Objective In contrast with the wealth of data on ischaemic heart disease in rheumatoid arthritis (RA), data on stroke are scarce and contradictory. Despite the high clinical and aetiological relevance, there is no data regarding when (if ever) after RA diagnosis there is an increased risk. Our objective was to assess the risk of stroke (by subtype) in contemporary patients with RA, particularly in relation to time since RA diagnosis.
Methods One incident RA cohort diagnosed between 1997 and 2009 (n=8077) and one nationwide prevalent RA cohort followed at Swedish rheumatology clinics between 2005 and 2009 ((n=39 065) were assembled). Each cohort member was matched to a general population comparator. Information on first-time hospitalisations for stroke up to 2009 was retrieved from the Swedish Patient Register. HR and 95% CI were estimated using Cox models.
Results In prevalent unselected RA, the HR of ischaemic stroke was 1.29 (95% CI 1.18 to 1.41). In the incident RA cohort, the overall risk increase was small and non-significant (overall HR 1.11, 95% CI 0.95 to 1.30). When stratified by RA disease duration, an increased risk of ischaemic stroke was indeed detectable but only after 10 or more years since RA diagnosis (HR>10 years: 2.33, 95% CI 1.25 to 4.34). Risk of haemorrhagic stroke was increased in prevalent but not in incident RA.
Conclusion The magnitude of stroke risk is lower than for ischaemic heart disease in RA, and the evolvement of this risk from RA diagnosis may be slower. This suggests different driving forces behind these two RA co-morbidities and has implications for the clinical follow-up of patients with RA.
Statistics from Altmetric.com
Patients with established rheumatoid arthritis (RA) are at well-documented increased risk of subclinical and clinical cardiovascular disease.1,–,4 It has recently been suggested that the progression of risk of myocardial infarction (MI) in RA may be more rapid than previously thought, and that the risk increase is already evident within a few years from RA diagnosis.4 ,5 The risk of clinical cerebrovascular events in RA has also been studied, but to a lesser extent than MI, and with varying and sometimes conflicting results (summarised in online supplementary table 1).3 ,6,–,14 In contrast with risks of MI in RA, there are little data on risks of cerebrovascular disease from cohorts of incident RA, and little data on when (if) in relation to RA diagnosis the risk of cerebrovascular disease is increased, data that would be of importance for our understanding of the driving forces behind the increased risk of cardiovascular co-morbidities in RA. Moreover, ischaemic stroke (IS) and haemorrhagic stroke (HS) have often been studied as a composite outcome.6 ,7 ,10,–,15 Given their different pathogeneses and risk-factor profiles,16 this might have concealed or diluted any increased risk specific to either subset. To our knowledge, only three studies have reported on risks specifically for IS,3 ,8 ,9 all showing increased risks of IS in RA (vs non-RA populations) with unknown or long-standing disease,8 ,9 and as defined from administrative3 or hospitalisation data.3 ,9
We therefore set out to assess the risk of stroke by subtype, and as a function of time since RA diagnosis, in two contemporary population-based cohorts: one nationwide cohort consisting of prevalent RA seen at Swedish rheumatology clinics between 2005 and 2009, and one incident cohort of patients with RA diagnosed between 1997 and 2009 in which we have recently demonstrated an increased risk of MI overall and shortly following RA diagnosis.5
We used cohort designs with incident and prevalent RA, respectively, as the exposures and subgroups of stroke (IS and HS) as the outcomes. The prevalent cohort should reflect the average level of (any) risk increase in patients with RA followed in routine rheumatology care; the incident cohort specifically allowed the assessment of risks as a function of time since onset of clinical RA.
In Sweden, patients with RA are treated by rheumatologists. Universal access to publicly funded healthcare, including inpatient care for all residents is available. Using the unique national identification number issued to all Swedish residents,18 data from national and virtually complete administrative or clinical registers on demographics, morbidity and mortality can be linked together. These registers also allow for unbiased identification of general population comparators and enable prospective assessments of morbidity and mortality independent of the status of RA disease.
