Abstract
Introduction
We studied the effect of Tumor Necrosis Factor-Alpha (TNF)-inhibitors on progressive spine damage in Ankylosing Spondylitis (AS) patients.
Methods
All AS patients (satisfying the modified New York criteria) prospectively followed and with at least two sets of spinal radiographs at a minimum gap of 1.5 years were included (n=334). Patients received clinical standard of care, which included non-steroidal anti-inflammatory drugs and TNF-inhibitors. Radiographic severity was assessed by the modified Stokes Ankylosing Spondylitis Spine Score (mSASSS). Patients with a rate of progression more than 1 mSASSS unit/year were considered progressors. Univariable and multivariable regression analyses were done. Propensity score matching (PSM) and sensitivity analysis were performed. A zero-inflated negative binomial (ZINB) model was used to analyze the effect of TNF-inhibitor on change in mSASSS with varying follow-up periods. Potential confounders like Bath AS Disease Activity Index (BASDAI), ESR, CRP, HLA-B27, gender, age of onset, smoking and baseline damage were included in the model.
Results
TNF-inhibitor treatment was associated with a 50% reduction in the odds of progression (OR: 0.52; CI: 0.30-0.88; p=0.02). Patients with a delay in starting therapy of more than 10 years were more likely to progress compared to those who started earlier (OR=2.4; 95% CI: 1.09-5.3; p=0.03). In the ZINB model TNF-inhibitor use significantly reduced progression when the gap between x-rays was more than 3.9 years. The protective effect of TNF-inhibitors was stronger after propensity score matching.
Conclusions
TNF-inhibitors appear to reduce radiographic progression in AS, especially with early initiation and longer duration of follow up.
Introduction
Ankylosing spondylitis (AS) is a chronic inflammatory arthritis affecting the sacroiliac joints and spine associated with new bone formation and spinal fusion. Patients with AS suffer from significant pain and loss of function with associated work disability 1. The introduction of Tumor Necrosis Factor Alpha (TNF)-inhibitors has significantly altered the landscape of treatment in inflammatory arthritis. It has proven to be an excellent treatment modality for reducing symptoms of AS 2-5. Unlike rheumatoid arthritis (RA), the benefits of TNF-inhibitor therapy on disease modification of AS has not been demonstrated to date.
Radiographic damage in AS is quantified by the number of bone spurs (syndesmophytes), squaring, erosions and sclerosis developing at vertebral corners. Quantified radiographic damage has been shown to correlate well with spinal mobility and overall physical function 6-9. Unlike rheumatoid arthritis and psoriatic arthritis, where TNF-inhibitors have demonstrated significant effect on progression of structural damage, the evidence to date is that the radiographic progression of AS is unaltered with the use of these agents 10-13. The only therapy showing promise for a disease modifying effect has been sustained use of non-steroidal anti-inflammatory drugs (NSAIDs) 14.
The impact of TNF-inhibitors on radiographic progression in AS has been difficult to resolve, in part because of the relatively slow tempo of radiographic change in AS, and the hurdles this imposes on longer-term placebo-controlled trials. Despite symptomatic improvement, 3 randomized controlled trials of TNF-inhibitors could not show significant benefit on structural progression when compared with historical controls. Prospective longitudinal cohorts can provide useful information in clinical settings in which longer periods of placebo treatment arms would not be feasible or ethically defensible. We studied the effect of TNF-inhibitors on radiographic progression in a well-characterized AS patient population enrolled in a protocol-based longitudinal study.
Methods
Patients
A prospective study of patients with AS satisfying the modified New York criteria included spinal radiographs every two years to assess structural progression. From this cohort, all patients having at least two sets of radiographs were included in this analysis. Three-hundred-and-thirty-four patients were included after excluding patients with total spinal ankylosis at baseline, as progression of disease cannot be assessed in this group. A comprehensive clinical evaluation and laboratory assessment was done on scheduled visits, at least once a year, using a standardized protocol.
Disease activity at baseline was assessed by a validated patient reported index, the Bath AS Disease Activity Index (BASDAI) as well as by erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP). In addition to these inflammatory markers, the following demographic variables were considered potential confounders in the model predicting progression of spine damage: age, age of onset of axial symptoms, duration of disease, HLA-B27 status, gender and smoking burden assessed by pack-year history. Radiographic disease severity in AS was assessed by a validated X-ray scoring method outlined below.
Radiographic scoring
Paired cervical and lumbar spine radiographs were available on all patients at a minimum interval of 1.5 years (mean 2.87±1.17 years; range 1.5 to 9 years). Independently one reader in USA (Reader 1) and two readers in Canada (Readers 2 and 3) scored the first and last available radiographs for each patient. All readers were blinded to the clinical details of the patient. The modified Stokes Ankylosing Spondylitis Spine Score (mSASSS) was used for scoring radiographic severity 15. Due to the unreliability of cervical spine squaring, this element was not scored in the radiographs 16. A change of 2 mSASSS units in 2 years (rate ≥ 1 unit/year) was defined as significant progression in AS and all patients who satisfied this criteria were labelled progressors 17,18. For this analysis, missing mSASSS corners were not considered for calculating progression as this could lead to false values from assigning average scores to these corners.
