Authors: Yumeko Kawano (1Brigham and Women’s Hospital, Boston, MA;; 2Harvard Medical School, Boston, MA;), Brittany N. Weber (1Brigham and Women’s Hospital, Boston, MA;; 2Harvard Medical School, Boston, MA;), Dana Weisenfeld (1Brigham and Women’s Hospital, Boston, MA;), Mary I. Jeffway (1Brigham and Women’s Hospital, Boston, MA;), Tianrun Cai (1Brigham and Women’s Hospital, Boston, MA;), Gregory C. McDermott (1Brigham and Women’s Hospital, Boston, MA;; 2Harvard Medical School, Boston, MA;), Qing Liu (1Brigham and Women’s Hospital, Boston, MA;), Jeffrey A. Sparks (1Brigham and Women’s Hospital, Boston, MA;; 2Harvard Medical School, Boston, MA;), Jennifer Stuart (1Brigham and Women’s Hospital, Boston, MA;; 3Harvard T. H. Chan School of Public Health, Boston, MA;), Jacob Joseph (4Veteran’s Affairs Healthcare System, Boston, MA;; 5Veterans Affairs Healthcare System, Providence, RI;; 6Brown University, Providence, RI, USA), Tianxi Cai (3Harvard T. H. Chan School of Public Health, Boston, MA;; 4Veteran’s Affairs Healthcare System, Boston, MA;), Katherine P. Liao (1Brigham and Women’s Hospital, Boston, MA;; 2Harvard Medical School, Boston, MA;; 4Veteran’s Affairs Healthcare System, Boston, MA;)
Categories: Article, Rheumatoid arthritis, heart failure, cardiovascular outcomes, inflammation
Source: Arthritis care & research
Doi: 10.1002/acr.25481
Authors: Yumeko Kawano, Brittany N. Weber, Dana Weisenfeld, Mary I. Jeffway, Tianrun Cai, Gregory C. McDermott, Qing Liu, Jeffrey A. Sparks, Jennifer Stuart, Jacob Joseph, Tianxi Cai, Katherine P. Liao
Patients with rheumatoid arthritis (RA) are at increased risk of cardiovascular disease (CVD) including heart failure (HF). However, little is known regarding the relative risks of heart failure subtypes such as HF with preserved (HFpEF) or reduced ejection fraction (HFrEF) in RA compared to non-RA.
We identified RA patients and matched non-RA comparators among participants consenting to broad research from two large academic centers. We identified incident HF and categorized HF subtypes based on ejection fraction (EF) closest to the HF incident date. Covariates included age, sex, and established CVD risk factors. Cox proportional hazards models were used to estimate the hazard ratios (HR) for incident HF and HF subtypes.
We studied 1445 RA patients and 4335 matched non-RA comparators (mean age 51.4 and 51.7 years, 78.7% female). HFpEF was the most common HF subtype in both groups (65% in RA vs. 59% in non-RA). RA patients had a HR of 1.79 (95% CI: 1.38–2.32) for incident HF compared to those without RA after adjusting for CVD risk factors. RA patients had a higher rate of HFpEF (HR 1.99, 95% CI: 1.43 – 2.77), but there was no statistical difference in HFrEF rate (HR 1.45, 95% CI 0.81 – 2.60).
RA was associated with higher rate of HF overall compared to non-RA, even after adjustment for established CVD risk factors. The elevated risk was driven by HFpEF, supporting a role for inflammation in HFpEF and highlighting potential opportunities to address this excess risk in RA.
Patients with rheumatoid arthritis (RA) are at increased risk of cardiovascular diseases (CVD) such as myocardial infarction and stroke compared to the general population^1, 2^. Compared to ischemic heart disease, heart failure (HF) has been relatively understudied in RA despite significant contributions to morbidity and mortality^3, 4^. Some prior studies have shown a heightened risk for HF in patients with RA compared to non-RA^5, 6^, but HF was generally studied as a single entity and the differential risk of HF subtypes were limited due to the need for large RA cohorts with detailed clinical documentation of phenotypic data (i.e. echocardiograms, cardiology notes) required for such studies.
