Article Text

Extended report
Impaired response to treatment with tumour necrosis factor α inhibitors in smokers with axial spondyloarthritis
  1. Adrian Ciurea1,
  2. Almut Scherer2,
  3. Ulrich Weber3,
  4. Pascale Exer4,
  5. Jürg Bernhard5,
  6. Giorgio Tamborrini6,
  7. Myriam Riek2,
  8. Rüdiger B Müller7,
  9. Bettina Weiss8,
  10. Michael J Nissen9,
  11. Rudolf Kissling8,
  12. Beat A Michel1,
  13. Axel Finckh9
  14. on behalf of the Rheumatologists of Swiss Clinical Quality Management Program for Axial Spondyloarthritis
  1. 1Department of Rheumatology, University Hospital, Zurich, Switzerland
  2. 2Swiss Clinical Quality Management Foundation, Zurich, Switzerland
  3. 3King Christian 10th Hospital for Rheumatic Diseases, Gråsten, Denmark
  4. 4Private Rheumatology Practice, Basel, Switzerland
  5. 5Department of Rheumatology and Rehabilitation, Bürgerspital, Solothurn, Switzerland
  6. 6Department of Rheumatology, Bethesda Hospital, Basel, Switzerland
  7. 7Department of Rheumatology, Cantonal Hospital, St. Gallen, Switzerland
  8. 8Division of Rheumatology, Uniklinik Balgrist, Zurich, Switzerland
  9. 9Department of Rheumatology, University Hospital, Geneva, Switzerland
  1. Correspondence to Dr Adrian Ciurea, Department of Rheumatology, University Hospital Zurich, Gloriastrasse 25, Zurich CH-8091, Switzerland; adrian.ciurea{at}


Objectives To investigate the impact of smoking on the response to treatment with a first tumour necrosis factor inhibitor (TNFi) in patients with axial spondyloarthritis (axSpA) in a real-life cohort.

Methods Patients fulfilling the Assessment of SpondyloArthritis international Society (ASAS) classification criteria for axSpA in the Swiss Clinical Quality Management Cohort were included in this study. The potential association between smoking status and differential response to TNFi in terms of the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI) and the Ankylosing Spondylitis Disease Activity Score (ASDAS) was analysed using multiple adjusted longitudinal mixed effect models. Binary response rates at 1 year were assessed with multiple adjusted logistic analyses.

Results A first TNFi was initiated in 698 patients with axSpA with available smoking status and a baseline or follow-up BASDAI assessment, of which 490 (70%) had complete covariate data. In comparison to non-smokers, current smokers demonstrated significantly smaller reductions in BASDAI and ASDAS scores upon treatment with TNFi (0.75 BASDAI units and 0.69 ASDAS units less, p=0.005 and 0.001, respectively) for patients with elevated baseline C-reactive protein (CRP) level. This effect was numerically smaller in patients with normal CRP. The odds for reaching a 50% improvement in BASDAI response or the ASAS criteria for 40% improvement after 1 year were significantly lower in current smokers than in non-smokers (0.54, 95% CI 0.31 to 0.95, p=0.03 and 0.43, 95% CI 0.24 to 0.76, p=0.004, respectively).

Conclusions Current smoking is associated with an impaired response to TNFi in axSpA.

  • Ankylosing Spondylitis
  • Disease Activity
  • Smoking
  • Spondyloarthritis

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


The influence of smoking on the development of rheumatoid arthritis (RA) is well established,1 and several therapeutic agents, including tumour necrosis factor inhibitors (TNFi), have been shown to be less effective in smokers with RA.2–4 Smokers have also been shown to have a shorter treatment adherence and a poorer response to TNFi in psoriatic arthritis.5 The impact of tobacco use in patients with axial spondyloarthritis (axSpA) is only beginning to emerge. Smoking was associated with incident ankylosing spondylitis (AS) in a recent longitudinal cohort study.6 In several cross-sectional analyses, smokers with AS and early axSpA showed increased levels of disease activity markers in comparison to non-smokers.7–10 Several predictors of good response have been identified after the introduction of TNFi to the therapeutic armamentarium in axSpA, but smoking was not among them: young age, short disease duration, male gender, elevated C-reactive protein (CRP), human leucocyte antigen (HLA)-B27 positivity, high baseline disease activity as assessed by the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI), as well as good functional status and spinal mobility expressed by the Bath Ankylosing Spondylitis Functional Index (BASFI) and Bath Ankylosing Spondylitis Metrology Index (BASMI), respectively.11–15 The aim of our study was to investigate the effect of smoking on the response to TNFi in a large cohort of patients with axSpA.


