Objectives Whether the increased risk of comorbidities, such as cardiovascular disease, in rheumatoid arthritis (RA) can be reverted by particular antirheumatic therapies, or response to these, is unclear but of critical clinical importance. We wanted to investigate whether response to tumour necrosis factor inhibitors (TNFi) translates into a reduced risk for acute coronary syndrome (ACS).
Methods A cohort of patients with RA initiating a first TNFi 2001–2012 was identified in the Swedish Biologics Register. The association between European League Against Rheumatism (EULAR) response after 3–8 months of treatment (assessed using the first, the best and the measurement closest to 5 months, respectively), and the risk of incident ACS during the subsequent year was analysed in Cox regression models. Adjustments included cardiovascular risk factors, joint surgery, RA duration, education and work disability.
Results During 6592 person-years among TNFi initiators (n=6864, mean age 55 years, 77% women), 47 ACS occurred. The adjusted HRs (95% CI), which were similar to the crude HRs, of the 1-year risk of ACS among EULAR good responders compared with non-responders were 0.5 (0.2 to 1.4), 0.4 (0.2 to 0.9) and 0.5 (0.2 to 1.2), for the first, the best and the evaluation closest to 5 months, respectively. EULAR moderate responders had equal risk to that of EULAR non-responders, who, compared with the general population referents (n=34 229), had a more than twice the risk of ACS. For good responders, there was no statistically significant difference in risk versus the general population.
Conclusions Optimised RA disease control has the potential to revert otherwise increased risks for ACS in RA.
- Rheumatoid Arthritis
- DMARDs (biologic)
- Cardiovascular Disease
Statistics from Altmetric.com
Patients with rheumatoid arthritis (RA) are at risk of cardiovascular comorbidity, in particular coronary heart disease.1 Since traditional risk factors only partly seem to explain this risk increase, factors related to the RA disease have been suggested to be of importance.2 Measurements of disease activity (eg, erythrocyte sedimentation rate (ESR) or Clinical Disease Activity Index (CDAI)) as well as disease severity (eg, extra-articular RA disease) have been associated with risk for cardiovascular disease (CVD).3–7
Drugs inhibiting tumour necrosis factor (TNFi) have proven to efficiently control inflammatory activity and prevent joint damage in RA.8 ,9 Since atherosclerosis share pathogenetic processes with chronic inflammatory diseases, inhibition of proinflammatory cytokines might be beneficial also in atherosclerotic coronary disease.10 Several studies have addressed the effect of pharmacological TNF inhibition on the risk of coronary artery disease in RA.11–21 In four of those, a statistically significant association between treatment with TNFi and a lower risk of coronary disease was observed, including our own previous study from the Swedish Biologics Register.14 ,17 ,20 ,21 These studies did not, however, distinguish between any direct effect of TNFi on the atherosclerotic heart disease, and any effect mediated via response to therapy, that is, via lowering of RA-related inflammation. In a previous study from the British Society for Rheumatology Biologics Register (BSRBR), including patients with RA who started TNFi therapy before 2006, therapy response was linked to a reduced risk of myocardial infarction (MI), although the risk of MI among the TNFi-treated patients overall did not differ from that in biologics-naïve patients.13 Reflective of British national guidelines for the use of biologic disease-modifying antirheumatic drug (DMARD),13 ,22 the TNFi-treated patients in that study had high disease activity (mean disease activity score 28-joint count (DAS28) 6.6 SD 1.0) and considerable functional impairment (mean Health Assessment Questionnaire (HAQ) 2.1 SD 0.6) at start of treatment. The results from the British study have not yet been replicated and may not, due to the extensive burden of RA disease in that cohort, be generalisable to other TNFi-treated RA populations.
Thus, while treatment with TNFi might lower the overall excess risk of cardiovascular comorbidity in patients with RA, it remains unclear, whether any such risk reduction applies to all treated patients, unrelated to disease activity, or is conditional on response to treatment. The aim of this study was therefore to investigate the effect of response to TNFi therapy on the short-term risk of acute coronary syndrome (ACS) in a contemporary cohort of patients with RA, and to contextualise these risks by comparison to those in the general population.
