Background Tumour necrosis factor (TNF) blockers are known to increase the risk of serious infections in rheumatoid arthritis. Despite wide use of TNF blockers in ankylosing spondylitis (AS), the infection risk has never been evaluated in this disease.
Objectives To assess serious infections in patients with AS not exposed and exposed to TNF blockers.
Methods A systematic literature review up to May 2008 using PubMed, EMBASE and Cochrane Library was performed. All randomised controlled trials (RCTs) published between 1995 and 2008 monitoring serious infections, treated with non-steroidal anti-inflammatory drugs (NSAIDs) or TNF blockers, were included. Infection risks were calculated by naive pooling and for 100 patient-years (pyrs) of exposure. To assess the serious infection risk with TNF blockers, a meta-analysis of RCTs was performed using Mantel–Haenszel's method with several sensitivity analyses.
Results Fourteen RCTs were included (3345 patients). With placebo or NSAIDs (N=2202), two serious infections were observed (0.09%, range 0.01% to 0.3%)—that is, 0.4/100 pyrs. In TNF blocker trials, two serious infections were observed with placebo (2/500, 0.4% (0.0% to 1.4%), ie, 1.0/100 pyrs) versus 14 serious infections with TNF blockers (14/996, 0.7% (0.3% to 1.4%), ie, 1.9/100 pyrs). Meta-analysis of the RCTs showed that the increase in serious infections with TNF blockers compared with placebo was not significant: risk difference=0.4% (−8% to 1.6%).
Conclusions The absolute risk of serious infections in patients with AS not exposed to TNF blockers is low. The absolute risk of serious infections in patients receiving TNF blockers is higher, but the difference was found to be not significant, possibly through lack of power. Continued monitoring is necessary.
Statistics from Altmetric.com
However, in RA, serious infections (defined as life-threatening or requiring intravenous antibiotics or hospitalisation) are a common adverse event with a prevalence of 6–18% and an incidence of 6/100 patient-years (pyrs)8,–,10 as reported in post-marketing surveillance and observational studies of TNF blockers. Moreover, Bongartz et al11 have published a meta-analysis of the risk of serious infections in patients with RA receiving TNF blockers, indicating that the rate of serious infections is increased in these patients compared with patients receiving placebo (OR, 2.0, 95% CI 1.3 to 3.1).
In AS, TNF blockers are being prescribed more and more often; indeed they are now recommended for patients with persistently high disease activity despite conventional treatments according to the ASsessment in Ankylosing Spondylitis/European League Against Rheumatism recommendations.12 In view of the increase in infections with these treatments in RA, it is important to evaluate the infection risk in AS.
Consequently, the first objective of this study was to assess the risk of serious infections in patients with AS not exposed to immunosuppressive drugs, using to this end data from the placebo groups and non-steroidal anti-inflammatory drug (NSAID) treatment groups in randomised controlled trials (RCTs). The second objective was to determine the risk of serious infections with TNF blockers in RCTs and to compare this risk of serious infections with TNF blockers with the control groups in the same RCTs by a meta-analysis.
This systematic literature review and meta-analysis was conducted according to the Cochrane Collaboration guidelines.13
A systematic literature search was performed in PubMed and EMBASE using the keyword ‘spondylarthritis’ [MeSH] with the following limitations: years of publication, 1995–2007; study design, ‘randomised controlled trials’, ‘clinical trials’, ‘clinical controlled trials’ or ‘clinical trials phase II, III or IV’; languages, ‘English’ or ‘French.’ Only articles published after 1995 were selected because TNF blockers appeared after this date and we wished to obtain populations that were comparable with regard to infection risks. The Cochrane Central Register of Controlled Trials was searched using the keywords ‘spondylarthritis’ alone, then ‘spondylarthritis’ and ‘infections’ and ‘spondylarthritis and ‘TNFα inhibitors’ or ‘TNFα blockers’. The last update of searches was performed on 23 March 2007 for EMBASE and on 9 May 2008 for PubMed. Moreover, to verify the completeness of the data on infections with TNF blockers, the manufacturers of the three available TNF blockers (Schering-Plough/Centocor (Levallois-Perret, France), Wyeth (Paris, France) and Abbott) were contacted. Because of missing data, authors of several (Rungis, France) trials were also contacted and provided complementary data.
