Abstract
Background
Immune regulation and inflammation play a role in the pathogenesis and progression of acute and chronic heart failure (HF). Although the clinical course of acute, severe inflammatory cardiomyopathy is well-described, the effects of chronic systemic inflammation on cardiovascular function over time are less clear. To investigate this question, we compared trajectories over time in left ventricular ejection fraction (LVEF) for HF patients with different chronic inflammatory diseases (CIDs): human immunodeficiency virus (HIV), systemic lupus erythematosus (SLE), systemic sclerosis (SSc), rheumatoid arthritis (RA), inflammatory bowel disease (IBD), and/or psoriasis.
Methods
Using a database of patients receiving care in a large metropolitan healthcare system since January 1, 2000, we analyzed serial, clinically indicated echocardiograms from HF patients with CIDs and frequency-matched HF patients without CIDs. We included patients with ≥3 serial echocardiograms (N=974; median 6.1 years between first and most recent echo). We assessed LVEF trajectories over time using latent trajectory models, then investigated differences in LVEF trajectories for specific CID subtypes compared with controls.
Results
Overall, the majority of patients studied (N=687; 70.5%) had LVEF trajectories consistent with heart failure with preserved or midrange EF (HFpEF/HFmrEF), whereas 255 (26.2%) had HF with reduced EF (HFrEF) and 32 (3.3%) had HF with recovered EF (HFrecEF). Compared with non-CID controls with HF, patients with RA, IBD, and SLE were significantly more likely than controls to have HFpEF/HFmrEF whereas patients with HIV were significantly more likely to have HFrEF.
Conclusions
Among HF patients with CIDs, distinct LVEF trajectory patterns associate with different specific individual CIDs. This highlights the heterogeneity of HF subtypes and changes over time across different CIDs.
Keywords: heart failure, chronic inflammatory diseases, autoimmune diseases, electronic cohort, latent trajectory analysis
Introduction
Immune regulation and inflammation play important roles in the pathogenesis and progression of heart failure (HF).1 In this context, human “models” of the immune dysregulation-inflammation-HF axis may provide clinically relevant insights. Whereas studies of extreme outcome-related phenotypes – e.g., infectious and inflammatory myocarditis – delineate acute, severe presentations of inflammation-mediated HF,2 investigation of clinical and subclinical myocardial dysfunction in chronic inflammatory diseases (CIDs) may provide insights into subacute and chronic manifestations of inflammation-mediated HF.
Recently, we observed significant differences in risk for incident heart failure across six distinct CIDs with unique pathophysiologies: inflammatory bowel disease (IBD), psoriasis, rheumatoid arthritis (RA), systemic sclerosis (SSc), systemic lupus erythematosus (SLE), and human immunodeficiency virus (HIV).3 People with SSc and SLE had the highest risk for HF after multivariable adjustment and there was heterogeneity in the predominant HF phenotype for each CID: the vast majority of HF among people with SSc was comprised of HF with preserved ejection fraction (HFpEF), whereas HF phenotypes for SLE patients were half HFpEF and half HF with reduced ejection fraction (HFrEF).
While these data provide insight into HF risk and presenting phenotype in different CIDs, little is known regarding trajectories over time of cardiac structural and functional parameters in patients with CIDs. Given the dynamic nature of inflammation and its overt clinical correlates in HF stabilization/improvement vs. worsening,1,2 as well as the dynamic nature of cardiac substrates in HF, the dearth of longitudinal data on cardiac substrates in CIDs represents a significant scientific limitation. Therefore, in this study, we leveraged a unique cohort comprised of several distinct CID populations and controls to evaluate the long-term trajectories of left ventricular ejection fraction (LVEF) in patients with CIDs and HF.
Methods
Overview.
We analyzed longitudinal echocardiographic data from a cohort of patients with CIDs and controls derived from electronic health records at Northwestern Medicine health system to test the hypothesis that changes or trajectories of LVEF over time differ across the CID types. The cohort creation and research protocol were approved by the institutional review board at Northwestern University. Because of the sensitive nature of the data analyzed for this study, qualified researchers trained in human subject confidentiality protocols may coordinate with the corresponding author to request access to the dataset and may need review by the Northwestern University Institutional Review Board.
Cohort.
The larger cohort includes individuals who were diagnosed with one of the following CIDs from January 1, 2000 to January 1, 2019: HIV, IBD, psoriasis, RA, SLE, and SSc; and controls without CIDs frequency-matched on age, sex, race, and baseline diabetes and hypertension status. Full details of this cohort have been previously described.3 The cohort creation and research protocol were approved by the institutional review board at Northwestern University (Chicago, IL, USA). A waiver of informed consent was applied.
For this nested analysis within the larger CID cohort, we included patients diagnosed with HF either at baseline (earliest visit with CID diagnosis for those with CID; earliest visit for controls) or during follow-up who had LVEF measurements from echocardiography performed on at least three different dates during follow-up. A minimum of three LVEF was selected because preliminary analysis showed that LVEF followed a quadratic trend, and a quadratic line needed at least three points to fit. HF diagnosis was based on having at least one previously validated inpatient or outpatient ICD 9/10 codes for HF3–6 (ICD9: 398.91, 401.01, 402.01, 402.11, 402.91, 404.03, 404.11, 404.13, 404.91, 404.93, 428.xx; ICD10: I09.81, I11.0, I13.0, I13.2, I50) at first visit in the electronic health record (EHR) (baseline) or on follow-up. We also set time zero as the date of heart failure diagnosis. We intentionally restricted our analyses to patients with HF to enhance internal validity, maintain the clinically relevant focus on HF phenotypes, and limit confounding by indication for serial echocardiographic examinations.
Statistical Analyses: Mixed Effects Models.