For quantifying reliability of radiographic scoring, we calculated the intra-class correlation coefficient (ICC) for status scores (one time point) and change scores. For this analysis, the three readers who were involved in mSASSS scoring read 30-paired radiographs. The readers were blinded to the sequence of films. The scores were analyzed using analysis of variance (ANOVA) statistics and the two-way mixed absolute agreement model with single measure estimate. ICC values for agreement are rated as poor (0.00–0.40), fair to good (0.40–0.75), and excellent (>0.75) 19. The ICC of status score for reader pairs 1 and 2 was 0.98 (95% C.I. 0.97 – 0.99; p < 1 × 10−3), for 1 and 3 was 0.97 (0.95-0.98) and for readers 2 and 3 it was 0.99 (0.98-0.99). The ICC of change scores for reader pairs 1 and 2 was 0.58 (95% C.I. 0.34 – 0.76; p < 1 × 10−3), for 1 and 3 was 0.64 (0.41-0.79) and for readers 2 and 3 it was 0.79 (0.63-0.88).
Treatment parameters
An NSAID index was calculated according to Assessment of SpondyloArthritis international Society (ASAS) recommendations 20. Patients taking the full, recommended dose of the respective NSAID for the entire period between radiographs receive a score of 100. Those not taking NSAIDs got a score of 0. Patients with varying NSAID use were given an average NSAID index for the entire period of follow up. An NSAID index of 50 was used as a cut off point to identify patients as high and low NSAID users as done previously 21.
TNF-inhibitors were assessed as a dichotomous variable initially. Patients who were exposed to these therapies were compared to biologic-naïve patients. TNF-inhibitor exposure was quantified as the duration of therapy and the proportion of disease duration (defined as the time since onset of axial symptoms) for which the patient was receiving the drug. Patients who were on TNF-inhibitors for more than 50% of their disease duration were considered sustained users. Delay in starting biologics was calculated as the time interval between the onset of AS symptoms and the start date of the TNF-inhibitor. If the first radiograph available was performed after starting therapy, the time on therapy before the first radiograph set was calculated.
Statistical Analysis
Student t-test, Chi-square test and logistic regression model were performed where applicable. For the multivariable analysis (MVA), variables were selected from those that were significant in the univariable analysis in addition to other potential confounders. In the MVA, exposure to TNF-inhibitors (yes/no) was used in the initial model followed by related variables such as duration and delay in starting therapy. As this was an observational study, the patients were not randomized to the treatment arm. Factors that could affect the decision to treat with TNF-inhibitors was addressed by doing a sub-analysis on a select group of patients who could be matched one to one in the two arms based on the propensity to receive the drug. To compare patients with the same propensity of being treated by a TNF-inhibitor, propensity score matching (PSM) was performed and an additional analysis was performed on the post-matched sample. As baseline BASDAI is one of the accepted criterion for initiating TNF-inhibitor therapy, PSM was done with baseline BASDAI as a predictor of receiving treatment with a caliper of 0.1. The standardized difference in BASDAI was 0.04 after matching. For tighter matching, a separate propensity matching was performed with a caliper of 0.05 and the following extended number of variables: gender, HLA-B27, baseline BASDAI, baseline ESR, baseline mSASSS, NSAID index, disease duration, baseline smoking pack-years, age of onset of symptoms. The standardized mean difference of all variables in the extended model post-matching was below 10% for all variables except HLA-B27 where the standardized mean difference was 0.11 (Supplementary-Table-1). The distribution of propensity scores post-matching was similar in the TNF-inhibitor exposed and unexposed groups (Supplementary-Figure-1). UVA and MVA analyses were performed as before in the post-matched samples.
Unmeasured confounders could still influence the results of observational studies and to evaluate how such an unmeasured potential confounder could affect the findings, we performed a sensitivity analysis. The sensitivity analysis examines the relationship between confounder prevalence in the non-progressor and progressor groups and how strong the effect of the confounder needs to be (odds ratio of the confounder) to negate the results of the study.
Considering the possibility of duration of follow up having an effect on the results we did a separate analysis to control for this variable. Interval between X-rays was added as a covariate in the MVA. To control for possible variability in radiographic scoring the readers were added as variables separately in the MVA. A separate analysis was done looking at the outcome of progression with an alternate definition (delta mSASSS= mSASSS at last visit - mSASSS at baseline) and accounting for the gap between radiographs. Age of onset, gender, HLA-B27 and NSAID index were controlled for in this analysis in addition to the baseline values for mSASSS, ESR, BASDAI and smoking (pack-years). More than 55% of patients in the cohort did not show any progression over the period of follow up and hence a Zero-Inflated Negative Binomial (ZINB) model was conducted to account for a significant proportion of zero values in over-dispersed delta mSASSS.
Statistical analyses were done using PASW 21, MatchIt package in R (version 2.14.2) and SAS 9.3 (SAS Institute Inc, Cary, NC). The respective institutional ethics boards approved the study and patients’ informed consents were obtained.
Results
The mean age of patients was 40.7 ± 13.8 years. In the AS cohort 76.7% were males and 83.4% were HLA-B27 positive. The mean disease duration was 16.4 ± 12.8 years and the mean age of onset of AS was 24.2±9.9 years. The mean baseline CRP was 1.48±1.96 mg/L and the mean ESR was 17.7±19.5mm per hour. The mean baseline mSASSS score was 9.6±14.5. No baseline spinal radiographic abnormality was seen in 144 patients (43.11%) patients and 102 (30.5%) patients progressed, as defined by a change of ≥ 1 mSASSS unit/year.
Out of 334 patients included in the study, 201 (60%) had received TNF-inhibitors for a mean duration of 2.5 ± 2.8 years. Four patients had stopped anti-TNF therapy before the baseline radiographs and 93 (27.8%) had initiated therapy before the baseline radiographs for a mean duration of 1.0 ± 1.8 years. Patients treated with TNF-inhibitors, as expected, had a higher BASDAI at baseline (Table 1). All other characteristics were similar between the groups, except for a higher percentage of males in the TNF-inhibitor treated group (Table 1). The mean delay in starting therapy after onset of disease was 15.2 ± 11.3 years, and 60% of those on therapy (N=120) had a delay of more than 10 years from disease onset to the initiation of therapy.