The trials were initially selected on the basis of their titles and abstracts. The inclusion criteria were adult populations with AS fulfilling modified New York classification criteria for AS.14 Only RCTs of NSAIDs or TNF blockers were included. In order to limit heterogeneity, articles assessing spondylarthritis other than AS according to the modified New York classification were excluded. After the initial selection on titles and abstracts, articles were analysed on the full text. All trials that did not monitor or report infectious events were excluded.
To assess the risk of serious infections in patients with AS not exposed to immunosuppressive drugs, data were obtained from placebo groups and NSAID treatment groups in RCTs. Regarding infections, NSAIDs were analysed with placebo. To determine the risk of serious infections with TNF blockers, RCTs of TNF blockers were selected and a meta-analysis compared treatment and control groups.
One investigator (AF-A) selected the articles, and two reviewers (AF-A and LJ-P) collected the data independently, using a predetermined form. Disagreements were resolved by consensus with a third reviewer (AB). The methodological quality of the selected studies was assessed using the Jadad Scale15 and intention-to-treat analysis.
For each trial, demographic characteristics (mean age, sex ratio), disease duration (weeks) and disease activity at baseline evaluated by the Bath Ankylosing Spondylitis Disease Activity index,16 therapy allocation (with doses, duration and numbers of patients in each group—NSAIDs and TNF blockers) and duration of follow-up were collected.
The number of patients with at least one serious infection was collected in placebo groups, NSAID treatment groups and TNF blocker groups of RCTs. In the literature, a serious infection is usually defined as life-threatening, fatal, requiring intravenous antibiotics or hospitalisation or resulting in persistent or significant disability. This definition was also used here; if another definition of serious infection was given in the trials, it was also recorded. When available, the characteristics of serious infections (localisation, organism and outcome) and deaths due to serious infections were noted.
Serious infections were expressed as percentage of patients with at least one infection. This percentage was given for each treatment group (placebo/NSAID and TNF blockers) by naïve pooling (sum of all serious infections divided by the sum of patients). CIs were assessed by the Clopper-Pearson method. For indicative purposes, incidence rates of infection for 100 pyrs were also calculated by naïve pooling (dividing the sum, on all the studies, of the number of serious infections by the sum of the exposure time). CIs were assessed using the method proposed by Ulm.17 To compare the percentage of patients with serious infection in TNF blocker and placebo groups of the TNF blocker RCTs, a meta-analysis of risk differences by Mantel–Haenszel's method was performed. The formula of the weight in Mantel–Haenszel's method for risk difference is (Narm1×Narm2)/(Narm1+Narm2), where Narmi is the number of patients in arm i (i=1,2). Note the weights are not sensitive to the absence of an event. This method was preferred to meta-analysis of ORs or RRs. Details of the method are given in Egger and Altman, p 294.18 Only the nine RCTs with a placebo control group were used. A fixed-effects model was used because of the absence of heterogeneity, tested using the Cochran test. Publication bias was explored by funnel plot. Data analysis was conducted using RevMan 4.2 statistical software (Revman, the Cochrane Collaboration, Oxford, UK). The number needed to harm (number of patients who should be treated to observe one extra event) was also calculated.
To check the robustness of the results of the meta-analysis, ORs and RRs were assessed using Mantel–Haentzel's technique with the continuity correction used by Bongartz et al11 and also using a continuity correction equal to 0.5 for each cell. Robustness was also tested by performing the meta-analysis, taking out one study at a time. Finally, to explore the impact of the study duration, a meta-analysis was performed on two subgroups of studies (risk difference by Mantel–Haenszel's method), according to the study duration (≤ vs > 16 weeks).