To inform the primary (latent trajectory) analyses, we first used mixed effects models to compare average LVEF trajectories across CIDs. These are preferred over other models of repeated measures (e.g., general estimating equations) since they are more flexible in handling differences in the total number and timing of echocardiograms per person in our cohort, which given its clinical nature had heterogeneous timing between echocardiograms. To test the hypothesis that different CIDs have different average rates of change in LVEF over time, we tested for interaction between time and CID type. Our base model also adjusted for matching variables age, sex, race, insurance status, hypertension (HTN) status, diabetes mellitus (DM) status, and year of entry into the database. We then used a stepwise procedure to develop the final model that assessed if including smoking status, statin use, and diagnosis of coronary artery disease (CAD) prior to HF, and timing of HF diagnosis (present at entry vs occurred during follow-up) improves model fit (assessed by checking if the new model had lower Akaike Information Criterion). Diagnosis of DM, HTN, and CAD were based on lab findings, medication use, and/or validated ICD codes as described in the original cohort (Supplemental Table I).3 Since LVEF is a percentage bound between 0 to 100%, we used a beta distribution.
Statistical Analyses: Latent Class Trajectory Models.
In mixed effects models, the underlying assumption is that HF patients belong to one homogenous group and the model estimates a single average LVEF change over time per CID type. However, this assumption may not be appropriate for HF patients given phenotypic heterogeneity and heterogeneity over time in LVEF change. Indeed, patients with reduced (<40%), mid-range (40% to 50%), or preserved (>50%) LVEF are already distinguished at time of diagnosis.7–11 To distinguish multiple longitudinal LVEF trajectories, we used latent trajectory analysis (LTA).
We first tested if the homogeneity assumption held true by testing the hypothesis: HF patients can be classified into more than one latent group based on the trajectory of LVEF over time. If this hypothesis held, we then planned to test whether CID type was associated with a latent group using logistic regression models.
Following guidelines and studies using LTA12–14, we performed modeling in stepwise manner: (1) assess need for polynomial terms for time (linear vs quadratic); (2) assess distribution; (3) assess need for random intercept terms; and (4) choose the optimal number of groups. For the 1st to 3rd step, we first assessed if the models converged and compared which of the two had a lower Bayesian information criterion score (BIC). The suggested specification was then carried over into the next step. For example, if we found that the linear specification had a lower BIC compared to the quadratic one, both models in step 2 would be linear. For the 4th step that involves determining the optimal group number, we considered a combination of BIC and group size, with the goal of having each group comprise at least 5% of the cohort and represent clinically distinct trajectories. To aid interpretation and assignment of a descriptive group label, we plotted observed LVEF trajectories for each group and included a smoothing line.
After determining the most optimal LTA model, we assigned each patient to a latent trajectory group using the highest post-probability method.14 Multivariable (logistic or multinomial) models were then used to test if CID remained a significant predictor of trajectory group after adjusting for frequency matching variables (age, sex, race, diabetes mellitus and hypertension at first visit, year of first visit) and smoking status in first visit, statin use at first visit, timing of HF diagnosis (at baseline or diagnosed on follow-up), and diagnosis of CAD prior to HF. The same stepwise fitting approach used for the mixed model above was used. Individuals with missing baseline variables (which consisted of <5% of eligible participants) were excluded from this analysis. For sensitivity analysis, we repeated the LTA including everyone that had ≥2 LVEF.
All analyses were run using R 4.0. All authors had full access to all the data in the study and takes responsibility for its integrity and the data analysis. Details of the modeling including sample computer code for the analysis are available as a supplemental file (Supplemental File).
Results
Sample characteristics
There were 2,046 patients with HF diagnoses out of 37,636 in the overall CID and non-CID controls cohort (18,278 with CIDs, 19,358 without CIDs). From this group, 974 patients had at least three LVEF measurements and were included in the analysis; 98.9% (963) of these patients had >2 months between their first and last echocardiogram during follow-up and the median time from first to last echocardiogram in the 974 patients with HF was 6.1 years. Compared to HF patients who did not have at least 3 LVEF measurements (and were therefore excluded from the analysis), the included sample was younger (mean age of 60.8 years (SD: 14.3) vs. 66.4 years (SD: 14.8)), comprised of a lower proportion of non-Hispanic whites (50.8% vs 66.0%), had lower mean LVEF at diagnosis (48.8 % (SD: 16.2) vs 55.1% (SD: 13.6), had a higher median number of cardiovascular admissions after heart failure diagnosis (7.0 [IQR 3.0, 12.5] vs 2.0 [IQR 1.0, 4.0]), but had comparable rates of CAD before HF diagnosis (31.5 vs 30.4%). (Supplemental Table II)
The CID sub-groups in our analytic sample (n=974) differed according to several demographic and clinical variables, including age, sex, and race-ethnicity and prevalence of co-morbidities at first visit (Table 1). We also observed differences in CAD preceding HF; CAD was more common in IBD and Psoriasis and less common in SLE and SSc compared with non-CID controls. There were also significant differences in medications use at first visit including use of disease modifying anti-rheumatic drugs (DMARDs). Nearly all RA (96%) and SLE (91%) patients were receiving a disease-modifying anti-rheumatic drug within one year of the first visit.
Table 1.