Study population and design

Patients recruited in the ongoing Swiss Clinical Quality Management Cohort for axSpA (SCQM-axSpA) were required to have a clinical diagnosis of AS or another form of SpA with predominantly axial disease according to the treating rheumatologist.16 Clinical and laboratory data were collected annually with the additional possibility to include data at intermediate visits.17 Smoking status was obtained through a patient questionnaire: non-smoker, previous smoker or current smoker, without reference to the quantity (eg, pack-years). Data on physical activity at each visit were also collected as a standardised patient questionnaire indicating the type of training: (a) none, (b) weekly 1 h group flexibility exercises offered by the Swiss AS Association throughout Switzerland, (c) flexibility exercises at home, (d) training in a gym and (f) other physical activity as training. The answer options for the frequency of training periods per week for exercise types c–e were 1–2x, 3–4x and 5–7x. The number of exercise sessions per week equalled the sum of the indicated exercise types (using the lower value of the indicated frequency interval). Therapeutic interventions after inclusion into SCQM were left to the discretion of the treating rheumatologist. However, patients eligible for treatment with TNFi were preferentially enrolled because regulatory authorities have called for continuous monitoring of patients receiving biologics and rheumatologists can deduct the costs of biologics used to treat patients included in SCQM-axSpA from their global treatment expenditures.

This was a longitudinal study within the SCQM-axSpA cohort analysing data collected between the start of data collection in January 2005 and the end of January 2014. The major inclusion criterion for this study was the fulfilment of the Assessment of SpondyloArthritis international Society (ASAS) 2009 classification criteria for axSpA18 with the following minor modifications as the cohort was initiated before publication of these criteria: the ASAS criterion ‘inflammatory back pain’ was defined as low back pain and morning stiffness for >3 months, improving with exercise but not relieved by rest, as well as age at onset <45 years. Furthermore, the ASAS criterion ‘good response to non-steroidal anti-inflammatory drugs’ was only added to the questionnaire in 2009. Additional inclusion criteria encompassed the availability of smoking status, the initiation of a first TNFi after recruitment into the cohort and at least one follow-up visit under the respective TNFi.

Outcomes and exposures

The study's outcome of interest was the response to a first TNFi. The exposure of interest was smoking status, categorised either as ever/never, or as current/previous/never. Response to TNFi was analysed either as the longitudinal evolution of disease activity as measured by BASDAI19 and AS Disease Activity Score (ASDAS-CRP)20 or as the binary response rate to a first TNFi at 1 year of treatment, ±6 months. This large window was mandated by the structure of SCQM as the yearly control visits recommended after inclusion did not necessarily match yearly intervals after initiation of treatment. The following response variables were assessed: 50% improvement in BASDAI (BASDAI50), major improvement in ASDAS (ASDAS-MI) and the ASAS criteria for 40% improvement (ASAS40).17

Statistical analysis

All analyses were prespecified. Correction for multiple testing was not performed. Baseline characteristics in terms of categorical variables were compared between smokers (ever or current/previous) and non-smokers using the χ2 test. For symmetrically distributed discrete or continuous variables, the t test was used for testing the difference in mean values. In other cases, the Mann–Whitney U test for differences in location of the distribution was used. All tests were two-sided, with a significance level set at 5%.

Several mixed effect models for the relationship of disease activity and time with the covariate smoking and a random intercept have been evaluated: fractional polynomials of time,21 conventional polynomials (including time+time2 and time1/2+time) and a ‘step model’, which only distinguished between the baseline visit and any follow-up visits. The fit of the models was analysed in terms of a Bayesian information criterion for comparing fixed effects in mixed effects models.22 The ‘step model’ fitted the observed data best. It includes a baseline and a follow-up component by using an indicator function, IFU, which has the value 1 if the visit is a follow-up and 0 for the baseline visit. It allows the estimation of both baseline BASDAI (or ASDAS) differences associated with different covariates (for IFU=0) and differences in BASDAI drop after initiation of TNFi associated with covariates, including smoking status (interaction terms of covariates with IFU). The following covariates were considered potential confounders for the effect of smoking based on conceptual considerations of their known association with smoking and potentially with response to treatment: sex, educational level as a proxy of socioeconomic status and physical exercise. Other covariates were included in the model because they were considered important in terms of explaining response to TNF inhibition: age, disease duration, HLA-B27 status, classification as AS or non-radiographic axSpA (nr-axSpA) and body mass index (BMI). Analyses were performed in subgroups of patients split by their baseline CRP status (elevated vs non-elevated). The Harrison method23 was used to adjust for baseline differences in BASDAI or ASDAS-CRP and BASFI in bivariate mixed effect models.24