HF is a clinical syndrome characterized by symptoms, e.g., dyspnea, lower extremity edema and evidence of congestion that result from structural or functional abnormalities of the heart^7–9^; this definition encompasses both HF with reduced ejection fraction (HFrEF), defined as EF ≤ 40%, as well as preserved ejection fraction (HFpEF) with EF ≥ 50%. Despite some overlap, there is increasing recognition of the different pathogenesis as well as treatment strategies of HFrEF compared to HFpEF^10, 11^. While HFrEF is typically caused by myocardial ischemic injury, HFpEF is thought to be associated with pro-inflammatory metabolic conditions including obesity, diabetes, and potentially systemic inflammatory conditions such as RA^12–14^. Abnormal coronary microvascular dysfunction due to endothelial inflammation has been observed in HFpEF and may be one of the mechanisms by which HFpEF develops in RA^15^. However, despite rising incidence in the general population, HFpEF remains relatively underdiagnosed in part due to its more subtle preclinical stages^11^. Understanding the risk of HF and specific HF subtypes in RA has the potential to improve screening, management, and ultimately cardiovascular outcomes for patients with RA.
The objective of our study was to investigate the risk of HF and HF subtypes, particularly HFpEF, among patients with RA compared to those without RA. As a prototypical inflammatory disease, we hypothesized that RA will be associated with increased risk of HF overall as observed in prior studies, and specifically with HFpEF. We also examined the differences in risk factors for HFpEF in RA compared to non-RA as an exploratory analysis.
We conducted a retrospective cohort study using data from the Mass General Brigham (MGB) Biobank, a cohort of subjects recruited from two large academic care centers who consented to broad-based research. The MGB biobank is linked with electronic health record (EHR) data. The EHR data include both structured data such as International Classification of Diseases (ICD) codes and laboratory values, as well as unstructured data including clinical notes and imaging reports. The narrative data were extracted using natural language processing (NLP) with the Narrative Information Linear Extraction (NILE) package^16^.
Patients with RA were classified using a previously validated RA algorithm that combines ICD codes and RA-related NLP concepts (terms related to the phenotype of interest) with a positive predictive value (PPV) of 91%^17^. The index date for each patient with RA was the date of the first RA ICD code if followed by another RA ICD code within 6 months, or the first RA NLP concept, whichever was earlier. Compared with a gold standard set of RA patients for whom the diagnosis date was confirmed with manual medical record review, this rule for the RA diagnosis date performed best with an intraclass correlation coefficient of 0.88. We also required at least 90 days between the date of first encounter in the EHR and RA diagnosis date (Figure 1) to identify patients with incident RA.
Each patient with RA was matched to three comparators without RA enrolled in the MGB Biobank. The comparators were matched based on sex, the year of birth (within the same year), and year of entry into the EHR (± 2-year window). The index date for each matched comparator was defined as an encounter date closest to the matched RA patient’s index date. Patients who were less than 18 years of age at the index date and those with less than 90 days of EHR history following the index date were excluded, as were patients with prevalent HF diagnoses prior to and up to 90 days following the index date (Figure 1).
Incident HF was ascertained using a previously validated algorithm (PPV 0.90) that incorporates a combination of structured and unstructured data from the EHR; the final algorithm included a weighted equation of the number of HF ICD codes, HF NLP, furosemide NLP, and total ICD counts^18^. Incident HF date was defined as the later date between the first HF ICD and the first furosemide NLP dates, as described in our prior study^18^.
We used the HF definition and classification criteria adopted by the American Heart Association, American College of Cardiology, Heart Failure Society of America, and the European Society of Cardiology^9,10^. HFpEF was defined as HF with an EF ≥ 50%, HFrEF as EF ≤ 40%, and EF values in the 41–49% range were classified as HF with moderately reduced EF (HFmrEF)^7^ (Figure 1). HF subtypes were categorized using EXTEND^19^, an NLP tool that extracts numerical data (i.e. EF) from clinical notes and cardiology reports (including echocardiograms, cardiac MRI, and cardiac PET scans). The EF value closest to the HF incident date was used in the analysis. Those with HFmrEF were included in the overall HF outcome but not specifically studied as a separate outcome due to small numbers.
Follow-up started from 90 days after the index date to the earliest of incident HF (any type) to reduce the potential for prevalent HF at the index date. Patients were followed until their last encounter in the EHR, death, end of study (October 26, 2021), or 15 years, whichever occurred earliest (Figure 2).