Comparison of Analytical Sample According to Chronic Inflammatory Disease Group
| Variables | Control (n=433) |
HIV (n=74) |
IBD (n=50) |
Psoriasis (n=83) |
RA (n=116) |
SLE (n=106) |
SSc (n=112) |
p-value |
|---|---|---|---|---|---|---|---|---|
| Male sex (%) | 243 (56.1) | 59 (79.7) | 32 (64.0) | 46 (55.4) | 27 (23.3) | 16 (15.1) | 22 (19.6) | <0.001 |
| Age at baseline (mean (SD)) |
64.0 (13.6) |
48.9 (11.4) |
60.8 (16.0) |
66.5 (11.0) |
66.1 (10.4) |
49.4 (13.6) |
57.0 (13.2) |
<0.001 |
| Race-Ethnicity (%) | ||||||||
| White Non-Hispanic | 232 (53.6) | 22 (29.7) | 37 (74.0) | 62 (74.7) | 50 (43.1) | 29 (27.4) | 63 (56.2) | <0.001 |
| Asian | 11 (2.5) | 0 (0.0) | 0 (0.0) | 5 (6.0) | 5 (4.3) | 4 (3.8) | 4 (3.6) | |
| Black Non-Hispanic | 139 (32.1) | 46 (62.2) | 8 (16.0) | 7 (8.4) | 50 (43.1) | 53 (50.0) | 22 (19.6) | |
| Hispanic | 30 (6.9) | 3 (4.1) | 1 (2.0) | 5 (6.0) | 6 (5.2) | 14 (13.2) | 11 (9.8) | |
| Other | 21 (4.8) | 3 (4.1) | 4 (8.0) | 4 (4.8) | 5 (4.3) | 6 (5.7) | 12 (10.7) | |
| Insurance at baseline (%) | ||||||||
| Self-pay | 70 (16.2) | 27 (36.5) | 9 (18.0) | 11 (13.3) | 20 (17.2) | 22 (20.8) | 26 (23.2) | <0.001 |
| Medicaid | 21 (4.8) | 9 (12.2) | 1 (2.0) | 4 (4.8) | 8 (6.9) | 11 (10.4) | 2 (1.8) | |
| Medicare | 282 (65.1) | 28 (37.8) | 26 (52.0) | 54 (65.1) | 70 (60.3) | 45 (42.5) | 46 (41.1) | |
| private | 60 (13.9) | 10 (13.5) | 14 (28.0) | 14 (16.9) | 18 (15.5) | 28 (26.4) | 38 (33.9) | |
| others | 130 (30.0) | 37 (50.0) | 23 (46.0) | 25 (30.1) | 38 (32.8) | 50 (47.2) | 64 (57.1) | |
| Year of first visit (mean (SD)) |
2011 (5) | 2009 (4) | 2013 (2) | 2014 (2) | 2013 (2) | 2012 (3) | 2013 (2) | <0.001 |
| Timing of HF diagnosis (incident/on follow-up) (%) | 193 (44.6) | 44 (59.5) | 22 (44.0) | 33 (39.8) | 40 (34.5) | 45 (42.5) | 17 (15.2) | <0.001 |
| Echo parameters near HF diagnosis | ||||||||
| LVEF (mean (SD)) | 47.2 (16.6) | 41.0 (16.4) | 47.2 (17.0) | 49.5 (16.8) | 51.2 (14.5) | 48.9 (16.0) | 57.8 (10.7) | <0.001 |
| LVEDD (mean (SD)) | 4.9 (1.0) | 5.2 (1.0) | 4.89 (0.8) | 5.0 (1.0) | 4.8 (1.1) | 4.8 (1.0) | 4.3 (0.7) | <0.001 |
| MEAR (mean (SD)) | 1.6 (1.2) | 1.51 (0.97) | 1.1 (0.4) | 1.6 (1.3) | 1.3 (0.8) | 1.3 (0.6) | 1.1 (0.6) | <0.001 |
| LAV (mean (SD)) | 75.1 (42.9) | 68.6 (33.7) | 65.5 (28.9) | 72.4 (24.3) | 62.2 (27.9) | 66.8 (44.5) | 54.5 (22.9) | <0.001 |
| LAD (mean (SD)) | 4.1 (1.7) | 3.9 (0.7) | 3.86 (0.65) | 4.0 (0.7) | 3.7 (0.6) | 3.72 (0.77) | 3.50 (0.65) | 0.002 |
| LVEF category near diagnosis (%) | ||||||||
| <40 | 137 (31.6) | 34 (45.9) | 16 (32.0) | 29 (34.9) | 25 (21.6) | 29 (27.4) | 6 (5.4) | <0.001 |
| >50 | 203 (46.9) | 25 (33.8) | 27 (54.0) | 46 (55.4) | 67 (57.8) | 54 (50.9) | 92 (82.1) | |
| 50–40 | 93 (21.5) | 15 (20.3) | 7 (14.0) | 8 (9.6) | 24 (20.7) | 23 (21.7) | 14 (12.5) | |
| Comorbidities at first visit (%) | ||||||||
| Hypertension | 224 (51.7) | 26 (35.1) | 24 (48.0) | 56 (67.5) | 72 (62.1) | 67 (63.2) | 35 (31.2) | <0.001 |
| Diabetes mellitus | 134 (30.9) | 6 (8.1) | 17 (34.0) | 30 (36.1) | 42 (36.2) | 15 (14.2) | 4 (3.6) | <0.001 |
| Stroke | 56 (12.9) | 3 (4.1) | 6 (12.0) | 13 (15.7) | 14 (12.1) | 13 (12.3) | 5 (4.5) | 0.06 |
| Smoker | 229 (52.9) | 37 (50.0) | 30 (60.0) | 49 (59.0) | 61 (52.6) | 49 (46.2) | 44 (39.3) | <0.001 |
| CAD before HF diagnosis (%) | 152 (35.1) | 23 (31.1) | 20 (40.0) | 34 (41.0) | 36 (31.0) | 28 (26.4) | 14 (12.5) | <0.001 |
| Medication use at First Visit (%) | ||||||||
| Statin | 273 (63.0) | 35 (47.3) | 27 (54.0) | 66 (79.5) | 89 (76.7) | 42 (39.6) | 48 (42.9) | <0.001 |
| Beta blocker | 330 (76.2) | 51 (68.9) | 38 (76.0) | 70 (84.3) | 96 (82.8) | 85 (80.2) | 50 (44.6) | <0.001 |
| ACE I or ARB | 281 (64.9) | 44 (59.5) | 25 (50.0) | 60 (72.3) | 82 (70.7) | 43 (38.4) | 70 (66.0) | <0.001 |
| MRA | 136 (31.4) | 15 (20.3) | 13 (26.0) | 27 (32.5) | 35 (30.2) | 23 (21.7) | 32 (28.6) | 0.28 |
| Use of Disease-modifying anti-rheumatic drugs (DMARD) | ||||||||
| Biologic DMARD | 0 (0.0) | 0 (0.0) | 11 (22.0) | 13 (15.7) | 58 (50.0) | 10 (9.4) | 8 (7.1) | <0.001 |
| Non-biologic DMARD | 0 (0.0) | 0 (0.0) | 12 (24.0) | 14 (16.9) | 53 (45.7) | 86 (81.1) | 55 (49.1) | |
| Not on DMARD or with HIV | 433 (100.0) | 74 (100.0) | 27 (54.0) | 56 (67.5) | 5 (4.3) | 10 (9.4) | 49 (43.8) | |
| Three-group solution membership (%) | ||||||||
| HFp/mrEF | 283 (65.4) | 40 (54.1) | 34 (68.0) | 59 (71.1) | 90 (77.6) | 78 (73.6) | 103 (92.0) | <0.001 |
| HFrEF | 135 (31.2) | 31 (41.9) | 13 (26.0) | 22 (26.5) | 21 (18.1) | 24 (22.6) | 9 (8.0) | |
| HFrecEF | 15 (3.5) | 3 (4.1) | 3 (6.0) | 2 (2.4) | 5 (4.3) | 4 (3.8) | 0 (0.0) | |
| Two-group solution membership (HFrEF, %) | 165 (38.1) | 34 (45.9) | 15 (30.0) | 30 (36.1) | 29 (25.0) | 35 (33.0) | 16 (14.3) | <0.001 |
| Outcomes | ||||||||
| CV hospitalization after HF diagnosis (median [IQR]) | 6.0 [2.0, 12.0] |