In Sweden, healthcare is publically funded, and use of biological treatment in RA is unrestricted, with a penetrance around 25%–30%. The Swedish Biologics Register has been used since 1998 for follow-up of patients receiving biological treatment for RA. All public and private rheumatology clinics report to the register, which has an estimated national coverage of 90%–95% regarding prescribed TNFi.23 ,24 Data on antirheumatic therapies, disease activity measurements (eg, ESR, patient’s global assessment of disease activity, swollen joint count, tender joint count, physician's global assessment of disease activity, HAQ and C-reactive protein (CRP)) are included in the register at start, 3 and/or 6 months and further on with 6–12 months intervals according to clinical practice. The DAS28 is calculated from the included variables.25 The register is based on the Swedish personal identification number allowing linkage with full coverage public registers collecting data on demography, mortality and morbidity (the Swedish National Population Register, Statistics Sweden, the Cause of Death Register, the National Patient Register).21 ,26 The International Classification of Disease codes and surgical procedure codes used in this study have been listed in online supplementary table S1.
Definition of exposure
The exposure, response to TNFi therapy, was defined according to the European League Against Rheumatism (EULAR) response criteria as none, moderate or good.27 Discontinuation of treatment within the evaluation time window with ‘lack of effect’ recorded as the reason for withdrawal was considered equivalent to ‘no response’. The time window for evaluation of response was set to 2–8 months (60–242 days) after the start of the treatment (see online supplementary figure S1 and S2). As 50% of the patients had two or more visits with DAS28 recorded within the evaluation visit time window, three separate definitions of response were defined using: (1) the first DAS28, (2) the best (lowest) DAS28 in the evaluation time window and (3) the DAS28 closest to 5 months (151 days). The latter was considered to correspond best with clinical practice and was therefore chosen for categories in descriptive information, subgroup and sensitivity analyses; however, the results were similar regardless of response definition. The mean time, days (SD), from start of the treatment to the evaluation visit with the first DAS28 was 120 (39) days, for the best DAS28 140 (48) days and correspondingly for the DAS28 closest to 5 months 139 (41) days.
The outcome, any first ever ACS, was defined as a primary discharge diagnosis of acute MI or unstable angina pectoris in the National Patient Register or acute MI as the underlying cause of death in the Cause of Death Register. This outcome definition has previously been validated in a Swedish early RA cohort indicating a positive predictive value of 95%.18
Follow-up started at the time point for evaluation of therapy response and ended at death, emigration, 31 December 2012, or at the time point for a first ACS and was limited to 1 and 2 years, respectively, following response evaluation.
From the Swedish Biologics Register 10 833 patients (≥18 years) with RA, with no previous ischaemic or congestive heart disease, who started treatment with a first biologic, a TNFi, between January 2001 and December 2012 were identified. After exclusions (figure 1) 7833 patients remained, of which 6864 patients had available data on EULAR response, including 343 patients who stopped treatment because of lack of effect. For each patient, five randomly assigned referents, matched for age, sex and county of residency at the time point for TNFi start, were identified from the Swedish Population Register resulting in a comparator cohort with 34 229 general population referents after exclusions.
Identification of covariates
From the National Patient Register, data on any previous hospitalisations or visits with a main or contributory diagnosis of diabetes mellitus, hypertension, cerebrovascular disease, other atherosclerotic disease and previous joint surgery were extracted. Data on socioeconomic factors such as education level, number of days on sick leave from work and disability pension in the year before the index date were retrieved from Statistics Sweden.
Sensitivity analyses evaluating alternative measurements of disease activity or response were included: categories of disease activity using DAS28, CDAI or Simplified Disease Activity Index (SDAI), CDAI and SDAI response and ACR/EULAR remission.28–33 In a sensitivity analysis, the follow-up was censored 30 days after the reported date of withdrawal of the TNFi (for any reason, ‘On-drug+30 days’). A sensitivity analysis with incident major cardiovascular event (MACE), including also stroke, transient ischaemic attack and cardiovascular death as outcome was performed. Sensitivity analyses with adjustments for additional cardiovascular risk factors (family history of premature CVD and lipid-lowering drug treatment) were performed in subsets of patients (see online supplementary tables S2 and S3). The unmeasured confounding by smoking habits was assessed in sensitivity analyses with quantitative bias modelling (see online supplementary table S4a and b).