We extracted demographic data and traditional cardiovascular risk factors. These included age at index date, sex, self-reported race, body mass index (BMI), smoking status, history of coronary artery disease, atrial fibrillation, hypertension, hyperlipidemia, diabetes, stroke, and chronic kidney disease at index date for RA and non-RA comparators. Smoking status was classified as ever smoker vs. never smoker based on positive mentions of smoking NLP concepts at or before the index date. Other comorbidities were ascertained based on ICD-9 and 10 codes as used in prior studies^6, 18, 20, 21^ (Supplementary Table S1).
For the RA cohort, we also extracted RA-specific variables including seropositivity (positive rheumatoid factor (RF) and/or antibodies to cyclic citrullinated peptide (anti-CCP) or NLP for seropositivity) at any point during the follow-up period. We also determined the use of corticosteroids and disease-modifying antirheumatic drugs (DMARDs) at baseline (within 90 days following the index date) (Figure 2). Post-baseline covariates such as subsequent DMARD choice, nonsteroidal anti-inflammatory medication use, and disease activity were not collected but may mediate the associations with outcomes.
The exposure of interest was RA vs. non-RA status. Summary statistics for baseline characteristics of the cohort were reported as mean (standard deviation [SD]), median (interquartile range [IQR]), or count (percentage), as appropriate; between-group comparisons were made using t-tests or chi-squared tests.
Incident rates for HF overall and specific HF subtypes (HFpEF and HFrEF) were reported per 1,000 person-years. Time to incident HFpEF and HFrEF outcomes were assessed and displayed with cumulative incidence curves treating death as a competing event. We constructed unadjusted and adjusted multivariable Cox proportional hazards model to estimate the hazard ratios (HR) for HF overall, HFpEF, and HFrEF. The models were adjusted by covariates selected a priori and included age, sex, and traditional CVD risk factors associated with HF: history of coronary artery disease, atrial fibrillation, hypertension, hyperlipidemia, diabetes, stroke, chronic kidney disease, and BMI. We used a cause-specific hazards model to examine the HF subtype outcomes (HFpEF and HFrEF), censoring patients who developed other HF subtypes at their HF diagnosis date. As an exploratory analysis, we stratified by RA status to enable a comparison of important clinical risk factors for HF and HFpEF among patients with and without RA.
We performed several sensitivity analyses. We examined a more stringent incident RA cohort using a longer run-in period of at least one year between the EHR entry date and the first RA code. We also used a Fine-Gray sub-distribution hazard model as an alternate model to account for competing risk for the different HF subtypes. All analyses were performed using R version 4.2.1 (http://www.r-project.org/). This study was approved by the MGB institutional review board.
Among 1963 patients who had incident RA, 1445 met our inclusion and exclusion criteria (Figure 1), and they were matched in a 3 ratio to 4335 comparators without RA. Baseline characteristics are shown in Table 1. The mean age was 51.4 years (13.5) for RA and 51.7 years (13.4) for non-RA cohorts, and 78.7% of both cohorts were female. Baseline comorbidities were similar between the two cohorts, except for coronary artery disease which was more common in RA (10.9% in RA vs. 8.5% in non-RA). The body mass index (BMI) was also slightly higher in RA (28.7 kg/m^2^ [6.7] vs. 28.1 [6.9]), and RA patients were more likely to have ever smoked (30.9% vs. 27.3%).
Over a combined 57,445 person-years of follow-up (median 10.3 years per patient), we identified 92 incident HF cases in RA and 157 incident HF in the non-RA cohorts (Table 2). HFpEF was the predominant HF subtype in both cohorts, accounting for 65.2% and 58.6% of all HF cases in RA and non-RA cohorts, respectively. The overall HF incidence rate was higher in RA than non-RA comparators (6.63 vs. 3.60 per 1,000 person-years). The incidence rate for HFpEF was higher in RA compared to non-RA (4.33 vs. 2.11 per 1,000 person-years). The incidence rate for HFrEF did not differ significantly between the two groups (1.23 vs. 0.80 per 1,000 person-years), though the total number of HFrEF events were small. 31 RA patients and 138 non-RA patients died during follow up. Cumulative incidence curves for HFpEF and HFrEF among patients with RA and non-RA comparators, adjusting for the competing risk of death, are shown in Figure 3.