10.0 [4.0, 17.0] |
8.0 [4.0, 12.0] |
5.0 [2.5, 8.0] |
7.0 [3.0, 13.0] |
6.5 [3.0, 15.0] |
4.0 [2.0, 10.0] |
<0.001 |
| Alive at Last Observation (%) | 365 (84.3) | 58 (78.4) | 40 (80.0) | 65 (78.3) | 90 (77.6) | 88 (83.0) | 84 (75.0) | 0.27 |
| Years of Observation (median [IQR]) | 5.3 [3.5, 8.0] |
6.6 [4.2, 11.6] |
4.9 [3.6, 6.7] |
4.3 [2.8, 5.5] |
4.9 [3.1, 7.0] |
5.4 [3.0, 7.7] |
5.0 [3.1, 6.8] |
<0.001 |
Notes: ACE I – angiotensin converting enzyme inhibitor; ARB – angiotensin II receptor blocker; CAD – coronary artery disease; HFp/mrEF – Heart failure with preserved/moderately reduced EF; HFrEF – HF with reduced EF; HFrecEF – HF with recovered EF; HIV – human immunodeficiency virus; IBD – inflammatory bowel disease; RA – rheumatoid arthritis; SLE – systemic lupus erythematosus; SSc – systemic sclerosis; LVEF – left ventricle ejection fraction; LVEDD – left ventricle end diastolic diameter; MEAR – mitral E/A ratio; LAV – left atrium volume; LAD – left atrium diameter; HF – heart failure; Mineralocorticoid Receptor Antagonist (MRA)
Trajectories of LVEF over time across different CIDs
We observed that different CID subtypes had different patterns over time in LVEF (Figure 1) and that CID subtype was associated with baseline LVEF and change over time (Supplemental Table III). Additional details regarding model and group selection for mixed-effects and latent trajectory modeling, as well as validation of the latent trajectory groups, are included in the supplement (Supplemental Methods and Supplemental Table IV).
Figure 1.
Estimated Left Ventricle Ejection Fraction over Time across Chronic Inflammatory Disease Types using Mixed Effects Model
Note: For estimation, other variables were set to the reference categories or to their mean values. HIV – human immunodeficiency virus; IBD – inflammatory bowel disease; RA – rheumatoid arthritis; SSc – systemic sclerosis; SLE – systemic lupus erythematosus
Regarding average changes in LVEF over time (mixed-effects models), IBD patients with HF had a significantly higher increase (recovery) in LVEF (βTime x IBD = 0.15, 95% CI: 0.005 to 0.30, p = 0.043) over time compared with HF patients without CIDs (controls). Meanwhile, SSc patients had significantly higher baseline LVEF (βSSc = 0.31, 95% CI: 0.20 to 0.43, p <0.001) but change in LVEF over time comparable to non-CID controls.
Investigation of heterogeneous, non-linear changes in LVEF over time via latent trajectory modeling demonstrated three distinct groups: patients with consistently preserved or midrange LVEF (HF with preserved/midrange ejection fraction, “HFp/mrEF”; N=687, 70.5% of patients), patients with consistently reduced LVEF (“HFrEF”; N=255, 26.2% of patients), and patients with reduced LVEF followed by substantial increase on subsequent echocardiography (HF with recovered ejection fraction, “HFrecEF”; N=32, 3.3% of patients) (Figure 2). In sensitivity analysis in which we relaxed our inclusion criteria to include patients with ≥2 LVEF measurements (rather than ≥3), similar latent trajectory patterns were observed.
Figure 2.
Observed Left Ventricle Ejection Fraction Over Time According to Latent Trajectory Groups.
Notes: HFp/mrEF – heart failure preserved/mid-range ejection fraction, HFreEF – heart failure reduced ejection fraction; HFrecEF – heart failure recovered ejection fraction
We observed significant differences across the groups in terms of demographic and clinical variables (Supplemental Table V). The HFrEF group was predominantly male, whereas the other groups were predominantly female, and the HFrecEF group was significantly younger than the other groups. Rates of diabetes and hypertension were comparable between groups, but rates of smoking and CAD prior to HF were higher in the HFrEF group. The HFrEF and HFrecEF groups had higher baseline left ventricle end-diastolic diameter and left atrium volume than HFp/mrEF. There were no significant differences in survival or cardiovascular disease-related hospitalizations between the groups.
Regarding the associations of CID subtypes with different LVEF trajectories, the majority of patients in each CID group and the control group were in the HFp/mrEF trajectory (Figure 3) but there were significant differences across CID subtypes. HF patients with RA, SLE, and IBD were significantly more likely than controls (HF patients without CIDs) to have HFp/mrEF (Table 2, Supplemental Table VI, Supplemental Figure I), whereas patients with HIV were significantly more likely than controls to be in the HFrEF group. Because no patients with SSc had HFrecEF, we performed a sensitivity analysis of two trajectory groups (HFp/mrEF vs. HFrEF) and observed that SSc patients were significantly more likely than controls to have HFp/mrEF (Supplemental Table VII and VIII).
Figure 3.