Crude incidence rates for ACS were calculated. HRs of ACS associated with treatment response were estimated by Cox proportional hazards regression, comparing the risk among moderate or good responders, respectively, with non-responders, and comparing the risk among patients in each response group with the risk among the general population referents. The fully adjusted regression models were adjusted for sex, age, county of residency, inclusion year, cardiovascular risk factors (diabetes mellitus, hypertension, previous cerebrovascular and/or other atherosclerotic disease), joint surgery, RA duration >10 years, education level, sick leave and disability pension the year before inclusion. The proportional hazards assumption was tested by introduction of an interaction term of exposure and follow-up time in the models, and found not to be violated.
In a complementary analysis (n=7833), missing values (n=2967, 5%) for disease activity variables needed for the calculation of DAS28 were replaced using multiple imputation in fully conditional specification (10) models through SAS command ‘Proc MI’ in 10 imputed datasets (see online supplementary table S5). Each variable was modelled by linear regression as a function of the disease activity variables, age, sex, year of inclusion and event status.
The risk of ACS among patients excluded because of lacking visit(s) compared with patients with visits and the general population, respectively, was evaluated with follow-up from the date of TNFi start, maximised to 1 year and 5 months (corresponding to the follow-up in the main analyses, see online supplementary table S6).
Analyses were performed using SAS V.9.3 (SAS Institute, Cary, North Carolina, USA). The Stockholm Ethics Review Board approved the study and waived the requirement for individual consent for this register-based study.
Of the 6864 patients included (mean age (SD), 55 (13) years, 77% women), 73% had a registered diagnosis of seropositive RA (table 1). A majority of the patients started infliximab (38%), etanercept (36%) or adalimumab (22%) as their first TNFi, and only a small number of patients started one of the more recent TNFi's: golimumab 1.8% and certolizumab pegol 2.8%. EULAR good response at evaluation was observed in 2682 patients (39%), moderate response in 2353 patients (34%) and 1829 (27%) were non-responders using the DAS28 closest to 5 months (table 1). The corresponding percentages using the first DAS28 (38% good, 36% moderate and 27% no response) and using the best DAS28 (44% good, 33% moderate and 23% no response) were largely similar.
Among the EULAR response groups, moderate responders had indications of a more severe RA disease at baseline with the highest disease activity (DAS28, ESR, high CRP) before treatment, and the highest frequencies of disability pension, joint surgery and long RA duration at baseline (table 1). The group with good EULAR response had the lowest frequencies of disability pension, joint surgery and of several comorbidities: diabetes, hypertension and other atherosclerotic disease (table 1).
The first year of follow-up in the RA cohort, starting at the visit closest to 5 months, comprised a follow-up of 6592 person-years during which 47 ACS events occurred, resulting in a crude incidence rate (95% CI) of 7.1 (5.4 to 9.5) events/1000 person-years (table 2). Extending to 2 years of follow-up, the crude incidence rate was 6.7 (5.4 to 8.3) events/1000 person-years, based on 84 ACS events during 12 571 person-years (table 2). The other response definitions resulted in similar incidence rates (table 2). The corresponding 1-year and 2-year incidence rates in the matched general population comparator were 2.9 (2.4 to 3.6) and 2.9 (2.5 to 3.4)/1000 person-years, respectively (figure 2).
The crude incidence rate was lowest among good responders and ranged from 2.8 to 3.5/1000 person-years for 1-year follow-up and from 4.3 to 4.4/1000 person-years for 2-year follow-up (table 2). Also among moderate and non-responders, the crude incidence rates were comparable across response definitions, and ranged from 8.6 to 10/1000 person-years, and from 7.7 to 9.2/1000 person-years, for the 1-year and 2-year follow-ups, respectively (table 2).