In the unadjusted Cox proportional hazards model, RA status was associated with increased hazard of HF overall (HR 1.84, 95% confidence interval [CI] 1.42–2.38) and HFpEF (HR 2.05, 95% CI 1.48–2.84), but not HFrEF (HR 1.54, 95% CI 0.86–2.74) (Table 2). In the multivariable model adjusting for age, sex, and traditional CVD risk factors (history of coronary artery disease, atrial fibrillation, hypertension, hyperlipidemia, diabetes, stroke, chronic kidney disease, and BMI), RA status was significantly associated with increased risk of HF overall (HR 1.79, 95% CI 1.38–2.32). Examining the HF subtypes, RA status was significantly associated with HFpEF (HR 1.99, 95% CI 1.43–2.77) but not HFrEF (HR 1.45, 95% CI 0.81–2.60).
In the exploratory analysis stratifying by RA status to compare clinical risk factors for HF and the HF subtypes among RA and non-RA comparators, we found that traditional CVD risk factors such as older age, BMI, and coronary artery disease (CAD) history increase the hazard for HF overall in both groups (Table 3). While female sex was associated with lower risk of HF overall in non-RA, it did not have as strong a protective effect among RA patients. With regards to incident HFpEF, older age and higher BMI increased risk for HFpEF in both groups; diabetes and stroke were also associated with increased risk of HFpEF among RA patients but not among non-RA comparators.
In the sensitivity analysis with a more stringent criteria for incident RA requiring at least 1 year of EHR history before the RA diagnosis date/index date, we identified 1329 RA patients matched to 3987 non-RA comparators. Similar to the primary analysis, we found that the incidence rate of HF overall was higher in RA compared to non-RA (6.63 vs. 3.92 per 1000 person-years), as was the incidence rate for HFpEF (4.35 vs. 2.69 per 1000 person-years). In multivariable Cox proportional hazards models (adjusting for the same covariates as in the primary analysis), we found similar results to the primary analysis, with higher rate of HF overall (HR 1.66, 95% CI 1.27–2.17) and HFpEF (HR 1.59, 95% CI 1.14–2.21) but not HFrEF (HR 1.47, 95% CI 0.77–2.81) in RA compared to non-RA (Supplementary Table S2).
Using a Fine-Gray subdistribution hazard model of HFpEF (treating other HF subtypes as competing risk), the results were similar to the primary cause-specific model. RA patients had higher hazard of HFpEF (aHR 1.93, 95% CI 1.38–2.68) but not HFrEF (HR 1.37, 95% CI 0.76–2.49) (Supplementary Table S3).
In this study investigating the association of RA and HF with a focus on HF subtypes, RA patients were at increased risk for HF overall and HFpEF compared to matched non-RA comparators, after adjusting for known risk factors related to cardiovascular disease. RA was not found to be an independent risk factor for HFrEF, though our findings regarding HFrEF are limited by small numbers. This study builds upon prior literature demonstrating RA as a risk factor for HF and provides further detail that the signal is largely driven by risk for HFpEF.
