ongoing pharmacovigilance registry of patients with psoriasis that was established in 2007 in the U.K. and Republic of Ireland to compare the safety of biologic therapies vs. nonbiologic systemic therapies. Establishing the risk of serious infections was a prespecified primary aim of the registry. The design of BADBIR[8] and the baseline patient characteristics[9] have been published previously. The National Institute for Health and Care Excellence (NICE) recommends that all patients with psoriasis on biologic therapies should be registered on BADBIR. Patients were selected using a data snapshot from October 2016. BADBIR was approved in March 2007 by the National Health Service Research Ethics Committee North West England (07/MRE08/9). All patients gave written informed consent for their participation in the registry.

Baseline Assessment

Baseline data were collected before or during the initial 6 months of treatment. Drug, clinical and comorbidity history along with anthropometric data were collected by a healthcare professional using a web-administered questionnaire, whereas lifestyle factors were collected using a patient-completed questionnaire.

Follow-up Assessments

Data from patients were collected every 6 months for the first 3 years, then annually thereafter up to 10 years. Follow-up data were collected and entered into a web-based system contemporaneously. Specific information about serious infections were collected, including descriptions of events, hospitalization, start and stop dates. Adverse events were classified using the Medical Dictionary for Regulatory Activities (MedDRA) system.

Data Analysis

The main inclusion criteria for this study were patients with chronic plaque psoriasis starting infliximab (Remicade®, Johnson & Johnson, New Brunswick, NJ, U.S.A.) and biologic-naïve patients with chronic plaque psoriasis on acitretin, psoralen-ultraviolet A, ciclosporin, fumaric acid esters, methotrexate or hydroxycarbamide, who were recruited in the nonbiologic systemic cohort. Owing to difficulties such as a low sample size for infliximab and a lack of comparable patients receiving other biologic therapies, adalimumab, etanercept and ustekinumab were analysed separately.[10] Analysis of patients who were biologic-naïve (i.e. first-line infliximab therapy) was performed separately to the aggregated analysis with all patients in the infliximab cohort (i.e. all-lines infliximab therapy).

Patients were included if follow-up data (at least one follow-up) were available. Overall, 844 patients were not included; three emigrated, 785 withdrew consent and 56 did not complete their questionnaire. Patients on infliximab contributed follow-up time from the first dose until the first occurrence of the following events: serious infection, discontinuation of treatment owing to other reasons, last registered follow-up, switch to other biologic therapy or death. Patients in the nonbiologic cohort contributed follow-up time from first dose of the index drug until the first event of any of the above, but were censored at the end of the last alternative nonbiologic therapy. Patients who switched from the nonbiologic therapy cohort to start infliximab contributed follow-up time to both cohorts.

A serious infection was defined as any infection that was associated with or prolonged hospitalization, use of intravenous antimicrobial therapy or led to death. The events were validated by separate review from two clinicians (Z.Z.N.Y. and R.B.W.) against the above criteria, and discrepancies were resolved through discussion. A clinical specialist in the specific type of infection was consulted in cases where there was uncertainty. The first serious infection was included for analysis in the current study, with a risk window period of 90 days following cessation of treatment applied for the attribution of the event to the drug.[11]

The impact of alternate dosing regimens was not analysed as the proportion of patients using cumulative doses that differ from the licensed dosing regimens is low in the U.K. (< 15%)[12] and the NICE-approved dosing regimen is in accordance with the licence. Within the infliximab cohort, the number of person-years receiving doses outside the licence was too low to make statistical inferences regarding the effect of dosing regimen on the risk of serious infection.

Primary Analyses

To provide a description of the rates of serious infections, crude incidence rates for each drug in the biologic cohort and in the nonbiologic cohort were calculated as the number of events per 1000 patient-years of follow-up. Survival modelling with Cox proportional hazards was used to compare event rates and estimate the effect of each exposure on the risk of serious infections. To investigate whether the risk of serious infections was time varying, we used the crude incidence rates at 0–6 months, 6–12 months and 12–24 months of follow-up, which are the designated follow-up data reporting time points.

The specific a priori potential confounders that were included in the multivariable analysis were based on expert opinion and a literature review.[5] These were age, sex, body mass index (BMI), waist circumference, alcohol use, disease severity (PASI), concomitant inflammatory arthritis (including psoriatic arthritis and ankylosing spondylitis), smoking, diabetes, chronic obstructive pulmonary disease, asthma and concomitant immunosuppressants. The total number of measured comorbidities was included as a separate covariate as a proxy for patient frailty. BMI was presented as a categorical variable to ease data description in Table 1, but was kept as a continuous variable in the statistical models. Adjustment for the baseline potential confounders was performed using a propensity score model. A probability score for having the treatment was derived from a logistic regression model based on the baseline relevant covariates listed above. The use of propensity score adjustment has various advantages over multivariable regression models, in particular the ability to check the balance of measured confounders between the comparator cohorts, and improve estimation when an outcome is rare by allowing for multiple covariates.[13]

Inverse probability treatment weighting, where the treatments were weighted for the distribution of the propensity score in the whole model cohort, was then performed using propensity score probabilities in both models. Balance between groups after weighting was assessed using expected bias from a logistic regression model estimating the effect of the variable on serious infection. Improvement in balance was achieved by an iterative process of fitting interaction terms involving the least balanced variables.

Concomitant therapies considered to be immunosuppressants were methotrexate, ciclosporin, fumaric acid esters and hydroxycarbamide. Concomitant immunosuppressants (defined as the exposure period to more than one immunosuppressant in the nonbiologic cohort) were treated exceptionally as time-varying covariates, allowing for the time on and off these drugs throughout follow-up.

Missing data (Table S1; see Supporting Information) were imputed in a multiple imputation model of 20 datasets in order to reduce bias.[14] We used multiple imputations to account for missing data for the potential confounders, as this preserves the variability and uncertainty of the missing data and avoids loss of power and bias that alternative ad hoc methods, such as a complete case analysis, may introduce. Propensity likelihood scores were calculated in each imputed dataset and combined after regression modelling using Rubin's rules. A key assumption for the Cox regression is the proportionality assumption, where the relative risk between the comparators is constant over time. Formal testing for proportionality using Schoenfeld's residuals in the Cox regression model was performed in five extracted imputed datasets, and, where the proportionality assumption did not hold, a time-stratified analysis using the prespecified time points was performed.

Secondary Analyses

A priori planned sensitivity analysis included methotrexate users as the comparator cohort as this was the most common systemic nonbiologic in use. Descriptive analysis was performed for soft tissue and skin infections and lower respiratory tract infections as these were identified as common infections associated with patients on biologic therapies, but the lower number of events did not allow for meaningful multivariable analysis of relative risks (Table S2; see Supporting Information).

All analyses were performed using Stata 14 (StataCorp LLC, College Station, TX, U.S.A.). The methods used in this analysis have been previously described.[10] The details of the methodology pertaining to this current study are reproduced in full for the benefit of the reader.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....