In Cox regression models comparing the 1-year risk of ACS among EULAR good responders with non-responders, the age-adjusted and sex-adjusted HRs (95% CI) were 0.5 (0.2 to 1.1), 0.3 (0.1 to 0.8) and 0.4 (0.2 to 1.0), for first response, best response and closest to 5 months response, respectively (table 2). Use of a fully adjusted model had limited impact on the HRs: EULAR good versus no response HR 0.5 (0.2 to 1.4), 0.4 (0.2 to 0.9) and 0.5 (0.2 to 1.2) (table 2). The 2-year follow-up resulted in similar results for first response, best response and closest to 5 months response: fully adjusted HRs 0.7 (0.4 to 1.3), 0.6 (0.4 to 1.2) and 0.7 (0.4 to 1.2), respectively (table 2). In all models, the point estimates for moderate responders compared with non-responders were close to 1.0 (table 2).
Compared with the general population, patients with no or moderate response had 2.5–3 times higher 1-year risk of ACS: age-adjusted and sex-adjusted HR 3.0 (1.8 to 5.1) and 3.1 (2.0 to 4.9), fully adjusted HR 2.7 (1.7 to 4.4) and 2.6 (1.5 to 4.4), respectively (figure 2). By contrast, no statistically significant difference in the risk of ACS was observed among good responders, age-adjusted and sex-adjusted HR 1.3 (0.7 to 2.6) and fully adjusted HR 1.2 (0.6 to 2.4) (figure 2). The 2-year follow-up resulted in similar HRs, however, with a borderline significant difference comparing good responders with the risk in the general population: age-adjusted and sex-adjusted HR 1.7 (1.1 to 2.6), fully adjusted HR 1.6 (1.0 to 2.6) (figure 2).
Subgroup and sensitivity analyses
Analyses stratified by sex, age ≤64 years/>64 years, presence/absence of cardiovascular risk factor(s) or disease duration >10 years/≤10 years, although the interpretation was limited by a low number of events in most strata, resulted in HRs similar to those in the main analyses (see online supplementary table S7). Sensitivity analyses evaluating alternative response or disease activity measurements (DAS28, CDAI and SDAI categories, ACR/EULAR remission, CDAI and SDAI response) generally indicated a lower risk among patients with lowest disease activity or best response irrespective of index (table 3). Sensitivity analyses with the broader outcome definition, MACE, gave results in line with the main analyses, both with respect to comparison among patients, and with the general population (table 4). Similarly, analyses employing the alternative risk window ‘On-drug+30 days’ resulted in HRs similar to those of the main analyses (see online supplementary table S7), as did the results based on data with missing variables handled by multiple imputation, and regression models adjusted for lipid-lowering drug treatment and family history of premature CVD, respectively (see online supplementary tables S2, S3 and S5). A quantitative assessment suggested that unmeasured confounding by smoking would only explain a small fraction, if any, of the observed association (see online supplementary table S4a and b).
In a population-based RA cohort with universal healthcare access and a current penetration of biological therapy around 30%, we found that good (but not moderate) EULAR response was associated with a 50% lower risk of ACS compared with non-responders, and that the short-term risk among good responders was similar to the risk of ACS in the general population, but more than doubled among non-responders and moderate responders.
The results of all models, irrespective of stratification, exposure definition or risk window were remarkably consistent, showing a robust pattern of a markedly lower risk of ACS among patients with the most favourable state of the RA disease at evaluation. In the fully adjusted main analyses, the analysis using the lowest observed disease activity within the evaluation time window was formally statistically significant, HR 0.4 (0.2 to 0.9).
The finding of a reduced risk of ACS among good responders to TNFi is in keeping with the previously published results from the BSRBR.13 There are, however, a number of important differences between the BSRBR study and ours: patients in the BSRBR cohort had a more severe RA disease and represented a mix of individuals with/without previous ischaemic heart disease, and the definitions of outcome, follow-up and exposure differ.13 Nevertheless, good or moderate response in the BSRBR study resulted in magnitude of risk reduction comparable with the risk reduction in good responders in our study.13 We have previously approached a similar question in a study of the risk of ACS and TNFi treatment in patients with early RA, in which neither treatment with, nor response to, TNFi was associated with the risk of ACS.18 The inclusion of patients with early RA implicated a low burden of disease in the study population and the statistical precision of the estimate in the case–control study was limited.18
The nationwide, population-based cohort study design and the high coverage of the Swedish Biologics Register imply a high degree of generalisability. Linkage to public registers with high coverage ensured a high case ascertainment and generalisability. The inclusion of a general population comparator enabled a contextualisation of the risks, and is a major addition to previous reports. Furthermore, the sensitivity analyses were in line with the main analysis, indicating that the results were not dependent on the evaluation method, and the analyses with MACE as outcome supported the main findings. To minimise the impact of other conditions, subsequent treatment and other interventions, one risk window after response evaluation was limited to 365 days. Other definitions of the follow-up (2 years or On-drug+30 days) did not considerably change the risk estimates, however, keeping in mind that treatment response is a transient state and that the median drug survival on TNFi is <3 years, the 2-year follow-up showed, not surprisingly, somewhat weaker associations.