To assess a potential association between smoking status and differences in response rates to a first TNFi, multiple adjusted logistic response analyses were performed for BASDAI50, ASDAS-MI and ASAS40 responses at 1 year (±6 months) for patients still on treatment and with an assessment of disease activity in the corresponding time window. Models were adjusted for gender, age, baseline CRP status (normal or elevated), HLA-B27, classification as AS versus nr-axSpA, BMI, educational level and physical exercise and for differences in baseline BASFI. The Hosmer–Lemeshow test was used to assess the goodness of fit of the models. The analyses of BASDAI over time and the BASDAI50 response after 1 year were repeated using multiple imputation of missing covariate data.25 Additional information on statistical methods and model choice is provided in the online supplementary appendix. R statistical software (R Development Core Team, 2011) was used for all analyses. The function ‘lmer’ (package lme4) was used for the linear mixed effects models (


Patient disposition and characteristics

Out of 2973 patients with SpA recruited into SCQM, 1880 fulfilled the ASAS criteria for axSpA and had available smoking status. A total of 549 patients were already treated with TNFi at inclusion. TNFi were initiated in 698 patients after recruitment (247 adalimumab, 216 etanercept, 169 infliximab, 64 golimumab and 2 certolizumab-pegol). The majority of these patients were smokers (38.4% current smokers, 23.9% past smokers and 37.7% non-smokers). Baseline differences between never-smoking patients and ever-smoking patients, as well as between current and previous smokers, are shown in table 1.

Table 1

Baseline characteristics in smokers and non-smokers with axSpA starting a first TNFi

Response to TNFi as assessed by the course of BASDAI and ASDAS

The observed course of BASDAI after the initiation of a first TNFi in ever smokers versus non-smokers is depicted in figure 1A, B. The analyses were stratified by CRP status at baseline as an interaction between smoking and CRP was found in a preliminary investigation. The median (IQR) observation period on a first TNFi was 2.1 (0.9 to 3.9) years with a median (IQR) of 2.0 (1.0 to 4.0) follow-up visits. Longitudinal mixed effects models, adjusted for baseline BASDAI levels, were used to analyse the response to TNFi in terms of the drop in BASDAI from baseline to follow-up (ΔBASDAI), as well as the difference in ΔBASDAI between smokers and non-smokers (figure 1C, D). Smoking was associated with a significantly lower reduction in BASDAI in patients with elevated baseline CRP levels (difference in ΔBASDAI 0.5 units, p=0.02, figure 1C). In the subgroup of patients with normal baseline CRP levels, the difference in ΔBASDAI was smaller and statistically not significant (0.3 units, p=0.21, figure 1D).

Figure 1

BASDAI evolution in patients treated with a first TNFi differentiated by baseline CRP status. Full lines: ever smokers; dashed lines: non-smokers. (A, B) Observed BASDAI evolution including all patients with available baseline CRP (665 of 698 patients). The total number of patients on treatment and with an observation around the respective nominal time point is indicated above the x-axis. At baseline, only data from the first visit are used. At later nominal time points, time windows of ±6 months around the nominal time point are used. p Values <0.05 and <0.1 for point-wise Mann–Whitney U test for a difference in distributions between smokers and non-smokers are indicated with ‘*’ and a ‘.’ symbol, respectively. (C, D) Modelled BASDAI evolution based on linear mixed effects models, solely adjusted for smoking status and differences in baseline BASDAI. BASDAI, Bath Ankylosing Spondylitis Disease Activity Index; CRP, C-reactive protein; TNFi, tumour necrosis factor inhibitor.