Table 1.
Baseline Variables in TNF-inhibitor users and non-users (naïve patients)
| Variable | TNF-inhibitor Naïve (N=133) | TNF-inhibitor Treated (N=201) | p value |
|---|---|---|---|
| Age (yrs) | 42.50±14.6 | 39.43±13.2 | ns |
| Female (%) | 32.33 | 17.41 | 0.002 |
| HLA-B27 (%) | 84.96 | 82.41 | ns |
| Disease Duration (yrs) | 16.38±14.4 | 16.47±11.8 | ns |
| Baseline mSASSS | 8.20±13.8 | 10.60±14.9 | ns |
| Smoking (pack-yrs) | 3.34±8.3 | 3.87±8.0 | ns |
| ESR (baseline) | 17.02±17.3 | 18.11±20.9 | ns |
| CRP (baseline) | 1.69±1.9 | 1.33±2.0 | ns |
| BASDAI (baseline) | 3.61±2.4 | 4.64±2.5 | 0.001 |
ns indicates p-value > 0.05
Baseline radiographic damage and radiographic progression
In the univariable logistic regression (Table 2), baseline radiographic damage (defined as the baseline mSASSS) was associated with progression (OR=1.07; 95% CI= 1.05-1.09; p<0.001). The rate of progression of the mSASSS scores was higher in males than females (1.2±1.9 vs 0.7±1.7 mSASSS unit/year; p=0.04). However, after controlling for baseline mSASSS, gender was not associated with progression significantly (Table 2). Similarly, age of onset of axial symptoms, duration of disease, HLA-B27 status and baseline BASDAI were not associated with progression after controlling for baseline mSASSS (Table 2).
Table 2.
Regression analysis in the full cohort (N=334) to identify factors associated with progression
| Variable | Odds Ratio | 95% CI | p value (UVA) | p value (MVA: All Variables) | |
|---|---|---|---|---|---|
| Lower | Upper | ||||
| Baseline mSASSS | 1.07 | 1.05 | 1.09 | <0.001 | <0.001 |
| Baseline ESR (mm/Hr) | 1.02 | 1.01 | 1.04 | <0.001 | 0.005 |
| Baseline BASDAI | 1.10 | 0.98 | 1.2 | 0.09 | 0.15 |
| Smoking (Pack Years) | 1.06 | 1.02 | 1.09 | 0.002 | 0.03 |
| Male vs Female | 1.52 | 0.78 | 2.97 | 0.22 | 0.25 |
| Age of onset | 1.02 | 0.99 | 1.05 | 0.07 | 0.02 |
| Disease duration | 1.01 | 0.99 | 1.03 | 0.26 | 0.08 |
| HLA-B27 | 1.08 | 0.54 | 2-17 | 0.82 | 0.58 |
| TNF-inhibitor use | 0.52 | 0.30 | 0.88 | 0.02 | 0.03 |
| NSAID-index | 1.01 | 0.99 | 1.01 | 0.15 | 0.71 |
| Multivariable Analysis including only variables significant in the univariable analysis# | |||||
| Baseline mSASSS | 1.06 | 1.04 | 1.08 | <0.001 | |
| Baseline ESR (mm/Hr) | 1.02 | 1.01 | 1.04 | 0.004 | |
| Baseline BASDAI | 1.10 | 0.95 | 1.25 | 0.09 | |
| Smoking (Pack Years) | 1.05 | 1.01 | 1.09 | 0.009 | |
| Age of onset | 1.03 | 0.99 | 1.06 | 0.07 | |
| TNF-inhibitor use | 0.47 | 0.24 | 0.94 | 0.03 | |
UVA: Univariate Analysis; MVA: Multivariable Analysis; CI: Confidence Interval
In the restricted MVA model only those variables that were significant in the UVA at p≤0.1 were included.
Relationship of baseline inflammatory markers to radiographic progression
Baseline values of ESR (OR=1.02; 95% CI=1.01-1.04; p<0.001) and CRP (OR=1.02; 95% CI=1.00-1.04; p=0.03) were associated with progression after controlling for baseline radiographic damage. Baseline ESR is more generalizable across different sites and used in the multivariable models of progression as a potential confounder. As CRP testing was done separately in the different centers, the values were standardized for all patients enrolled and retested. The standardized baseline CRP (ZCRP) remained a significant factor associated with radiographic progression (OR=1.59; 95% CI=1.24-2.03; p<0.001).
Smoking and radiographic progression
In the entire cohort 61.1% (N=204) had never smoked while 13.5% (N=45) had smoked more than 10 pack-years. In the univariable analysis, smokers were more likely to progress (OR=1.7; 95% CI = 1.1-2.8; p=0.02). The rate of mSASSS progression increased with increasing pack-years of smoking with the highest rate seen in smokers with more than a 10 pack-year history (Figure 1). Compared to those patients who smoked less, patients who smoked more than 10 pack-years had a significantly higher rate of progression (1.9 ± 2.0 vs 0.9 ± 1.9, p=0.005). In the univariable model, smoking pack-years was significantly associated with progression (OR=1.06; 95% CI=1.02-1.09; p=0.002).
Figure 1. Smoking and Rate of Radiographic Progression.
Comparison of the rate of mSASSS progression in patients who smoked compared to those who do not. The rate of progression was significantly different from the group of patients who either did not smoke or smoked less than 10 pack-years. A clear increase in progression rate is seen with increasing total burden of smoking.