Some limitations should be taken into consideration. The fully adjusted regression models were adjusted for traditional cardiovascular risk factors: diabetes, hypertension and previous atherosclerotic disease(s) as identified by diagnoses in the National Patient Register, and for socioeconomic variables such as work-ability, whereas other risk factors (such as dyslipidaemia and heredity) were only available for subsets of patients. The results of the sensitivity analyses, however, suggest that adjustments for these potential confounders would only have limited, if any, impact on the risk estimates. Similarly, a quantitative estimation of confounding by smoking based on data from previous studies,4 ,34–39 argue against anything but a marginal potential for unmeasured confounding by smoking in our study. Further, the likelihood of achieving response to TNFi may be multifactorial. Because of the observational design, causality between TNFi-mediated response and ACS risk cannot be assumed, as the association could be influenced by other changes in pharmacotherapy (eg, corticosteroids or cyclo-oxygenase inhibitors) or lifestyle factors among responders. Alternatively, responsiveness as such could describe a phenotype with a lower CVD risk.
In summary, we found a reduction in the short-term risk of ACS in patients with RA and good response on TNFi treatment, indicating that the ability to reach optimal control of disease activity, rather than merely improved disease control, or TNFi per se, could potentially normalise the pattern of ischaemic heart disease in RA.
The authors would like to thank Jonas Eriksson and Thomas Frisell, Department of Medicine Solna, Clinical Epidemiology Unit, Karolinska Institutet, Stockholm, for help with linkage, data extraction and assistance with statistical analyses.
Handling editor Tore K Kvien
Collaborators The Anti-Rheumatic Therapy in Sweden (ARTIS) Study Group conducts scientific analyses using data from the Swedish Biologics Register. It also safeguards the quality and handling of the nationwide data collected. The following are the members of the ARTIS Study Group: Johan Askling, Lars Klareskog and Ronald van Vollenhoven (Karolinska Institutet, Stockholm, Sweden); Eva Baecklund (Uppsala University, Uppsala, Sweden); Alf Kastbom (Linköping University, Linköping, Sweden); Lennart Jacobsson (Sahlgrenska Academy, Gothenburg, Sweden); Carl Turesson and Elisabeth Lindqvist (Lund University, Malmö and Lund, Sweden); Solbritt Rantapää-Dahlqvist and Helena Forsblad-d'Elia (Umeå University, Umeå, Sweden); Nils Feltelius (Chairman of the Medical Products Agency, Sweden) and Sofia Ernestam (Register Holder, Swedish Rheumatology Quality Register, Sweden).
Contributors All authors have made substantial contributions to conception and design, acquisition of data or analysis and interpretation of data; drafting the article or revising it critically for important intellectual content and final approval of the version to be published.
Funding This study was supported by grants from the Swedish Research Council (K2013-52X-20307-07-3 and K2008-52X-20611-01-3), the Swedish Rheumatism Association, King Gustav V's 80-Year Foundation, the Västerbotten and Stockholm County Councils, the Swedish Heart-Lung Foundation, the Swedish Foundation for Strategic Research and the Swedish public-private COMBINE research consortium (http://www.combinesweden.se). The ARTIS Study Group conducts scientific analyses using data from the Swedish Biologics Register ARTIS run by the Swedish Society for Rheumatology. ARTIS has entered into agreements with Abbvie, BMS, MSD, Pfizer, Roche and UCB.
Competing interests None declared.
Ethics approval The Stockholm Ethics Review Board.
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.