Background/Purpose: Sick leave (SL) is a critical socioeconomic consequence of ankylosing spondylitis (AS). This study aimed at exploring the role of presenteeism relative to BASDAI and BASFI in predicting SL in patients adequately treated with the TNF-α inhibitor infliximab using longitudinal data analysis.

Methods: The EASIC database  is an open label extension of the ASSERT trial and includes 103 patients with AS from six European countries that were treated with infliximab. Demographic, clinical and work-related data were collected at baseline and every six months over a 2-year period. Sick leave represented the number of non-working days in the last six months. Presenteeism represented the AS impact on productivity of patients at work and is measured as a single item  (0: no effect; 10: strong effect). We fitted a number of statistical models, including standard/zero-inflated Poisson/Negative Binomial (NB) fixed/mixed effects models. Based on exploratory data analysis and expert opinion, covariates considered in the models included time, gender, baseline age, disease duration, BASDAI, BASFI, BASMI, presenteeism and work-status (part-/full- time), and pair-wise interactions between them. Model selection was based on Akaike information criterion (AIC), Bayes information criterion (BIC) and likelihood ratio test. SAS 9.2 was used for data analysis.  

Results: Of all patients (n=103), 84% had full-time job; 86% were male; mean (SD) age was 41 (9.6); mean monthly presenteeism ranged from 2 to 3; mean (SD) BASDAI and BASFI were 3.2 (1.7) and 3.5 (1.4), respectively. Of patients with paid jobs, 35% had SL > 0 with a mean (SD) of 8 (10) days over two years.  The ratio of the deviance of Poisson regression to its degrees of freedom indicated overdisperson, suggesting that the standard Poisson models were not appropriate. The likelihood ratio tests comparing the NB models with the Poisson models were significant (p-value < 0.001) in favor of the NB over the Poisson models. Because the overdispersion was due to excess of zero, the zero-inflated NB (ZIBN) models were considered better than the standard NB models. AIC resulting from the ZINB fixed model was smallest, suggesting that the ZINB fixed effect model was the best choice. 

In the first part of the best fitting ZINB fixed effect model (which predicts the probability of having zero day of SL), the intercept and the only slope for presenteeism at 6 months (PRES.LAG6) in the past were significantly different than zero. In the second part (which predicts the number of non-zero SL), the intercept and only slopes for time and PRES.LAG6 were significantly different than zero. An increase of one unit in presenteeism yielded an increase by 29% in the expected SL.

Conclusion: In patients with adequately controlled AS, our findings show that presenteeism is a better predictor for SL over time than BASDAI and/or BASFI. Limitations of this study include small sample size, short follow-up time and the absence of work-related factors among the predictors.

---

« Back to 2013 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/presenteeism-predicts-sick-leave-better-than-basdai-andor-basfi-in-a-longitudinal-cohort-of-patients-with-ankylosing-spondylitis-easic/