Non-steroidal anti-inflammatory drugs and radiographic progression
The NSAID index was calculated as described above. There was significant variability in the intake of NSAIDs and the mean index was 35.36 ± 37.51. Twenty seven percent of patients had not taken NSAIDs during the observation period. A total of 117 (35%) had an NSAID index of above 50 and were considered high users. The NSAID index was not associated with radiographic progression. There was no significant difference in the rate of progression between high or low NSAID users.
TNF-inhibitors and radiographic progression
After correcting for the baseline mSASSS, those who had received TNF-inhibitors had a 50% reduction in odds of progression compared to those who had not been on TNF-inhibitors (OR: 0.52; CI: 0.30-0.88; p=0.02). The total duration of treatment was inversely associated with progression (OR=0.90; CI: 0.82-0.99; p=0.04). The duration of treatment expressed as a percentage of total disease duration was significantly associated with radiographic progression after adjusting for baseline mSASSS. Patients who were on biologics for more than 50% of their disease duration had lower odds of progression (OR=0.2 95% CI: 0.04-0.92; p=0.04) compared to patients who were not. Patients who were not on TNF-inhibitors for the greater part of their disease duration had increasingly higher rates of mSASSS progression (Figure 2A).
Figure 2. TNF-inhibitors and rate of radiographic progression.
This shows the effect of duration and point of starting therapy among AS patients on TNF-inhibitors. (A) The duration of therapy in relation to the total duration of disease is plotted against the rate of progression. As patients remained on therapy for a greater proportion of their disease duration, the rate of radiographic progression decreased. There was a significant difference in the rate of progression in those patients who remained on the drug for more than 50% of their disease duration compared to others. (B) The delay in starting TNF-inhibitors was significant in determining the effect it had on the rate of progression of radiographic damage. The earlier the therapy was initiated, the better the results. (C) Cumulative probability plots show a striking difference in radiographic progression, with patients who got TNF-inhibitors within 10 years of onset of disease progressing at a much slower rate compared to those who received the drug after 10 years.
In the patients who were on TNF-inhibitors, the rate of mSASSS progression increased with an increasing delay in starting treatment (Figure 2B). In the regression analysis after correcting for the baseline mSASSS, patients with a delay in starting therapy of more than 10 years were more likely to progress compared to those who started earlier (OR=2.4; 95% CI: 1.09-5.3; p=0.03). There was a significant impact observed when the probability plots of patients who received TNF-inhibitors within 10 years of symptom onset were compared to those who started treatment later (Figure 2C). Thus not only the use of TNF-inhibitors but also the total duration and delay in starting therapy were important in determining the rate of mSASSS progression.
Independent factors associated with radiographic progression
Stringent MVA was done with multiple models (Table 2). In the final model, we adjusted for variables that were considered as potential confounders. Significant variables in the final model included baseline mSASSS baseline ESR, smoking pack-year history, and TNF-inhibitors (Table 2). In the final model multivariable model, TNF-inhibitor use was significantly associated with more than 50% reduction in the adjusted odds of progression (Adj. OR: 0.47; 95% CI: 0.24-0.94; p=0.03).
Sensitivity analysis was used to evaluate the magnitude of any unknown confounders on measures of associations reported in this study. The sensitivity analysis (Supplementary-Figure-2), examines the relationship between prevalence of confounders in the non-progressor and progressor groups and the odds ratio of the confounders with respect to TNF-inhibitor use. For example, if the prevalence of an unmeasured confounder in the non-progressor group was 30% and 5% in the progressor group (green line), the odds ratio of the confounder would need to be about 6.5 or higher to account for the significant differences observed in TNF-inhibitor use. So the odds ratio of the confounder required to nullify the observed differences in TNF-inhibitor use is more than 6.5.
Propensity score matching to confirm the effect of TNF-inhibitors on radiographic progression
A total of 142 patients could be included in the post-match analysis after PSM with the extended variables described in the methods. A separate PSM with BASDAI alone including 194 patients was performed. Univariable followed by multivariable logistic regression analysis in the post-match samples were done (Table 3). The final multivariable model here included smoking (in pack-years), gender, age of onset, TNF-inhibitor use and baseline values of mSASSS and ESR. As disease duration was strongly correlated with baseline mSASSS in the post match cohort (R=0.52; p<0.001), disease duration was not included in the complete model. However, neither inclusion of disease duration in the complete model nor other potential confounders in the final model influenced the association between use of TNF-inhibitors and progression.
Table 3.
Logistic Regression Analysis in the post-matched samples (N=142, Extended Model PS)
| Variable | Odds Ratio | 95% CI | p value (UVA) | p value (MVA: All Variables) | |
|---|---|---|---|---|---|
| Lower | Upper | ||||
| Baseline mSASSS | 1.05 | 1.03 | 1.08 | <0.001 | 0.002 |
| Baseline ESR (mm/Hr) | 1.02 | 0.99 | 1.04 | 0.06 | 0.07 |
| Baseline BASDAI | 1.06 | 0.90 | 1.24 | 0.51 | 0.71 |
| Smoking (Pack Years) | 1.1 | 1.03 | 1.15 | 0.005 | 0.01 |
| Male vs Female | 2.01 | 0.72 | 5.7 | 0.18 | 0.21 |
| Age of onset | 1.03 | 0.99 | 1.07 | 0.12 | 0.50 |
| Disease duration | 1.02 | 0.98 | 1.05 | 0.32 | NI* |
| HLA-B27 | 0.64 | 0.23 | 1.77 | 0.39 | 0.44 |
| TNF-inhibitor use | 0.38 | 0.16 | 0.89 | 0.03 | 0.01 |
| NSAID-index | 1.01 | 0.99 | 1.02 | 0.35 | 0.53 |
| Multivariable Analysis including only variables significant in the univariate analysis* | |||||
| Baseline mSASSS | 1.05 | 1.02 | 1.08 | 0.001 | |
| Baseline ESR (mm/Hr) | 1.02 | 0.99 | 1.05 | 0.08 | |
| Male vs Female | 2.3 | 0.75 | 6.98 | 0.15 | |
| Smoking (Pack Years) | 1.09 | 1.02 | 1.17 | 0.01 | |
| Age of onset | 1.02 | 0.98 | 1.07 | 0.37 | |
| TNF-inhibitor use | 0.30 | 0.11 | 0.78 | 0.01 | |
UVA: Univariate Analysis; MVA: Multivariable Analysis; CI: Confidence Interval NI: Not included (Disease duration was strongly correlated with baseline mSASSS in the post-match cohort and was not included in the MVA. Inclusion of disease duration in the MVA did not however affect the level of significance of TNF-inhibitor use).