Literature search results and trial characteristics
Figure 1 shows the selection process of published trials for this systematic review. Of 356 potentially relevant publications retrieved during the initial PubMed, EMBASE and Cochrane search, 348 were excluded, mainly because they were not RCTs, reported no data on infections, reported interventions other than medication, or did not concern adult patients with AS. In addition, one of the pharmaceutical companies provided eight articles. Of these, six were literature reviews and only one reported appropriate data and was included. The Cochrane search yielded no additional studies. The last update of searches in May 2008 provided five more articles following the previous criteria. Finally, 14 trials2 3 19,–,30 were analysed—that is, a total of 3445 patients. In all, 74% were men, mean ± SD age was 41 ± 3 years, and mean duration of follow-up was 20 ± 14 weeks. The quality of the articles met the required standards (mean Jadad, 3.6 ± 1.0). The characteristics of the trials are summarised in table 1, and table 2 shows the characteristics of the nine RCTs reporting TNF blocker infections versus placebo.
Serious infection risks
Serious infections in patients with AS not exposed to TNF blockers
The risk of infection ranged from 0% to 0.9%, but only two studies2 26 reported infections in the placebo group (table 1). Of 2202 patients with AS treated with placebo or NSAIDs, two (0.09%, range 0.01% to 0.3%) had serious infections (table 3), which corresponds to 0.4 (0.0 to 4.5) serious infections per 100 pyrs. No deaths due to infection and no tuberculosis were reported. One study reported a serious infection in the placebo group but not in the treatment group.2
Serious infections with TNF blockers
The risk of serious infection ranged from 0% to 2.9% (table 3). Of 1243 patients with AS treated with TNF blockers, 14 (1.1%, range 0.6% to 1.9%) had at least one serious infection (table 3), which corresponds to 2.2 (1.2 to 3.6) serious infections for 100 pyrs. Serious infections with TNF blockers as reported in the articles are described in supplementary table 4 (supplementary data online). Serious infections occurred in patients treated with either infliximab (511 patients exposed) or etanercept (524 patients exposed). No serious infection was reported in patients treated with adalimumab (208 patients exposed). No deaths due to infection were reported with the three treatments.
Comparison of serious infection risks in TNF blocker trials by meta-analysis
In the meta-analysis of the nine RCTs of TNF blockers versus placebo (table 2), the rate of serious infections with TNF blockers was not significantly increased compared with placebo (risk difference=0.4%, 95% CI −8% to 1.6%; p=0.54) (figure 2). With this difference in risk, it would be necessary to treat 257 patients for 6 months to observe one additional infection with TNF blockers compared with placebo (number needed to harm). The CI around this result is, however, large: number needed to harm=125 to infinity to number needed to benefit 59.
No heterogeneity was detected (p=0.98 (Cochran's test) and I2 = 0). A fixed-effects model was therefore used. The funnel plot did not show any publication bias (data not shown).
Robustness of results
Several sensitivity analyses were performed; the results were similar. For example, Mantel–Haenszel's OR using the continuity correction used in Bongartz et al11 was 1.41 (95% CI 0.48 to 4.17). The other sensitivity analyses confirmed the results (data not shown).
The results of the meta-analysis were not influenced by the length of follow-up of the studies. The risk difference in the subgroup of studies with a length of follow-up of 16 weeks or less was 1% (range −2% to 3%) and 0.5% (−1% to 2%) in the studies with a longer follow-up.
According to the differences in risk of serious infections assessed in the meta-analysis, for the difference in risk to become significant, it would be necessary to add to the meta-analysis one balanced study with a total sample size of 5500 patients.