Prevalent RA cohort
Using the outpatient component of the National Patient Register initiated in 2001, and including outpatient visits in non-primary care in Sweden, we identified a cohort of individuals with two or more visits to a rheumatology or internal medicine clinic listing RA (see online supplementary appendix) between January 2005 and December 2009 (n=39 065). This calendar period was chosen to provide us with a cohort of contemporary unselected and prevalent RA patients. Coverage of this part of the patient register for somatic care, regardless of specialisation, is approximately 80%. The missing 20% is mostly due to lack of data from private practitioners.19 In 2008, only 7% (17/232) of active rheumatologists in Sweden were private practitioners,20 indicating that the register coverage pertaining to RA is high. Diagnoses are reported using the International Classification of Diseases (ICD) version 10.21 An index date–the date of the second non-primary care outpatient visit listing an ICD-10 code of RA between 2005 and 2009 (regardless of number of visits before 2005)–was assigned to all individuals, and defined the start of follow-up.
Incident RA cohort
The Swedish Rheumatology Quality Register was initiated in 1995. It includes patients with RA who were aged 16 years or older, fulfilling the 1987 American College of Rheumatology criteria for RA. At diagnosis, patients are entered into the register with information on age, sex, rheumatoid factor (RF) status, date of first symptom of RA and date of diagnosis of RA. For this study, we identified all individuals diagnosed with RA between January 1997 and December 2009 within 12 months of first symptoms of RA (n=8077). The index date for the incident RA cohort was the date of diagnosis of RA.
For each patient in each of the two RA cohorts, we randomly selected five individuals from the Swedish Population Register (which includes all Swedish residents) matched on sex, year of birth and residential area. Each population-based comparator had to be alive at the index date of their corresponding RA patient, and were assigned the same index date as start of follow-up.
Data sources used for detecting outcomes during follow-up
Using the national registration number, we linked the two cohorts of patients with RA and their two matched cohorts of comparators to the following data sources, for which data were available up to 31 December 2009: The National Patient Register and the Swedish Population Register. The National Patient Register contains, in addition to the outpatient component described above, information on inpatient care since 1964, with nationwide 100% coverage since 1987.22 The register lists date of admission, date of discharge and the discharge diagnosis (primary and secondary diagnoses) as set by the discharging physician and classified according to the calendar-year specific ICD. The Swedish Population Register includes information on deaths, emigration and immigration for the entire Swedish population.23
By linking the cohorts of patients with RA and their two cohorts of matched comparators to these registers, we identified all hospitalisations listing stroke before and after the index date, all deaths and all emigrations from Sweden during follow-up.
Definition of outcome
Primary outcome for this study was first-time IS and HS after the index date (see online supplementary appendix for ICD codes used). In the main analyses, all individuals listed with an IS before the index date were excluded from the analysis of IS after the index date. Correspondingly, all individuals who were listed with a HS before the index date were excluded from the analysis of HS after the index date. As a sensitivity analysis, we also analysed stroke of unspecified type.
Validity of ICD codes indicating stroke
Stroke codes in the ICD-8 and ICD-9 have been validated and found to have a high validity24; the positive predictive value of a code indicating stroke was found to be 94% when using medical charts and current clinical criteria to validate register information. To assess whether this high validity applied in the context of RA, we conducted a validation study of the ICD codes indicating IS and HS. All charts of patients with RA, and controls in the Epidemiological Investigation of Rheumatoid Arthritis (EIRA) case-control study of incident RA who, after inclusion in EIRA, were hospitalised with an ICD-10 code indicating IS or HS (n=36 patients with RA and 40 controls) were reviewed. Detailed information on the EIRA study has been published previously.25 Using the MONICA criteria26 to classify events, we found that the ICD-10 codes, I61 and I63, both used as primary and secondary discharge diagnosis, had a positive predictive value of 92% among patients with RA, and 89% among non-RA controls.
To prevent detection of outcome being related to a visit at the rheumatologist for patients with RA, follow-up started 60 days after the index date in the prevalent cohort and its comparators. As a sensitivity analysis, we also started follow-up one year after the index date. In the incident cohort, follow-up started at the index date. In both cohorts, follow-up ended at the first-ever recorded IS or HS, death, first emigration, or 31 December 2009, whichever was earliest.
Crude rates were calculated by dividing the number of events during follow-up by the corresponding person-time at risk. All rates are presented as number of events per 1000 person-years.