Restricted model includes variables that were significant in the univariate analysis with p value ≤ 0.2. After including only variables significant at p ≤ 0.1, TNF-inhibitor use remained strongly associated with less progression (not shown).
TNF-inhibitor use was associated with a 70% reduction in odds of progression in the MVA model (OR: 0.30; 95% CI: 0.11-0.78; p=0.01). Smoking and baseline mSASSS were also significantly associated with progression. ESR at baseline was marginally significant (p=0.08) in the MVA model. Standardized CRP, used in place of ESR in MVA, was significantly associated with progression (OR= 1.6; 95% CI: 1.01-2.51; p=0.05). The use of standardized CRP in place of ESR did not have a significant effect in the MVA model. In additional PSM studies with BASDAI alone as predictor, 194 patients could be matched and TNF-inhibitor use remained significantly associated with progression (OR: 0.43; 95%CI: 0.20-0.94; p=0.03) in the MVA.
The duration of TNF-inhibitor use, when used in the model in place of the dichotomous variable ‘TNF-inhibitor use’, was inversely associated with progression (OR=0.84; 95% CI: 0.73-0.96; p=0.01) after adjusting for baseline radiographic damage, ESR, smoking (in pack-years), gender and age of onset. Except for 3 patients, all TNF-inhibitor treated patients had at least 1 year of treatment between the first and last X ray. When these 3 patients were excluded from the analysis, TNF-inhibitor use remained inversely associated with progression (OR: 0.30; 95% CI: 0.11-0.82; p=0.02).
As this was not a randomized controlled trial, patients were followed for varying lengths of time. The gap between first and last x-rays scored was similar between the TNF-inhibitor user and non-user groups in terms of mean duration ± SD (2.9±1.03 vs 2.8±1.1 years) and range (1.5-5.5 vs 1.6-6.3 years). The results remained significant after adjusting for the duration of follow up in the MVA model with the odds of progression on TNF-inhibitors being 0.29 (95% CI 0.11-0.79; p=0.01). In the zero inflated negative binomial model (Figure 3) when the gap between X-rays increased above 3.9 years, TNF-inhibitor use was significantly associated with a lower delta mSASSS (Relative Ratio (RR)=0.42, 95% CI= [0.18, 0.98], p=0.04). There was no significant difference between the groups when the gap between radiographs was less than 3.9 years.
Figure 3. Zero-Inflated Negative Binomial Model for change in mSASSS scores over time.
There were several patients who did not progress during the follow up period. Zero- inflated negative binomial model is conducted to account for many portions of zeros (55.4%) in over dispersed delta mSASSS. TNF-Inhibitor use was significantly associated with less progression (defined by change in mSASSS), when patients were followed up for more than 3.9 years. There was no significant difference between two groups during early time period. The expected decrease of delta mSASSS in the TNF-inhibitor user group compared to non-user group was 58% (Relative Ratio (RR)=0.42, 95% CI= [0.18, 0.98], p=0.044).
Discussion
This is the first study to show a clear association between the use of TNF-inhibitors and progression of damage in AS. Both the duration of therapy and the timing of treatment initiation are important in the ultimate effect on the rate of damage in the spine of AS patients. The long duration of follow-up and large number of patients and controls in this study have contributed to its strength. The comparison with a control group of AS patients, not receiving TNF-inhibitors, from the same cohorts and followed concurrently in a real world setting was a noteworthy advantage. The large number of patients in our cohort allowed us to do PSM. This is an important aspect of the study as PSM reduces bias in the selection of patients in observational studies. Thus patients in the two groups had equal propensity to receive TNF-inhibitors. An additional sensitivity analysis indicates that unobserved confounders have to be very strong to negate the findings of this study, suggesting our observations are true with a very high likelihood. Previous studies examining the effect of TNF-inhibitors on patients included in randomized controlled trials were followed for shorter duration and the comparison group was a historical cohort 10-12.
An important observation arising from this study is the therapeutic advantage of both early intervention with TNF-inhibitors and of total duration of treatment. Prior studies that attempted to demonstrate disease-modifying effects of TNF-inhibitors were hampered by the structure of the randomized controlled trials, which included patients with longstanding disease, and short observation periods. More than 40% of our AS patients on biologics had started TNF-inhibitor therapy with less than 10 years of disease duration, which is a different profile from many other reports of biologics in AS, and this could have influenced our results. Any effect of duration of follow-up has been minimized in an additional MVA model controlling for gap between X-rays. Moreover in the post-matched sample the duration of follow-up was essentially identical in the two cohorts. In an additional analysis, patients were grouped based on the gap between x rays and a different dependent variable for progression (delta mSASSS) was used. This analysis showed a significant effect of TNF-inhibitors in reducing the progression of mSASSS with longer follow up.