Distribution of Three Latent Trajectory Groups Per Chronic Inflammatory Disease Type
Note: HIV – human immunodeficiency virus; IBD – inflammatory bowel disease; RA – rheumatoid arthritis; SSc – systemic sclerosis; SLE – systemic lupus erythematosus; HFp/mrEF – heart failure preserved/mid-range ejection fraction, HFreEF – heart failure reduced ejection fraction; HFrecEF – heart failure recovered ejection fraction
Table 2.
Predicted Probabilities for Trajectory Group Membership using Multinomial Regression Model
| LVEF trajectory and CID group | Odds Ratio (95% CI)* |
Estimated Probability (%, 95% CI) |
Difference in Probability Estimated Difference between Groups (ref: None) (CID - none) |
|
|---|---|---|---|---|
| Estimate (%, 95% CI) |
p-value | |||
| HF with preserved or moderately reduced EF (HFp/mrEF) | ||||
| None | Reference | 61.1 (56.1 to 66.1) | Reference | - |
| HIV | 62.5 (52.4 to 72.5) | 1.4 (−7.5 to 10.2) | 0.76 | |
| IBD | 68.4 (62.6 to 74.2) | 7.3 (3.9 to 10.7) | <0.001 | |
| Psoriasis | 66.2 (55.6 to 76.8) | 5.1 (−4.8 to 15.0) | 0.30 | |
| RA | 71.1 (62.7 to 79.5) | 10 (2.4 to 17.7) | 0.012 | |
| SLE | 71.9 (63.3 to 80.5) | 10.8 (3.9 to 17.8) | 0.003 | |
| HF reduced EF (HFrEF) | ||||
| None | Reference | 35.1 (30.0 to 40.2) | Reference | - |
| HIV | 0.95 (0.61 to 1.46) | 34.1 (23.3 to 44.8) | −1 (−10.5 to 8.4) | 0.83 |
| IBD | 0.69 (0.56 to 0.83) | 27.6 (21.6 to 33.7) | −7.5 (−11.2 to −3.7) | <0.001 |
| Psoriasis | 0.81 (0.49 to 1.33) | 31.1 (20.0 to 42.2) | −4 (−14.4 to 6.4) | 0.44 |
| RA | 0.57 (0.36 to 0.90) | 24.2 (15.3 to 33.1) | −10.9 (−19.2 to −2.6) | 0.012 |
| SLE | 0.63 (0.44 to 0.92) | 27.0 (18.3 to 35.8) | −8.1 (−15.2 to −1.0) | 0.027 |
| HF recovered EF (HFrecEF) | ||||
| None | Reference | 3.8 (2.1 to 5.5) | Reference | - |
| HIV | 0.89 (0.84 to 0.95) | 3.5 (1.8 to 5.2) | −0.3 (−1.0 to 0.4) | 0.35 |
| IBD | 0.93 (0.88 to 0.98) | 4 (2.1 to 5.9) | 0.2 (−0.2 to 0.6) | 0.36 |
| Psoriasis | 0.65 (0.63 to 0.67) | 2.7 (1.4 to 4.1) | −1.1 (−1.8 to −0.4) | 0.002 |
| RA | 1.06 (1.02 to 1.10) | 4.7 (2.6 to 6.8) | 0.9 (0.1 to 1.7) | 0.022 |
| SLE | 0.22 (0.21 to 0.23) | 1 (0.51 to 1.6) | −2.8 (−3.9 to −1.6) | <0.001 |
Notes: CID – chronic inflammatory disease; HIV – human immunodeficiency virus; IBD – inflammatory bowel disease; RA – rheumatoid arthritis; SLE – systemic lupus erythematosus. Scleroderma excluded from modeling due to zero recovered EF cases. Results are adjusted for age, sex, diabetes, and hypertension at first visit, race, year of entry into the database, and timing of HF diagnosis (at first visit vs incident on follow-up). The odds ratio for multinomial regression has two references: the reference outcome (HFp/mrEF) and the reference category for each predictor (no CID for CID variable). Interpreting the odds ratio for IBD in HFrEF section would then be: An individual with IBD had significantly lower odds of being in the HFrEF group than being in the HFp/mrEF group compared to individuals with no CID if all other variables were held equal. Estimated probabilities for that same group can be read as follows: IBD have 27.6% probability of being in the HFrEF group and this was significantly lower than the probability of being in the HFrEF group by a similar individual without CID.
We also observed that other clinical variables besides CID subtype were differentially associated with LVEF trajectories (Supplemental Table VI). All matching variables (age, sex, race, year of first entry, DM and HTN at baseline) were significant predictors of latent trajectory membership, as was a diagnosis of CAD prior to HF, which was associated with lower odds of HFp/mrEF. Incorporating timing of HF diagnosis (i.e., having HF at or up to 90-days postbaseline vs incident or diagnosis of HF), first visit smoking status, or statin use at first visit did not improve fit and were not included in the final model.
Discussion
In a nested cohort consisting of HF patients with CIDs and frequency-matched controls with HF, we observed that different CIDs have different LVEF trajectories as assessed by serial echocardiograms. We identified three latent trajectory groups based on LVEF that correspond to known subtypes of HF: HFrEF, HFp/mrEF, and HFrecEF.7–11 Compared with controls, patients with RA, IBD, and SLE were more likely to be in the HFp/mrEF trajectory groups and less likely to be in the HFrEF trajectory group. Compared with controls, patients with RA were also more likely to be in the HFrecEF group while patients with SLE and psoriasis were less likely to be in the HFrecEF group. While the HFrEF trajectory was more common in patients with HIV than controls, the predicted probability of an HIV patient being classified into the HFrEF group was not significantly different than the control group in adjusted models. Patients with SSc were more likely to be in the HFp/mrEF group compared with the HFrEF group than controls.