As the course of BASDAI over time upon TNF inhibition may be influenced by a variety of factors (age, gender, symptom duration, classification as AS or nr-axSpA, HLA-B27 positivity, education, BMI and physical exercise), multiple adjusted models were used in patients with complete covariate data (70%, N=490, baseline characteristics shown in table 2). Models were additionally adjusted for underlying differences in baseline BASDAI and BASFI as these parameters represent important predictors of response to TNFi. The models estimate the difference in ΔBASDAI in BASDAI units after the start of TNFi for each covariate, including smoking (table 3). A positive value for the difference in ΔBASDAI means that the respective covariate impairs the response to TNFi, while a negative value represents a more favourable BASDAI response. Upon TNF inhibition, current smoking led in patients with elevated baseline CRP levels to a significantly lower BASDAI reduction from baseline to follow-up (0.75 BASDAI units less, p=0.005). A similar effect size was observed when comparing patients classified as nr-axSpA versus AS, with an estimated difference in ΔBASDAI of 0.82 BASDAI units (table 3). Past smoking had no significant effect on the BASDAI reduction after the start of TNF inhibition (0.36 BASDAI units, p=0.27). The effect of smoking (ever, current or previous) on the difference in ΔBASDAI is summarised in table 4. An analysis based on multiple imputation of missing covariate data led to similar results for the effect of smoking on ΔBASDAI (data not shown). The multiple adjusted analyses confirmed the crude BASDAI evolution displayed in figure 1A, B: smoking was significantly associated with a reduced response to TNFi in terms of BASDAI improvement, and this effect was numerically stronger in patients with elevated baseline CRP (the difference in the smoking effect in patients with normal vs elevated baseline CRP being statistically not significant; data not shown).

Table 2

Baseline characteristics of axSpA smokers and non-smokers with complete covariate information starting a first TNFi differentiated by baseline CRP status

Table 3

Estimated difference in BASDAI or ASDAS-CRP drop following start of a first TNFi induced by different covariates after stratification for baseline CRP levels

Table 4

Estimated difference in BASDAI or ASDAS-CRP drop after initiation of a first TNF inhibitor between smokers (ever, current, previous) and non-smokers

In addition, the negative influence of smoking on treatment response was confirmed in all analyses using the course of ASDAS-CRP upon TNF inhibition as an outcome (see online supplementary figure S2, tables 3 and 4). The influence of smoking on treatment response in patients with elevated baseline CRP was driven again by current smoking (0.69 units, p=0.001) as no statistically significant influence on ASDAS-CRP reduction could be detected in past smokers (0.3 BASDAI units, p=0.22). The effect size tended to be larger for ASDAS-CRP than for BASDAI, considering the range of these two scores. As CRP is used in the calculation of the ASDAS-CRP score, this difference might be explained by the finding that normalisation of CRP after start of TNFi occurred in a significantly lower proportion of smokers than non-smokers (Fisher's test p=0.04). As for BASDAI, past smoking did not significantly affect improvement in ASDAS-CRP. Comparable effect size estimates and identical conclusions were obtained from analyses using only data up to 2, 3 or 4 years after the start of a first TNFi.

Response rates to TNFi after 1 year

The effect of smoking on BASDAI and ASDAS-CRP reduction after initiation of TNFi was mirrored in dichotomised response rates (BASDAI50, ASDAS-MI and ASAS40 response rates) after 1 year (±6 months). Eighty-six patients (12.3%) were excluded from these analyses as the only available follow-up visit had not been scheduled within this time window and only 33 patients had stopped TNFi treatment within the first six months of therapy. We found significant differences between smokers and non-smokers for all effectiveness outcomes (table 5). OR for achieving a BASDAI50 response in current smokers compared with never smokers was 0.54 (95% CI 0.31 to 0.95), p=0.03. Analyses with multiple imputation for missing covariates yielded similar BASDAI50 results (data not shown). Past smoking did not significantly affect any of the outcomes. Differences in BASDAI50 response rates between current smokers and non-smokers ranged from 8% to 21% (table 5).

Table 5

Multiple adjusted response analysis after 1 year of treatment with a first TNFi


This longitudinal study of a large cohort of patients with axSpA followed in real-life conditions suggests that current smoking impairs the response to TNFi treatment, especially in patients with elevated baseline CRP. Current smokers had approximately 0.7 BASDAI or ASDAS-CRP units less improvement in disease activity in response to a first TNFi. The effect of smoking was lower than the minimally important difference in ASDAS-CRP (decrease of ≥1 units).26 However, a combination of several predictors of response has been shown to more adequately predict relevant treatment responses in axSpA and matrix models have been developed for this purpose.14 We show here that the combination of smoking (effect size 0.69 ASDAS-CRP units) and of classification as nr-axSpA (effect size 0.53 ASDAS-CRP units) in the context of elevated baseline CRP levels outreaches the minimally important ASDAS-CRP difference of 1.1 units. The relevance of the finding that smoking impairs clinical improvement after TNFi initiation was confirmed in dichotomised response rates, which demonstrated significant differences for achieving ASDAS-MI, ASAS40 and BASDAI50 responses in current smokers compared with never smokers. Whether the response to individual TNFi might be differentially affected by smoking2 remains to be investigated in larger numbers of patients treated with the respective agents.

