Background/Purpose: Only scarce date is available on the long term treatment effect in a real-life setting (i.e. effectiveness) of TNFi in early axial SpA forms and its predisposing associated factors; furthermore, unbiased evaluation of treatment effect in non-randomized clinical trials is challenging, and new methods have been developed to overcome prescription bias. 
The objectives of the study were a) to estimate the probability to initiate a TNFi over 5 years of follow-up in real life setting using novel statistical methods to overcome prescription bias; b) to determine the long term effectiveness of first TNFi and its predisposing factors.

Methods: Observational prospective French cohort (DESIR) with 5 years of follow-up, including 708 TNFi-naïve patients with early axial spondyloarthritis. Study visits were scheduled every 6 months in the first two years of follow-up then yearly up to 5 years. Treatment (TNFi or other) was at the discretion of the treating rheumatologist’s. The probability to initiate a TNFi was estimated by the Kaplan Meier Method, assuming non informative dropouts. Effectiveness of the first TNFi was defined as the probability to reach an ASAS40 response in both groups (TNFi vs. any other treatment) after at least 10 months of exposure. To evaluate treatment effect and overcome prescription bias repeatedly occurring over time, we have applied an iterative method based on inverse propensity score (PS) weighting using a marginal structural model, that allows the integration of the repeated weights derived from the propensity score at each visit (i.e. the probability to receive the treatment at each visit). The structural model used for this analysis was a PS-weighted cox regression, to estimate the probability to present an ASAS40 response after at least 10 months of treatment. Factors predicting first TNFi effectiveness, were explored by Cox (univariate and then multivariate) regression models.

Results: Of the 708 patients included in the analysis, 258 patients initiated a first TNFi during the first five years of follow-up. The probability to initiate a TNFi treatment was 41.3% [95%CI 37.2-45.1]. Among the 258 patients who received a first TNFi, 163 (63.2%) were exposed for at least 10months. On the original data, ASAS40 response was observed in 50/163 (30.7%) vs.58/450 (12.9%) patients from the TNFi and usual care groups, respectively. The likelihood of an ASAS40 response was greater in the TNFi exposed group (HR= 3.3[95%CI 2.9-3.8], p< 0.001). Male gender (HR=1.5[95%CI 1.1-2.1]), HLAB27+ (HR= 1.4[95%CI 1.1-2.0]) and the presence of at least one objective sign of inflammation (MRI or CRP) or structural damage (radiographic sacroiliitis) (HR = 1.7[95%CI 1.2-2.4]) were jointly predictive of an improved outcome.

Conclusion: Our study, applying novel statistical techniques to overcome prescription bias, confirms the 5-year effectiveness of TNF alpha inhibitors (TNFi) in patients with early axial spondyloarthritis (axSpA), and we confirm that male gender, with HLAB27 positive and the presence of at least one objective sign of inflammation or structural damage are more frequently associated with such effectiveness.

ACR Meeting Abstracts - https://acrabstracts.org/abstract/5-years-treatment-effect-of-tnf-alpha-inhibitor-in-early-axial-spondyloarthritis-and-associated-factors-an-inverse-probability-weighting-analysis-of-the-desir-cohort/