Our findings of an estimated 79% increased risk for HF in RA compared to non-RA are in line with prior epidemiologic studies^22^. In one large population-based study conducted in the pre-biologic era, the risk for HF in RA was estimated to be as high as HR of 1.87 (95% CI 1.47–2.39)^5^. In a more contemporary cohort, the HR was 1.21 (95%CI 1.03–1.42)^6^. Furthermore, in the present study, the increased risk for HF in RA was driven by HFpEF rather than HFrEF. Few prior studies have examined HF subtypes specifically, likely due to the difficulty in identifying a sufficiently sized RA cohort with EF data in a format that can be used for large cohort analyses. Several studies have observed that RA patients with HF tended to have higher EFs and evidence of diastolic dysfunction (one of the features of HFpEF) compared to non-RA HF patients^4, 23^, which aligns with our findings. In one prior study, RA patients were noted to have a heightened risk of nonischemic HF (defined by the absence of ICD codes for ischemic heart disease) compared to ischemic HF in the year following their RA diagnosis^24^; although EF values were not available in their study, ischemic heart disease is typically associated with HFrEF. HF tended to have higher EFs and evidence of diastolic dysfunction (one of the features of HFpEF) compared to non-RA HF patients^4, 23^, which aligns with our findings. In one prior study, RA patients were noted to have a heightened risk of nonischemic HF (defined by the absence of ICD codes of ischemic heart disease) in the year following their RA diagnosis compared to ischemic HF^24^; though EFs were not available in their study for HF subtyping, ischemic heart disease is typically associated with HFrEF. HF tended to have higher EFs and evidence of diastolic dysfunction (one of the features of HFpEF) compared to non-RA HF patients^4, 23^, which aligns with our findings. In one prior study, RA patients were noted to have a heightened risk of nonischemic HF (defined by the absence of ICD codes of ischemic heart disease) in the year following their RA diagnosis compared to ischemic HF^24^; though EFs were not available in their study for HF subtyping, ischemic heart disease is typically associated with HFrEF. HF tend to have higher EFs and evidence of diastolic dysfunction (one of the features of HFpEF) compared to non-RA HF patients^4, 23^, which aligns with our findings. In one prior study, RA patients were noted to have a heightened risk of nonischemic HF (defined by the absence of ICD codes of ischemic heart disease) in the year following their RA diagnosis compared to ischemic HF^24^; though EFs were not available in their study for HF subtyping, ischemic heart disease is typically associated with HFrEF. A recent Swedish registry-based study found that RA was more strongly associated with HF with EF ≥40% than EF <40% (odds ratio [OR] 1.7, 95% CI 1.4–2.0 vs. OR 1.5, 95% CI 1.2–1.8)^25^; however, this association was driven more by HFmrEF (EF 40–49%), rather than by HFpEF (EF ≥50%). Our study corroborates prior studies where no increased risk for HFrEF was observed among the RA cases, however the lack of signal in our study may also be due to a relatively small number of individuals with RA who developed an event.
HFpEF was the most common subtype of HF in our study, accounting for 65.2% of HF cases among RA patients and 58.6% in the non-RA comparators. Prior studies have similarly noted that HFpEF was the predominant subtype among RA patients and non-RA comparators^6, 23^; this may be a reflection of the rising incidence of HFpEF in the general population^11, 26^, but nevertheless points to the importance of recognizing the HFpEF subtype in RA patients. Some of the clinical risk factors for HFpEF in the general population include older age, higher BMI, atrial fibrillation, and diabetes mellitus^27, 28^. Obesity and insulin resistance were particularly noted to be associated with HFpEF among women^29^. In the few studies investigating risk factors for HFpEF in the RA population, high levels of inflammatory markers (erythrocyte sedimentation rate and C-reactive protein) and high disease activity (Disease Activity Score-28) were noted to be associated with HFpEF^6, 18, 23, 24^. HFpEF was the most common subtype of HF in our study, accounting for 66.3% of HF cases among RA patients and 56.4% in the non-RA comparators. Prior studies have similarly noted that HFpEF was the predominant subtype among RA patients as well as among non-RA comparators^6, 23^; this may be a reflection of the rising incidence of HFpEF in the general population^11, 26^, but nevertheless points to the importance of recognizing the HFpEF subtype in RA patients. Some of the clinical risk factors for HFpEF in the general population include older age, higher BMI, atrial fibrillation, and diabetes mellitus^27, 28^. Obesity and insulin resistance were particularly noted to be associated with HFpEF among women^29^. In the few studies investigating risk factors for HFpEF in the RA population, high levels of inflammatory markers (erythrocyte sedimentation rate and C-reactive protein) and high disease activity (Disease Activity Score-28) were noted to be associated with HFpEF^6, 18, 23, 24^. Our study builds on these studies and suggests that known cardiometabolic risk factors for HF such as diabetes and obesity represented by higher BMI conferred similar risk in RA and non-RA. BMI however has a complex relationship with RA and inflammation, and may have a mediating role between RA and CV disease. Furthermore, BMI has been found to have divergent effects on inflammatory markers based on sex in patients with RA^30^, and whether or not these sex differences translate into different CV risk profiles for men and women with RA requires further study.