We have shown the effect of smoking on radiographic progression independent of baseline inflammation, damage and the use of TNF-inhibitors. This is the first study in Ankylosing Spondylitis to show a strong effect of level of smoking on progression. Several studies have shown that AS patients who smoke have worse functional outcomes and greater radiographic severity22-25. In early disease cohorts, smoking has recently been assessed. In the Devenir des Spondylarthropathies Indifférenciées Récentes (DESIR) cohort smokers had more inflammation on MRI in both the sacroiliac joints and spine in addition to more structural damage by the mSASSS 26. In the German Spondyloarthritis Inception Cohort (GESPIC), smoking was significantly associated with radiographic progression at 2 years in patients with axial spondyloarthritis or AS with disease duration ≤ 10 years 17. None of these patients were exposed to TNF-inhibitors.
It is not surprising that we did not find baseline demographic factors that significantly modify the risk of progression of structural damage. Several studies looking at candidate predictors of mSASSS progression have not identified baseline demographic features, including HLA-B27 and gender 17,18,27. The strong effect of baseline radiographic severity likely negates other minor effects. In a recent study, when baseline radiographic severity was not included in the MVA, gender and age were important predictors of progression 28. An earlier cross-sectional study from our cohort assessed gender differences in AS 29. In patients with long standing AS, radiographic severity (assessed by Bath Ankylosing Spondylitis Radiographic Index [BASRI]) was significantly worse in males. Baraliakos et al found no gender effect on mSASSS progression among 146 AS patients 27. They did report that rapid progressors were more likely to be males, however this was not controlled for baseline radiographic severity. A more recent study of radiographic progression in axial spondyloarthritis showed only a trend to more progression in men compared to women 17.
NSAIDs have been proposed to be disease modifying in AS 21,30,31. Two studies and one re-analysis of the data have reported this finding. We have not been able to show the same effect, though these are not similar cohorts or study designs. There was a suggestion that NSAIDs are effective when used continuously in patients who had higher baseline inflammation and more radiographic damage. We could not demonstrate any effect of NSAID therapy after correcting for baseline CRP or ESR values. There was no difference in the high NSAID index (>50/100) patients when compared to those with low NSAID index (<50/100). There was no effect when the biologic-naïve and TNF-inhibitor user groups were tested separately. Despite controlling for several potential confounding factors and after multiple analyses, we did not see a disease modifying effect of NSAIDs on radiographic progression in AS.
The limitations of this study include potential unmeasured confounders and potential biases that cannot be adjusted for, as this was not a randomized controlled trial. In this study, TNF-inhibitors were given to patients who were not well controlled on NSAIDs and had high disease activity. It is not clear if the effect on preventing radiographic progression would be present in patients who are otherwise well controlled. Hence the decision to start TNF-inhibitors should still be made based on patients’ symptoms and lack of response or toxicity to NSAIDs.
Conclusions
This is the first study to show a protective effect of TNF-inhibitors on radiographic progression in AS. TNF-inhibitors are most effective when started sufficiently early in the disease course and when treatment is sustained for a longer period of time. Smoking and baseline inflammation continue to be important risks and radiographic baseline damage remains the strongest predictor of future radiographic progression in AS.
Supplementary Material
Supplementary-Figure-1. Propensity score matching of patients exposed or not to TNF-inhibitors
Propensity matching was done with a caliper of 0.05 and the following variables: gender, HLA-B27, baseline BASDAI, baseline ESR, baseline mSASSS, NSAID index, disease duration, baseline smoking pack-years, age of onset of symptoms. The standardized mean difference of all variables in the extended model post-matching was below 15% for all variables. As seen in the figure, the distribution of propensity scores post-matching was similar in the TNF-exposed and unexposed groups.
Supplementary-Figure-2. Sensitivity Analysis by assessing unmeasured potential confounder effect needed to negate study findings
We evaluated how an unmeasured potential confounder could affect the findings of our study by conducting a sensitivity analysis. The sensitivity analysis examines the relationship between confounder prevalence in the non-progressor and progressor groups and the odds ratio of the confounder with respect to TNF-inhibitor use. For example, if the prevalence of an unmeasured confounder in the non-progressor group was 30% and 5% in the progressor group (green line), the odds ratio of the confounder would need to be about 6.5 or higher to account for the significant differences observed in TNF-inhibitor use. So the odds ratio of the confounder required to nullify the observed differences in TNF-inhibitor use is more than 6.5.
Acknowledgements and disclosures
The authors thank Dr. John Boscardin for his biostatistics expertise in addition to Grace Yoon, BA, University of California, San Francisco, Adele Carty, Renise Ayearst and Ammepa Anton, Toronto Western Research Institute, Vera Wirawan and Laura Diekman from University of Texas at Houston, Lorie Guthrie from the National Institute of Health, and Kelly Tillery and Tessa Scaffide from Cedars-Sinai Medical Center for help with study coordination and database management. NH is supported by the Arthritis Society, Canada and the CIBC Young Investigator Award (Arthritis Research Foundation, Toronto). LSG is supported by Rosalind Russell Medical Research Center for Arthritis, Spondylitis Association of America (Young Investigator award) and NIH/NIAMS P01 AR052915-06A1. Disclosures for NH and RDI include consultant/speaker honoraria received from Abbott Immunology, Canada, Janssen Rheumatology, Canada and Amgen/Pfizer/Wyeth Canada. LSG has been a consultant for Abbott USA and UCB. There are no disclosures for TJL, MHW, ML, MHR, MMW and JDR. The patients from Toronto are also included in the Spondyloarthritis Research Consortium of Canada (SPARCC) database. Administrative costs for the SPARCC database are supported in part by The Arthritis Society of Canada and by unrestricted education grants from by Janssen, Abbvie, Amgen and Pfizer. Industry had no role in the design of the study, nor in the analysis and interpretation of the data, nor in manuscript preparation. NH and LSG had complete access to all the data and take responsibility for the integrity of the data and the accuracy of the data analysis. The project described was supported by the National Center for Research Resources, Grant UL1RR033176, and is now at the National Center for Advancing Translational Sciences, Grant UL1TR000124. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.