In the main analyses, only individuals free from the outcome of interest before the index date were included, but rates without exclusions were also estimated. To compare the risk of IS or HS in patients with RA and in comparators, Cox models were used with time since the index date as time scale, stratified by birth year, sex and residential area, and adjusted for age at index date. In the incident cohort, we further assessed the role of RA duration by using time-dependent covariates. We also present stratum-specific hazard ratios (HR) for the calendar period of RA diagnosis, RF status, sex and age (quartiles).The proportional hazards assumption was tested by introducing an interaction term between the exposure (RA) and time since study entry. The study had more than 80% power to detect an HR for IS of 1.2, and 75% power to detect an HR for HS of 1.3 with a two-sided 5% Z-alpha. All analyses were performed using The SAS software package, V.9.2 (SAS Institute, Cary, North Carolina, USA).
Incidence and HR in prevalent RA
After excluding individuals with IS before the index date, 37 872 patients with RA and 167 388 comparators remained. Of those, 640 (1.7%) patients with RA and 2040 (1.2%) comparators were hospitalised with a first IS after the index date. The crude rates (95% CI) of IS were 5.8 (5.1 to 6.6) for RA and 4.1 (3.8 to 4.4) in the comparators, corresponding to an adjusted HR of 1.29 (95% CI 1.18 to 1.41). Analyses stratified by sex or RF status did not reveal any heterogeneity with respect to risk of IS. HR, 95% CI and person-time estimates based on the prevalent cohort and their comparators are presented in table 2.
After excluding individuals with HS before the index date, 38 786 patients with RA and 170 763 comparators remained. Of those, 133 (0.3%) patients with RA and 434 (0.2%) comparators were hospitalised with HS after the index date. The crude rates (95% CI) were 1.2 (0.8 to 1.5) for RA and 0.9 (0.7 to 1.0) for the comparators, corresponding to an adjusted HR of 1.32 (95% CI 1.09 to 1.60). Analyses stratified by RF status did not reveal any heterogeneity, but the risk of HS among women with RA increased 44% compared with the comparator women, but increased only 10% in men with RA when compared with the comparator men (table 2).
Adding a lag of 12 months before start of follow-up did not alter our findings (online supplementary table 2). Rates and HR of HS and IS in the prevalent cohort without excluding individuals with event prior to the index date, are found in the online supplementary table 3a.
Incidence and HR in incident RA, including HR by time since RA diagnosis
In the incident RA cohort, the crude rate (95% CI) of IS was 4.8 (3.3 to 6.3) in RA and 4.2 (3.6 to 4.9) in the comparators. The corresponding rate (95% CI) for HS was 0.8 (0.2 to 1.3) in RA and 1.0 (0.6 to 1.3) in the comparators. There was no significantly increased risk for IS or HS in patients with incident RA compared with their comparators (adjusted HR 1.11, 95% CI 0.95 to 1.30 for IS and 0.79, 95% CI 0.54 to 1.14 for HS). Subgroup analyses based on RF status did not reveal any significant heterogeneity. All HRs and 95% CI estimated based on the incident cohort and their comparators are found in table 3.
Rates and HR of HS and IS in the incident cohort without excluding individuals with event prior to index date, are found in the online supplementary table 3b.
When we assessed the time-dependent effect of duration of RA on the risk of IS and HS, we noted that the overall risk of IS was not increased in RA compared with comparators until after 10 years with RA (online supplementary table 5). Stratifying on calendar period of RA diagnosis and time since RA diagnosis did not reveal any pattern of generally increased risks for stroke among patients diagnosed with RA more than 10 years ago (data not shown). With respect to risk of HS, there were no indications that the duration of RA would modify the HRs (table 4). We could not detect any effect of calendar period of RA diagnosis (online supplementary table 4).
Rates and HRs of stroke of unspecified type are presented in online supplementary table 5.
Based on two of the world's largest cohorts of RA, this study suggests that contemporary unselected patients with RA are at a moderately increased risk of stroke. It also suggests that the risk of stroke in RA evolves more slowly, counting from RA diagnosis, than the risk of MI, and is detectable a decade after diagnosis of RA (although an increased risk earlier in the course of the disease cannot formally be ruled out). Apart from their clinical relevance, these findings are important for our understanding of the aetiologies of cardiovascular co-morbidities such as stroke and MI in RA.
Our study is the first to specifically report on risks, counting from RA diagnosis in contemporary population-based RA. Our finding of an increased risk in prevalent and unselected RA (HR=1.3 for both IS and HS) is somewhat lower, but reasonably in keeping with data from previous studies.3 ,8,–,10 At the same time, there are several studies indicating that there is no increased risk of stroke in RA.4 ,6 ,11 ,15 The more precise results from our cohorts may settle some of the uncertainty whether there is an increased risk of stroke in contemporary RA or not.