The underlying mechanisms of the influence of smoking on clinical improvement after TNFi start largely remain unclear. In addition to interference with pharmacokinetic and pharmacodynamic characteristics of TNFi, other factors might be involved.27 Smoking may not only raise CRP levels in a dose-dependent manner,28 it may also increase pain levels by influencing neurological processing of sensory information or by unspecific tissue damage due to hypoxia or vasoconstriction.29 Additionally, physical exercise has been often regarded as a confounder, as exercise may improve disease activity and function, and smokers may exercise less.30 Our analyses have been adjusted for differences in physical activity. Other hypotheses put forward to explain the association between smoking and future spinal radiographic progression31 ,32 include periodontal disease, citrullination and epigenetic influences on local remodelling pathways.27 Importantly, smoking might amplify the effect of disease activity on radiographic damage, as has been recently suggested.33

Previous smoking was neither associated with differences in BASDAI or ASDAS-CRP improvement following TNFi initiation, nor with differences in response rates. In addition to reducing cardiovascular morbidity and mortality,34 patients with axial SpA might, therefore, further benefit from quitting tobacco use. As the time-point of smoking cessation was not recorded in SCQM, it remains unknown how long the detrimental effect of smoking on a future response to TNF inhibition persists after quitting tobacco use. Analyses of major smoking-related diseases have shown that smoking cessation at any age dramatically lowers mortality and that nearly all excess risk can be avoided if a person quits smoking before 40 years of age.35 ,36

This study has limitations inherent to observational data, especially informative (non-random) dropout in the longitudinal study. However, similar effect size estimates and identical conclusions were obtained from sensitivity analyses using only data up to 2, 3 or 4 years after the baseline visit. Although we adjusted our analyses for many disease characteristics potentially associated with the course of disease activity and smoking, we cannot exclude the possibility of confounding by unmeasured factors.

In conclusion, our findings suggest that current smokers with axSpA may have a reduced clinical response to TNFi, particularly in the context of elevated baseline CRP levels. Whether quitting smoking might ameliorate the course of disease during treatment with TNFi remains to be confirmed in prospective studies.


We thank the participating rheumatologists and patients who made this study possible and the entire SCQM staff for data management and support. A list of rheumatology private practices and hospitals that are contributing to the SCQM registries is available at We furthermore would like to thank Prof. Leonhard Held and Dr. Eva Furrer from the Department of Biostatics of the University of Zurich for statistical support in setting up the Harrison adjustment for baseline differences and Julia Braun for sharing her R code for this analysis. The additional statistical support of Dr. Daniel J. Stekhoven was very much appreciated.


View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:

  • Press release

    Press release

    Files in this Data Supplement:

  • Lay summary

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Handling editor Tore K Kvien

  • AC and AS contributed equally.

  • Contributors AC, AS and AF designed the study. All investigators substantially contributed to the acquisition, analysis or interpretation of data. AC wrote the article and all coauthors revised the manuscript critically for important intellectual content. AS was responsible for the implementation of the statistical analyses. AC had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors agreed on the final content of the submitted manuscript.

  • Funding This study was supported by grants from the ‘Stiftung für Rheumaforschung’ and the ‘Schweizerische Bechterew-Stiftung’. The SCQM Foundation was financially supported by the Swiss Society of Rheumatology, the Balgrist Foundation, the ARCO Foundation and by Abbvie, Bristol-Myers-Squibb, Merck Sharp & Dohme, Pfizer, Roche and UCB. The study sponsors had no role in the study design or in the collection, analysis or interpretation of the data, the writing of the manuscript or the decision to submit the manuscript for publication. Publication of this article was not contingent upon approval by the study sponsors.

  • Competing interests AC has received consulting and/or speaking fees from Abbvie, Merck Sharp and Dohme, Pfizer, UCB, Celgene, Jannsen-Cilag. UW has received honoraria from Abbvie for being a workshop convenor of the International Course on MRI in Spondyloarthritis and for contributing to a slide kit on spondyloarthritis. PE has received speaking fees from Abbvie.

  • Patient consent Obtained.

  • Ethics approval Ethical approval for the collection of patient data was given by the Supraregional Ethics Commission for Clinical Research of the Swiss Academy of Medical Sciences. Ethical approval for the current study was obtained from the Ethics Commission of the Canton of Zurich.

  • Provenance and peer review Not commissioned; externally peer reviewed.