Footnotes
Publisher's Disclaimer: This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process which may lead to differences between this version and the Version of Record. Please cite this article as an ‘Accepted Article’, doi: 10.1002/art.38070
References
- 1.Khan MA, van der Linden SM. Ankylosing spondylitis and other spondyloarthropathies. Rheum Dis Clin North Am. 1990;16(3):551–579. [PubMed] [Google Scholar]
- 2.Revicki DA, Luo MP, Wordsworth P, et al. Adalimumab reduces pain, fatigue, and stiffness in patients with ankylosing spondylitis: Results from the adalimumab trial evaluating long-term safety and efficacy for ankylosing spondylitis (ATLAS). J Rheumatol. 2008;35(7):1346–1353. [PubMed] [Google Scholar]
- 3.van der Heijde D, Dijkmans B, Geusens P, et al. Efficacy and safety of infliximab in patients with ankylosing spondylitis: Results of a randomized, placebo-controlled trial (ASSERT). Arthritis Rheum. 2005;52(2):582–591. doi: 10.1002/art.20852. [DOI] [PubMed] [Google Scholar]
- 4.Braun J, van der Horst-Bruinsma IE, Huang F, et al. Clinical efficacy and safety of etanercept versus sulfasalazine in patients with ankylosing spondylitis: A randomized, double blind trial. Arthritis Rheum. 2011;63(6):1543–1551. doi: 10.1002/art.30223. [DOI] [PubMed] [Google Scholar]
- 5.Inman RD, Davis JC, Jr, Heijde D, et al. Efficacy and safety of golimumab in patients with ankylosing spondylitis: Results of a randomized, double-blind, placebo-controlled, phase III trial. Arthritis Rheum. 2008;58(11):3402–3412. doi: 10.1002/art.23969. [DOI] [PubMed] [Google Scholar]
- 6.Boonen A, vander Cruyssen B, de Vlam K, et al. Spinal radiographic changes in ankylosing spondylitis: Association with clinical characteristics and functional outcome. J Rheumatol. 2009;36(6):1249–1255. doi: 10.3899/jrheum.080831. [DOI] [PubMed] [Google Scholar]
- 7.Landewe R, Dougados M, Mielants H, van der Tempel H, van der Heijde D. Physical function in ankylosing spondylitis is independently determined by both disease activity and radiographic damage of the spine. Ann Rheum Dis. 2009;68(6):863–867. doi: 10.1136/ard.2008.091793. [DOI] [PubMed] [Google Scholar]
- 8.Machado P, Landewe R, Braun J, Hermann KG, Baker D, van der Heijde D. Both structural damage and inflammation of the spine contribute to impairment of spinal mobility in patients with ankylosing spondylitis. Ann Rheum Dis. 2010;69(8):1465–1470. doi: 10.1136/ard.2009.124206. [DOI] [PubMed] [Google Scholar]
- 9.Wanders A, Landewe R, Dougados M, Mielants H, van der Linden S, van der Heijde D. Association between radiographic damage of the spine and spinal mobility for individual patients with ankylosing spondylitis: Can assessment of spinal mobility be a proxy for radiographic evaluation? Ann Rheum Dis. 2005;64(7):988–994. doi: 10.1136/ard.2004.029728. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.van der Heijde D, Salonen D, Weissman BN, et al. Assessment of radiographic progression in the spines of patients with ankylosing spondylitis treated with adalimumab for up to 2 years. Arthritis Res Ther. 2009;11(4):R127. doi: 10.1186/ar2794. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.van der Heijde D, Landewe R, Baraliakos X, et al. Radiographic findings following two years of infliximab therapy in patients with ankylosing spondylitis. Arthritis Rheum. 2008;58(10):3063–3070. doi: 10.1002/art.23901. [DOI] [PubMed] [Google Scholar]
- 12.van der Heijde D, Landewe R, Einstein S, et al. Radiographic progression of ankylosing spondylitis after up to two years of treatment with etanercept. Arthritis Rheum. 2008;58(5):1324–1331. doi: 10.1002/art.23471. [DOI] [PubMed] [Google Scholar]
- 13.Gran JT, Skomsvoll JF. The outcome of ankylosing spondylitis: A study of 100 patients. Br J Rheumatol. 1997;36(7):766–771. doi: 10.1093/rheumatology/36.7.766. [DOI] [PubMed] [Google Scholar]
- 14.Haroon N, Kim TH, Inman RD. NSAIDs and radiographic progression in ankylosing spondylitis bagging big game with small arms? Ann Rheum Dis. 2012 doi: 10.1136/annrheumdis-2012-201844. [DOI] [PubMed] [Google Scholar]
- 15.Creemers MC, Franssen MJ, van't Hof MA, Gribnau FW, van de Putte LB, van Riel PL. Assessment of outcome in ankylosing spondylitis: An extended radiographic scoring system. Ann Rheum Dis. 2005;64(1):127–129. doi: 10.1136/ard.2004.020503. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Ward MM, Learch TJ, Weisman MH. Cervical vertebral squaring in patients without spondyloarthritis. J Rheumatol. 2012;39(9):1900. doi: 10.3899/jrheum.120322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Poddubnyy D, Haibel H, Listing J, et al. Baseline radiographic damage, elevated acute-phase reactant levels, and cigarette smoking status predict spinal radiographic progression in early axial spondylarthritis. Arthritis Rheum. 2012;64(5):1388–1398. doi: 10.1002/art.33465. [DOI] [PubMed] [Google Scholar]
- 18.Haroon N, Maksymowych W, Rahman P, Tsui F, Inman R. LMP2 and ERAPI variants are associated with radiographic severity in ankylosing spondylitis in the SPARCC cohort. J Rheumatol. 2011;38(6):1147–1147. [Google Scholar]