Prior studies have described a wide range of myocardial involvement in CIDs ranging from acute myocarditis to a more subacute inflammatory cardiomyopathy.15–20 These processes are associated with an increased risk of HF in CID populations.21–23 However, comparatively sparse data exist on trajectories of myocardial dysfunction after HF diagnosis in these conditions. We were able to leverage a unique cohort to describe LVEF trajectories after HF diagnosis across a number of CID populations. Our findings provide important clinical insight and may also have mechanistic implications. The HFp/mrEF trajectory was the predominant group in all CIDs and was significantly more likely in patients with RA, IBD, SLE, and SSc. The HFpEF/HFmrEF predominance likely reflects chronic low-level systemic inflammation and immune dysregulation leading to myocardial fibrosis and diastolic dysfunction.17
We observed RA to be the only CID in which patients were more likely to belong to the HFrecEF group compared with controls. This may relate to the underlying pathophysiology of HF in RA, cohort characteristics, or differences in treatment. Approximately half of the patients with RA were on biologic DMARDs, greater than that observed for other CIDs. Immuno-modulatory treatment may play a role in improving HF outcomes in CIDs but short-term trials have been inconclusive.1,24 Prolonged anti-inflammatory RA treatment may be needed to affect inflammatory mechanisms that drive HF progression17,22 translating to EF recovery, and may explain some of the higher rates of LVEF recovery for RA patients. Reversal of adverse remodeling, a possible reason for observed EF recovery, likely requires a balance of suppression and modulation of innate and adaptive immunity that likely has variable clinical penetrance for a given CID25–27 along with any neuro-hormonal effects of guideline-directed medical therapy (GDMT).11
Changes over time in usage of immune-modulatory drugs and GDMT for CID, HF, and comorbidities including HTN likely also played a role in our findings. Certainly, it is possible that changes in treatment guidelines during the time studied influenced EF trajectories especially since GDMT has been associated with some degree normalization of LVEF.11,28 Indeed, we included year of first visit (baseline) as a covariate and observed an association of year of first visit with latent group membership; specifically, later year of entry was associated with higher likelihood of HFrecEF, which may suggest increased uptake over time of GDMT and/or improved therapies for underlying (ischemic and non-ischemic) etiologies of HF. Consistent with this, we observed higher baseline use of guideline-directed HFrEF medications in more individuals coming in more recent years (e.g. beta blocker use at baseline before 2010: 59% vs after 2010: 77%; full table not shown). These hypothesis-generating findings highlight the importance of future analyses of effects of immunomodulatory and GDMT on serial cardiac function, which will need to carefully address confounding by timing, type, and intensity of immunomodulation, and also ideally would separately consider CIDs due to distinct pathophysiologies and treatments over time (e.g., less toxicity and longer life expectancy with newer HIV therapies).
History of CAD prior to HF diagnosis was associated with lower estimated average LVEF in the mixed model and people with CAD were significantly less likely to be in the HFp/mEF group in the latent trajectory model. This is consistent with previous findings in the Swiss cohort that having a history of CAD was associated with higher odds to be in the HFrEF group (as well as the HFrecEF group, which may be a result of myocardial recovery following revascularization) compared with the HFp/mEF group.29 It also aligns with the pathophysiologic cascade of CAD leading to ischemic events, pathologic remodeling, and pump failure.30
Our study has several limitations. This is a cohort derived from a healthcare system in a single metropolitan area and our findings might not be generalizable to all centers. We also relied on diagnosis codes for ascertainment of HF and comorbidities. While we used previously validated codes, some degree of misclassification cannot be fully eliminated. Our use of EHR captures outcomes from routine clinical practice but we cannot fully control for selection bias arising from differences in retention in care, treatment received, and timing/frequency of echocardiography as well as medical center-specific context. We sought to address this to some extent by restricting analyses to patients with HF, for at least one, and often more, echocardiographic examinations are indicated. Nevertheless, differences in timing of HF diagnosis (baseline vs on follow-up) and frequency of echoes are also insufficient to explain why these LVEF patterns were uncovered. Timing of diagnosis was not an important predictor in the multinomial model and most individuals had echoes spread over six years of follow-up with very few (<1%) having all their echoes done within two months of follow-up. Still, our findings would have limited generalizability and our analyses should be replicated in other centers or using data of clinical research networks. Given the size and nature of the cohort (which made clinical physician adjudication of diagnoses from >30,000 people outside the scope of these analyses), we were unable to granularly define CID severity during follow-up or in response to management and other management changes that may occur over time, which may lead to inadequate capture of disease severity and heterogeneity and may explain why we did not detect significant differences in clinical outcomes across groups unlike those found in other studies.9,10,31,32 Relatedly, we also cannot fully assess the impact of heart failure medications and GDMT (baseline or changes over time) on HF trajectories. While we observed that medication use at baseline aligned with predicted trajectory groups (e.g., lower use of beta blocker and angiotensin converting enzyme inhibitor/angiotensin II receptor blockers use among SSc where mostly had a HFpEF trajectory), we cannot make causal claims due to inability to adequately control for treatment selection. We also did not have data on GDMT changes over time and recommend the conduct of more research that assess their role in LVEF trajectories. We also faced the risk over-extraction of groups during LTA; however, we show that even if we use the two-group or one-group solution, CID remains a significant factor in LVEF trajectory (see Supplemental table IV and V). Finally, due to our analytic sample – small in absolute terms, though comparatively large when considered in the context of limited data on HF patients with CIDs – and potential residual confounding, we were unable to see significant differences in outcomes across the latent groups in contrast to previous research suggesting that those who recover LVEF tend to have better outcomes.7,32
Conclusion
Among patients with CIDs and HF, there appear to be distinct patterns of LVEF trajectory over time which are associated with different CIDs. These findings may relate to distinct underlying pathophysiologic processes across CIDs as well as differences in treatment for the conditions. Future analyses focused on effects of CID severity and immunomodulatory therapy on HF patterns in CIDs are warranted to inform research on immune-mediated HF contributors as well as clinical care of patients with chronic inflammatory conditions.
Supplementary Material
Supplemental File.
A. Supplemental Methods
B. Supplemental Results
Supplemental Table I. Operational Definitions for Diseases and Co-morbidities
Supplemental Table II. Medication Use at Baseline According to Chronic Inflammatory Disease
Supplemental Table III. Mixed effects model: Predictors of LVEF over time
Supplemental Table IV. Latent Trajectory Analysis Model Selection
Supplemental Table V. Comparison of Latent Groups in Three-Group Solution.
Supplemental Table VI. Results of Multinomial Regression
Supplemental Table VII. Comparison of Latent Groups in Two-Group Solution.