Our study has several strengths. By using contemporary nationwide population-based non-primary care visit data (rather than, eg, hospitalisation data) to identify the prevalent RA cohort, our cohort comprises virtually all patients with RA who were seen by a rheumatologist/internist during the study period, which was more recent (2005–2009) than in most previous studies. Although hospitalisation data may be used to identify RA, such designs carry an inherent risk of selection bias (of those patients with RA who need be hospitalised) and that hospitalisation (through pathways other than the RA disease) is related to the risk of stroke. Similarly, using the Swedish Rheumatology Quality Register allowed identification of a large clinical incident RA cohort of less than 12 months history of symptom at the time of RA diagnosis by a rheumatologist. The size of our study populations, the number of events during follow-up, and the duration of follow-up are strengths that when combined will provide our study with high power to detect even moderate differences in risks between RA and the general population. Our primary outcome has been validated and found to have a high positive predictive value, also in the context of RA (>90%, unpublished data). This reduces the risk of misclassification of outcome, and that the observed HRs would represent a dilution of the true HR. Finally, the capture of the outcome was independent of RA status, and similar for the RA and the general population cohort, which reduces the risk of differential misclassification of outcome and associated bias.
The Swedish Rheumatology Quality Register provided a large clinical cohort, but it may be that rheumatologists are less likely to include patients in a register who were intended for long-time follow-up, if the predicted survival of the patient is very short. This could lead to an underestimation of the true rates of IS/HS in the incident RA cohort, at least in the short term. Conversely, it may be that although we did not use hospitalisation data to identify the prevalent RA cohort, the two rheumatology outpatient visits used to identify the patients with RA were somehow related to factors that also lead to the stroke, and that our 60-day lag period was too short. However, use of a 1-year lag period did not materially alter the HRs (online supplementary table 2). We lacked information on traditional cardiovascular risk factors and, thus, the ability to assess their association with IS/HS in RA, which is a limitation of our study, although its absence does not affect the validity of our results, as our aim was to assess risks, not to attribute them to specific risk factors. Whereas the aim of the present study was to assess risks versus the general population, it will be an important task for future studies to assess risks in relation to exposures to specific treatment, and in relation to stroke phenotypes and risk factors.
In the general population, there are studies indicating that the risk factors for IS and HS are different.27,–,31 The discrepant risk factor profiles emphasise the need to study HS and IS as separate entities, and could explain why some of the reports using composite outcomes have failed to demonstrate a risk increase.11 ,15 Smoking is a well-established risk factor for RA.32 Smoking and inflammatory activity has also been demonstrated to increase levels of fibrinogen, a known risk factor for thromboembolic events, such as ischaemic stroke.33 It could be that factors related to hyper-coagulation and formation of thrombi/emboli trigger the increased risk of IS noticed in RA. There are studies suggesting that pro-coagulant factors are up-regulated in established RA,34 and it could be that this is part of the explanation for our observation that the risk of IS, in contrast with the rapid increase in risk of MI,5 was increased until 10 years after RA diagnosis. One of the strongest risk factors for IS in the general population is atrial fibrillation.31 This could perhaps be one of the explanations as to why the risk of IS is higher in RA than in the general population, but to be able to answer questions pertaining to the cause of IS in RA, studies aimed at attributing risk to risk factors must be performed. A recent meta-analysis of hypertension in RA failed to detect any difference in the prevalence of hypertension in RA and the general population,35 making it less likely that hypertension would be the primary cause of the vascular changes resulting in an increased risk of IS after 10 years of disease. Inflammation-induced accelerated atherosclerosis including increased arterial stiffness, a well-documented finding in incident cohorts of RA36 might, on the other hand, explain the somewhat slower evolvement of stroke (vs MI) risks observed in our cohort.17 In the absence of detailed clinical data pertaining to vessel function in our cohort, this will remain a speculation.
In conclusion, contemporary patients with RA are at increased risk of stroke, which is of a lesser magnitude than the typically reported risk of MI. In contrast with the rapid increase in risk of MI that we have reported previously in the incident RA cohort,5 the evolvement of risk of stroke seems slower, suggesting different driving forces behind these two RA co-morbidities, and a different need for clinical follow-up pertaining to risk and primary prevention measures.
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:
- Web Only 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.
- Web Only 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.
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.