Over the past decade, there has been increasing awareness of different phenotypes of HFpEF beyond diastolic dysfunction related to systemic hypertension. Particularly relevant to the rheumatology patient population is the inflammatory-metabolic phenotype, wherein systemic inflammatory states (such as obesity, diabetes, and systemic inflammatory diseases) are thought to induce coronary microvascular dysfunction (CMD)^13, 15, 31^. Specifically, proinflammatory cytokines including tumor necrosis factor α (TNFα), TNFα Receptor 1, Interleukin (IL)-6, IL-1, and others reduce endothelial production of nitric oxide (NO), resulting in downstream effects that increase cardiomyocyte stiffening, hypertrophy, and myocardial fibrosis^32, 33^ and the clinical syndrome of HFpEF. Indeed, CMD as measured by impaired coronary flow reserve on cardiac positron tomography (PET) scans have been observed in RA patients without clinical cardiovascular disease, especially in those with high IL-6 levels^34, 35^. Whether the CMD portends future HF risk (particularly HFpEF) and whether targeted anti-inflammatory therapies mitigate HFpEF risk in this population require further study. Nevertheless, minimizing comorbidities such as diabetes and obesity and controlling systemic inflammation in RA undoubtedly contribute to lowering the high CV risk burden in this population and should be pursued aggressively^36^.
While HFpEF was once considered to have limited treatment options compared to HFrEF, the paradigm shifted after a landmark study in 2021 found that empagliflozin, a sodium-glucose cotransporter 2 inhibitor, reduced the risk of cardiovascular death and HF hospitalizations among patients with HFpEF^37^. Moreover, there has been increasing interest in the use of anti-inflammatory therapies for treatment of HFpEF. An analysis of the CANTOS trial showed a dose-dependent decrease in HF hospitalizations with the use of the IL-1 β monoclonal antibody canakinumab^38^. Following promising recent studies showing the association of IL-6 levels with incident HFpEF in the general (non-rheumatologic) population^39, 40^, a large clinical trial is currently underway to study a novel IL-6 ligand monoclonal antibody (ziltivekimab) for the treatment of HFpEF and HFmrEF in the general cardiology population^41^. In the future, such studies could inform treatment of RA patients with HFpEF or cardiometabolic risk factors for HFpEF, and highlight the importance of screening for HFpEF in this population.
Our study has several limitations. The study was conducted at an academic tertiary hospital system among participants in a Biobank which may introduce selection bias as well as limitations in diversity and geography. Thus, the findings may not be generalizable to the general population. The study also did not account for post-baseline variables that may mediate associations such as chronic NSAID or steroid use, or specific DMARD use. Disease activity was also unavailable but likely mediates the association with HF outcomes. In addition, while the HF algorithm and the NLP method for extracting EF data were previously validated^18, 19^ and allowed us to effectively utilize EHR data, we rely on the availability of clinically-performed cardiology studies for HF subtyping. Therefore, if patients received cardiology care or had cardiac imaging performed outside of the MGB system, we would not be able to ascertain their HF status or subtype. Additionally, while we sought to identify incident RA patients by requiring 90 days of EHR history prior to any RA ICD code or NLP concept for RA, it is possible that there are patients whose RA diagnosis dates preceded the EHR entry date. We performed a sensitivity analysis using a more stringent criteria for incident RA, and the results were similar to the primary analysis. In our exploratory analysis comparing different risk factors for HF among RA and non-RA, we were underpowered to study HFrEF risk due to the small number of events. Finally, CV comorbidities were ascertained using the common approach of using ICD codes, however the PPVs can range from 71–95%^6, 18, 20, 21^.
In summary, we observed that RA was an independent risk factor for incident HF, specifically HFpEF and not HFrEF, after adjusting for traditional CV risk factors associated with HF. While the clinical risk factors for HFpEF did not vary substantially between RA and non-RA comparators, older age, higher BMI, and diabetes were associated with greater HFpEF risk in RA. RA can be considered a human model for inflammation, and findings from this study support the notion that chronic inflammation increases risk for HFpEF. Since inflammation is modifiable with anti-inflammatory medications, further studies are needed to determine whether anti-inflammatory therapies have the potential to reduce risk of HFpEF in RA and other individuals with chronic inflammation.