Supplemental Table VIII. Predictors of Belonging in HFrEF group (vs HFp/mrEF) using Logistic Regression
Supplemental Figure I. Differences of Predicted Probabilities between CID and Control Latent Group using the Multinomial Regression Model
What’s new?
Among people with heart failure and chronic inflammatory conditions, such as rheumatoid arthritis or systemic lupus erythematosus, different chronic inflammatory conditions were associated with different trajectories of left ventricular ejection fraction (LVEF) over time.
These findings suggest that disease-specific patterns of immune dysregulation and inflammation correspond to different heart failure presentations and substrates.
Methodologically, latent trajectory analyses may be useful for uncovering heterogeneity in disease substrate over time in chronic, dynamic conditions such as heart failure.
What are the clinical implications?
Different chronic inflammatory diseases, as well as the treatments for these diseases, may differentially impact heart failure presentation and prognosis.
Future analyses should investigate disease modifying therapy-associated changes in cardiac substrate and clinical outcomes among people with heart failure and chronic inflammatory diseases. Such analyses may yield clinically relevant insights for chronic inflammatory disease patients while also highlighting potential targets for intervention in the broader population with heart failure.
Acknowledgements
We acknowledge the Northwestern Medicine Enterprise Data Warehouse for logistical and analytic support.
Sources of Funding
The authors thank the American Heart Association Fellow‐to‐Faculty Award (16FTF31200010; PI: Feinstein) and Predoctoral Fellowship Award (825793; PI: Rivera, 2021–2023), as well as the National Institutes of Health (P30AI117943 and UL1TR001422) for funding support.
Non-standard Abbreviations and Acronyms
- ACE I
angiotensin converting enzyme inhibitor
- ARB
angiotensin II receptor blocker
- CAD
coronary artery disease
- CID
chronic inflammatory diseases
- DM
diabetes mellitus
- DMARDs
disease modifying anti-rheumatic drugs
- EHR
electronic health record
- GDMT
guideline-directed medical therapy
- HF
heart failure
- HFpEF
heart Failure with preserved ejection fraction
- HFrEF
heart Failure with reduced ejection fraction
- HFrecEF
heart Failure with recovered ejection fraction
- HFp/mrEF
heart Failure with preserved or midrange LVEF
- HIV
human immunodeficiency virus
- HTN
hypertension
- IBD
inflammatory bowel disease
- LTA
latent trajectory analysis
- LVEF
left ventricular ejection fraction
- LVEDD
Left ventricle end diastolic diameter
- MEAR
Mitral E/A ratio
- LAV
Left atrium volume
- LAD
left atrium diameter
- MRA
Mineralocorticoid Receptor Antagonist
- RA
rheumatoid arthritis
- SSc
systemic sclerosis
- SLE
systemic lupus erythematosus
Footnotes
Disclosures
AR: None
AS: None
FA: Dr. Ahmad has received consulting fees from Amgen and Teladoc Livango.
ET: None
JW: None
DLJ: None
MJF: Dr. Feinstein serves on the medical advisory board for Novartis.
References
- 1.Adamo L, Rocha-Resende C, Prabhu SD, Mann DL. Reappraising the role of inflammation in heart failure. Nat Rev Cardiol 2020;17:269–285. [DOI] [PubMed] [Google Scholar]
- 2.Trachtenberg BH, Hare JM. Inflammatory cardiomyopathic syndromes. Circ Res 2017;121:803–818. [DOI] [PubMed] [Google Scholar]
- 3.Prasada S, Rivera A, Nishtala A, Pawlowski AE, Sinha A, Bundy JD, Chadha SA, Ahmad FS, Khan SS, Achenbach C, et al. Differential Associations of Chronic Inflammatory Diseases With Incident Heart Failure. JACC Heart Fail 2020;8:489–498. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ahmad FS, Chan C, Rosenman MB, Post WS, Fort DG, Greenland P, Liu KJ, Kho AN, Allen NB. Validity of cardiovascular data from electronic sources: The multi-ethnic study of atherosclerosis and HealthLNK. Circulation 2017;136:1207–1216. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Huang H, Turner M, Raju S, Reich J, Leatherman S, Armstrong K, Woods P, Ferguson RE, Fiore LD, Lederle FA. Identification of Acute Decompensated Heart Failure Hospitalizations Using Administrative Data. Am J Cardiol 2017;119:1791–1796. [DOI] [PubMed] [Google Scholar]
- 6.McCormick N, Lacaille D, Bhole V, Avina-Zubieta JA. Validity of heart failure diagnoses in administrative databases: A systematic review and meta-analysis. PLoS One 2014;9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Basuray A, French B, Ky B, Vorovich E, Olt C, Sweitzer NK, Cappola TP, Fang JC. Heart failure with recovered ejection fraction clinical description, biomarkers, and outcomes. Circulation 2014;129:2380–2387. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Clarke CL, Grunwald GK, Allen LA, Barón AE, Peterson PN, Brand DW, Magid DJ, Masoudi FA. Natural history of left ventricular ejection fraction in patients with heart failure. Circ Cardiovasc Qual Outcomes 2013;6:680–686. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Lupón J, Gavidia-Bovadilla G, Ferrer E, De Antonio M, Perera-Lluna A, López-Ayerbe J, Domingo M, Núñez J, Zamora E, Moliner P, et al. Heart Failure With Preserved Ejection Fraction Infrequently Evolves Toward a Reduced Phenotype in Long-Term Survivors: A Long-Term Prospective Longitudinal Study. Circ Hear Fail 2019;12:1–9. [DOI] [PubMed] [Google Scholar]
- 10.Tsuji K, Sakata Y, Nochioka K, Miura M, Yamauchi T, Onose T, Abe R, Oikawa T, Kasahara S, Sato M, et al. Characterization of heart failure patients with mid-range left ventricular ejection fraction—a report from the CHART-2 Study. Eur J Heart Fail 2017;19:1258–1269. [DOI] [PubMed] [Google Scholar]