- 19.Salaffi F, Carotti M, Garofalo G, Giuseppetti GM, Grassi W. Radiological scoring methods for ankylosing spondylitis: A comparison between the bath ankylosing spondylitis radiology index and the modified stoke ankylosing spondylitis spine score. Clin Exp Rheumatol. 2007;25(1):67–74. [PubMed] [Google Scholar]
- 20.Dougados M, Simon P, Braun J, et al. ASAS recommendations for collecting, analysing and reporting NSAID intake in clinical trials/epidemiological studies in axial spondyloarthritis. Ann Rheum Dis. 2011;70(2):249–251. doi: 10.1136/ard.2010.133488. [DOI] [PubMed] [Google Scholar]
- 21.Poddubnyy D, Rudwaleit M, Haibel H, et al. Effect of non-steroidal anti inflammatory drugs on radiographic spinal progression in patients with axial spondyloarthritis: Results from the german spondyloarthritis inception cohort. Ann Rheum Dis. 2012 doi: 10.1136/annrheumdis-2011-201252. [DOI] [PubMed] [Google Scholar]
- 22.Kaan U, Ferda O. Evaluation of clinical activity and functional impairment in smokers with ankylosing spondylitis. Rheumatol Int. 2005;25(5):357–360. doi: 10.1007/s00296-004-0451-3. [DOI] [PubMed] [Google Scholar]
- 23.Averns HL, Oxtoby J, Taylor HG, Jones PW, Dziedzic K, Dawes PT. Smoking and outcome in ankylosing spondylitis. Scand J Rheumatol. 1996;25(3):138–142. doi: 10.3109/03009749609080003. [DOI] [PubMed] [Google Scholar]
- 24.Ward MM, Hendrey MR, Malley JD, et al. Clinical and immunogenetic prognostic factors for radiographic severity in ankylosing spondylitis. Arthritis Rheum. 2009;61(7):859–866. doi: 10.1002/art.24585. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Doran MF, Brophy S, MacKay K, Taylor G, Calin A. Predictors of longterm outcome in ankylosing spondylitis. J Rheumatol. 2003;30(2):316–320. [PubMed] [Google Scholar]
- 26.Chung HY, Machado P, van der Heijde D, D'Agostino MA, Dougados M. Smokers in early axial spondyloarthritis have earlier disease onset, more disease activity, inflammation and damage, and poorer function and health-related quality of life: Results from the DESIR cohort. Ann Rheum Dis. 2012;71(6):809–816. doi: 10.1136/annrheumdis-2011-200180. [DOI] [PubMed] [Google Scholar]
- 27.Baraliakos X, Listing J, von der Recke A, Braun J. The natural course of radiographic progression in ankylosing spondylitis: Differences between genders and appearance of characteristic radiographic features. Curr Rheumatol Rep. 2011;13(5):383–387. doi: 10.1007/s11926-011-0192-8. [DOI] [PubMed] [Google Scholar]
- 28.van Tubergen A, Ramiro S, van der Heijde D, Dougados M, Mielants H, Landewe R. Development of new syndesmophytes and bridges in ankylosing spondylitis and their predictors: A longitudinal study. Ann Rheum Dis. 2012;71(4):518–523. doi: 10.1136/annrheumdis-2011-200411. [DOI] [PubMed] [Google Scholar]
- 29.Lee W, Reveille JD, Davis JC, Jr, Learch TJ, Ward MM, Weisman MH. Are there gender differences in severity of ankylosing spondylitis? results from the PSOAS cohort. Ann Rheum Dis. 2007;66(5):633–638. doi: 10.1136/ard.2006.060293. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wanders A, Heijde D, Landewe R, et al. Nonsteroidal antiinflammatory drugs reduce radiographic progression in patients with ankylosing spondylitis: A randomized clinical trial. Arthritis Rheum. 2005;52(6):1756–1765. doi: 10.1002/art.21054. [DOI] [PubMed] [Google Scholar]
- 31.Boersma JW. Retardation of ossification of the lumbar vertebral column in ankylosing spondylitis by means of phenylbutazone. Scand J Rheumatol. 1976;5(1):60–64. [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary-Figure-1. Propensity score matching of patients exposed or not to TNF-inhibitors
Propensity matching was done with a caliper of 0.05 and the following variables: gender, HLA-B27, baseline BASDAI, baseline ESR, baseline mSASSS, NSAID index, disease duration, baseline smoking pack-years, age of onset of symptoms. The standardized mean difference of all variables in the extended model post-matching was below 15% for all variables. As seen in the figure, the distribution of propensity scores post-matching was similar in the TNF-exposed and unexposed groups.
Supplementary-Figure-2. Sensitivity Analysis by assessing unmeasured potential confounder effect needed to negate study findings
We evaluated how an unmeasured potential confounder could affect the findings of our study by conducting a sensitivity analysis. The sensitivity analysis examines the relationship between confounder prevalence in the non-progressor and progressor groups and the odds ratio of the confounder with respect to TNF-inhibitor use. For example, if the prevalence of an unmeasured confounder in the non-progressor group was 30% and 5% in the progressor group (green line), the odds ratio of the confounder would need to be about 6.5 or higher to account for the significant differences observed in TNF-inhibitor use. So the odds ratio of the confounder required to nullify the observed differences in TNF-inhibitor use is more than 6.5.