In this systematic review, the risk of serious infections in patients with AS not exposed to immunosuppressive drugs was very low (0.09%, ie, 0.4 per 100 pyrs). It should be noted that these infection risks refer to patients who are selected for clinical trials, and consequently cannot be considered as baseline infection risks. For patients with AS in TNF blocker trials, the risk of serious infections was higher—that is, 0.7% (1.9 per 100 pyrs) with TNF blockers versus 0.4% (1.0 per 100 pyrs) in the control group. However, this difference did not reach statistical significance on meta-analysis.
Few studies have investigated the incidence or prevalence of serious infections in patients with AS. A prospective cohort study indicated a high incidence of serious infections with infliximab treatment in AS, psoriatic arthritis or undifferentiated spondylarthropathy of ∼4.2/100 pyrs.31 However, they had a small population of 191.5 pyrs and no comparison with placebo. RCTs often have a small number of analysed patients with a brief exposure time, so they are often not suitable for demonstrating an increased risk of rare side effects. Open-label studies usually lack a control group; consequently establishing causality between a treatment and an event is impossible. An interesting and powerful alternative consists of pooling results of RCTs by a meta-analysis.18 This systematic review was performed using all available literature sources and included all published data to date. Furthermore, the meta-analysis was performed in accordance with the recommendations of the Cochrane Collaboration. The definition of serious infections was consensually used. No evidence for significant publication bias was apparent on a funnel plot32 (data not shown) and no statistical heterogeneity was detected. Furthermore, several methods of meta-analysis were applied, assessing the risk difference versus the OR. Both methods confirmed the robustness of the results. Several sensitivity analyses were also performed and confirmed the results. Therefore we consider these results to be valid.
However, this study does have shortcomings. Only published data were analysed, only English and French trials were selected, and, in spite of large searches on databases, some data may have been missed. Finally, biases in patient selection in RCTs may influence the final results in terms of risk of infections. This was noticeable when risks of infections in the placebo groups of trials were assessed, which varied according to the trials analysed. We observe that serious infections varied between 0.4 events per 100 pyrs (in NSAID trials) and 1.0 event per 100 pyrs (in TNF blocker trials) in the control groups. However, these figures are all low and lower, for example, than in RA.
The lack of significant increase in risk of serious infections with TNF blockers may be explained by a lack of power. Indeed, more infections were detected in the TNF blockers group; it would, however, necessitate a RCT of more than 5000 patients to reach statistical significance with the difference in risks observed. However, even in this case, the magnitude of serious infections would seem smaller than in RA as the observed risk of serious infections with TNF blockers was around 1% in the present meta-analysis (ie, around 2 events per 100 pyrs, which is much lower than in patients with RA treated with TNF blockers).
The few serious infections reported in the articles reviewed here were very heterogeneous, both viral and bacterial infections. Of the 1243 patients treated with TNF blockers, one case of tuberculosis was reported, without any case of opportunistic infection. No deaths due to infection were reported.
It is known that patients with RA are at increased risk of developing infectious diseases—around 6/100 pyrs in open-label studies.8,–,10 There are many possible explanations for the lower incidence of infections in patients with AS not exposed to immunosuppressive drugs compared with RA patients. Firstly, patients with AS are often younger, with a mean age around 40; consequently they have fewer comorbidities than patients with RA. Secondly, the physiopathology of the two diseases is very different. RA is characterised by an inflammatory synovitis due to the activation of T and B lymphocytes and proinflammatory cytokines. AS is characterised by both inflammation and new bone formation, with a strong genetic effect partially explained by the major histocompatibility complex class I antigen HLAB27. Finally, the treatment of these diseases also differs; the first-line treatment of AS is NSAIDs, whereas RA is more often treated with steroids, which are associated with increased rates of bacterial infection.33
In conclusion, for the first time, this meta-analysis shows that the risk of serious infection is very low in patients with AS included in trials and not exposed to immunosuppressive drugs, and indicates that the risk of serious infection was not increased with TNF blockers in RCTs, but that the difference was not significant. However, as TNF blockers are being prescribed more and more often in AS, longitudinal prospective studies are required to confirm these results; physicians should be aware of infections when prescribing these treatments.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
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.