- 11.Wilcox JE, Fang JC, Margulies KB, Mann DL. Heart Failure With Recovered Left Ventricular Ejection Fraction: JACC Scientific Expert Panel. J Am Coll Cardiol 2020; 76:719–734. [DOI] [PubMed] [Google Scholar]
- 12.Loughran T, Nagin DS. Finite sample effects in group-based trajectory models. Sociol Methods Res 2006;35:250–278. [Google Scholar]
- 13.Bengtson AM, Pence BW, Powers KA, Weaver MA, Mimiaga MJ, Gaynes BN, O’Cleirigh C, Christopoulos K, Christopher Mathews W, Crane H, et al. Trajectories of Depressive Symptoms Among a Population of HIV-Infected Men and Women in Routine HIV Care in the United States. AIDS Behav 2018;22:3176–3187. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Nagin DS, Odgers CL. Group-Based Trajectory Modeling in Clinical Research. Annu Rev Clin Psychol 2010;6:109–138. [DOI] [PubMed] [Google Scholar]
- 15.Alexandroff AB, Pauriah M, Camp RDR, Lang CC, Struthers AD, Armstrong DJ. More than skin deep: Atherosclerosis as a systemic manifestation of psoriasis. Br J Dermatol 2009;161:1–7. [DOI] [PubMed] [Google Scholar]
- 16.Bigeh A, Sanchez A, Maestas C, Gulati M. Inflammatory bowel disease and the risk for cardiovascular disease: Does all inflammation lead to heart disease? Trends Cardiovasc Med 2019;30:463–469. [DOI] [PubMed] [Google Scholar]
- 17.Błyszczuk P, Szekanecz Z. Pathogenesis of ischaemic and non-ischaemic heart diseases in rheumatoid arthritis. RMD Open 2020;6: e001032. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Dhakal BP, Kim CH, Al-Kindi SG, Oliveira GH. Heart failure in systemic lupus erythematosus. Trends Cardiovasc Med 2018;28:187–197. [DOI] [PubMed] [Google Scholar]
- 19.Psarras A, Soulaidopoulos S, Garyfallos A, Kitas G, Dimitroulas T. A critical view on cardiovascular risk in systemic sclerosis. Rheumatol Int 2017;37:85–95. [DOI] [PubMed] [Google Scholar]
- 20.Sinha A, Feinstein MJ. Coronary Artery Disease Manifestations in HIV: What, How, and Why. Can J Cardiol 2019;35:270–279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Prasada S, Rivera A, Nishtala A, Pawlowski AE, Sinha A, Bundy JD, Chadha SA, Ahmad FS, Khan SS, Achenbach C. Differential associations of chronic inflammatory diseases with incident heart failure. JACC Hear Fail 2020;8:489–498. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Khalid U, Egeberg A, Ahlehoff O, Lane D, Gislason GH, Lip GYH, Hansen PR. Incident heart failure in patients with rheumatoid arthritis: A nationwide cohort study. J Am Heart Assoc 2018;7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Khalid U, Ahlehoff O, Gislason GH, Kristensen SL, Skov L, Torp-Pedersen C, Hansen PR. Psoriasis and risk of heart failure: A nationwide cohort study. Eur J Heart Fail 2014;16:743–748. [DOI] [PubMed] [Google Scholar]
- 24.Riehle C, Bauersachs J. Key inflammatory mechanisms underlying heart failure. Herz 2019;44:96–106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Azevedo PS, Polegato BF, Minicucci MF, Paiva SAR, Zornoff LAM. Cardiac Remodeling: Concepts, Clinical Impact, Pathophysiological Mechanisms and Pharmacologic Treatment. Arq Bras Cardiol 2016;106:62–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Januzzi JL, Hauptman PJ. Reversed reverse remodeling: Can biomonitoring solve the clinical conundrum of the 3Rs? Circ Hear Fail 2014;7:388–390. [DOI] [PubMed] [Google Scholar]
- 27.Mann DL, Barger PM, Burkhoff D. Myocardial recovery and the failing heart: Myth, magic, or molecular target? J Am Coll Cardiol 2012;60:2465–2472. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Albert J, Lezius S, Stork S, Morbach C, Guder G, Frantz S, Wegscheider K, Ertl G, Angermann CE. Trajectories of Left Ventricular Ejection Fraction After Acute Decompensation for Systolic Heart Failure: Concomitant Echocardiographic and Systemic Changes, Predictors, and Impact on Clinical Outcomes. J Am Heart Assoc 2021;10: e017822. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Vedin O, Lam CSP, Koh AS, Benson L, Teng THK, Tay WT, Braun OO, Savarese G, Dahlström U, Lund LH. Significance of Ischemic Heart Disease in Patients with Heart Failure and Preserved, Midrange, and Reduced Ejection Fraction: A Nationwide Cohort Study. Circ Hear Fail 2017;10:1–9. [DOI] [PubMed] [Google Scholar]
- 30.Elgendy IY, Mahtta D, Pepine CJ. Medical Therapy for Heart Failure Caused by Ischemic Heart Disease. Circ Res 2019;124:1520–1535. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Dunlay SM, Roger VL, Weston SA, Jiang R, Redfield MM. Longitudinal changes in ejection fraction in heart failure patients with preserved and reduced ejection fraction. Circ Hear Fail 2012;5:720–726. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Kalogeropoulos AP, Fonarow GC, Georgiopoulou V, Burkman G, Siwamogsatham S, Patel A, Li S, Papadimitriou L, Butler J. Characteristics and outcomes of adult outpatients with heart failure and improved or recovered ejection fraction. JAMA Cardiol 2016;1:510–518. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental File.
A. Supplemental Methods
B. Supplemental Results
Supplemental Table I. Operational Definitions for Diseases and Co-morbidities
Supplemental Table II. Medication Use at Baseline According to Chronic Inflammatory Disease
Supplemental Table III. Mixed effects model: Predictors of LVEF over time
Supplemental Table IV. Latent Trajectory Analysis Model Selection
Supplemental Table V. Comparison of Latent Groups in Three-Group Solution.
Supplemental Table VI. Results of Multinomial Regression
Supplemental Table VII. Comparison of Latent Groups in Two-Group Solution.
Supplemental Table VIII. Predictors of Belonging in HFrEF group (vs HFp/mrEF) using Logistic Regression
Supplemental Figure I. Differences of Predicted Probabilities between CID and Control Latent Group using the Multinomial Regression Model