Identification of fatigue subtypes and their correlates in prevalent heart failure: A secondary analysis of the Atherosclerosis Risk in Communities Study

Abstract

Background:

Among HF patients, fatigue is common and linked to quality of life and functional status. Fatigue is hypothesized to manifest as multiple types, with general and exertional components. Unique subtypes of fatigue in HF may require differential assessment and treatment to improve outcomes. We conducted this study to identify fatigue subtypes in persons with prevalent HF in the Atherosclerosis Risk in Communities (ARIC) study and describe the distribution of characteristics across subtypes.

Methods:

We performed a cross-sectional analysis of 1,065 participants with prevalent HF at ARIC Visit 5 (2011-13). We measured exertional fatigue using the Modified MRC Breathlessness scale, and general fatigue using the PROMIS fatigue scale. We used latent class analysis to identify subtypes of fatigue. Number of classes was determined using model fit statistics and classes were interpreted and assigned fatigue severity rating based on the conditional probability of endorsing survey items given class. We compared characteristics across classes using multinomial regression.

Results:

Overall, participants were 54% female and 38% Black with a mean age of 77. We identified four latent classes (fatigue subtypes): 1) high general/high exertional fatigue (18%), 2) high general/low exertional fatigue (27%), 3) moderate general/moderate exertional fatigue (20%), and 4) low/no general & exertional fatigue (35%). Female sex, Black race, lower education level, higher BMI, increased depressive symptoms, and higher prevalence of diabetes were associated with higher levels of general and exertional fatigue.

Conclusions:

We identified unique subtypes of fatigue in HF patients that have not been previously described. Within subtype, general and exertional fatigue were mostly concordant in severity, and exertional fatigue only occurred in conjunction with general fatigue, not alone. Further understanding these fatigue types and their relationships to outcomes may enhance our understanding of the symptom experience and inform prognostication and secondary prevention efforts for persons with HF.

Introduction

Heart failure (HF) affects approximately 6 million people in the US and is associated with marked morbidity and mortality. Among HF patients, fatigue is a common and distressing symptom that significantly impacts patient quality of life and function.¹ Fatigue is associated with both patient-reported and clinical outcomes such as worse life satisfaction², decreased quality of life^2,3, worse perceived physical, general and emotional health⁴, and impaired physical and role functioning.⁵ Increased severity of fatigue is also associated with worse occupational performance, increased activities of daily living (ADL) and instrumental activities of daily living (IADL) dependence^6,7, and worse HF self-care⁸.

Fatigue in HF has two types: 1) general fatigue and 2) exertional fatigue.^9,10 General fatigue in multiple chronic conditions, including HF, is described as a multi-dimensional, persistent, and full-body experience not related to exertion. In addition to being unrelated to activity, it is typically not relieved by normal recuperative methods such as rest, nutrients, or sleep. General fatigue is also described as a multi-dimensional experience involving physical, emotional, and social aspects.⁹ Although some HF studies have found general fatigue to be associated with stroke, anemia, and depression, the underlying mechanisms of general fatigue in HF are largely unknown.¹¹ Exertional fatigue, however, is directly related to exertional activity and is often linked to dyspnea and shortness of breath in patients with HF.¹² Exertional fatigue is hypothesized to involve reduced cardiac output, poor heart rate response, and inadequacy of peripheral vascular dilation leading to exertional symptoms that involve respiratory and muscle fatigue.¹²

Although fatigue in HF has been described as having general and exertional components, the majority of current studies examine fatigue in HF as a single type. The few that do examine both general and exertional fatigue analyze them individually and do not address their co-occurrence.^8,10,13,14 Further, these studies do not provide a consistent understanding of the predictors of these types of fatigue. Therefore, we have a poor understanding of the prevalence and distribution of general and exertional fatigue, and their co-occurrence, in populations with HF.

Understanding different fatigue subtypes in individuals with HF is particularly important because different subtypes may require differential assessment methods and interventions. Further, to develop new, clinically meaningful hypotheses regarding the fatigue symptom experience in those with HF, it is important to identify potential fatigue subtypes, and evaluate their prevalence and related characteristics for further investigation. Therefore, we conducted a study to identify unique subtypes of fatigue in older adults with prevalent HF in the Atherosclerosis Risk in Communities (ARIC) study using latent class analysis and described the distribution of demographic and clinical characteristics across fatigue subtypes.

Methods

The data, analytic methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure. However, the ARIC study data are publicly available through the database of Genotypes and Phenotypes and the National Heart, Lung, and Blood Institute Biological Specimen and Data Repository Information Coordinating Center. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Parent Study:

This analysis was conducted using data from Visit 5 (2011-2013) of the Atherosclerosis Risk in Communities (ARIC) study. Visit 5 was the first study visit to take place after the start of HF adjudication in ARIC in 2005. ARIC is an ongoing, prospective epidemiologic cohort study which enrolled an initial sample size of 15,792 participants between 1987 to 1989 from 4 different US communities in North Carolina, Mississippi, Minnesota, and Maryland.¹⁵ Participants were subsequently examined at serial study visits. The ARIC study was approved by institutional review boards at each participating site. All participants provided written informed consent.

Sample:

Visit 5 of the ARIC study included 6,538 participants. We excluded all participants without a diagnosis of prevalent HF (n=5,426) and the small number who were non-black or non-white (n=2). This provided us with a sample of 1,112 participants. Of the 1,112 participants with prevalent HF, latent class analysis was performed on 1,065 participants (47 were excluded due to majority missing data on fatigue items).

Measures:

Prevalent Heart Failure:

We used the ARIC definition of prevalent ‘definite’ or ‘possible’ HF prior to visit 5 to encompass all participants with an existing diagnosis of HF rather than an acute decompensation. Prevalent definite or possible HF is defined as: 1) HF-related hospitalizations occurring since 2005 that were adjudicated by trained and certified physicians¹⁶, or 2) a physician HF survey endorsing “has this patient ever had HF or cardiomyopathy?” with a HF onset date prior to 2005, or 3) International Classification of Diseases, Ninth Revision, Clinical Modification code 428 for hospitalizations in first or any position prior to 2005¹⁷, or 4) HF self-report from visits 3-4 or on annual follow-up call that is not refuted by the physician HF survey, or 5) self-report of treatment for HF from any study visit or annual follow up prior to visit 5.

Exertional fatigue:

We used the modified Medical Research Council (MRC) breathlessness scale measured at visit 5 (2011-2013) to operationalize exertional fatigue in the latent class analysis. This scale is validated for the measurement of dyspnea on exertion. ^18,19,20 Patient reported dyspnea on exertion is the closest construct to exertional fatigue that is measured in ARIC. Subjective perception of exertion and dyspnea (eg. Borg perceived exertion scales) are closely interrelated and used together in cardiopulmonary exercise testing to evaluate perceived exertion level and to identify fatigue limitations with exertion.^21,22 Therefore, in this analysis, we have used exertional dyspnea as a reflection of exertional fatigue. The Modified MRC Breathlessness scale includes 5 items asking 1) are you troubled by shortness of breath when hurrying on the level or walking up a slight hill? 2) do you have to walk slower than people of your age on the level because of breathlessness? 3) do you ever have to stop for breath with walking at your own pace on the level? 4) Do you ever have to stop for breath after walking about 100 yards (or after a few minutes) on the level? 5) are you too breathless to leave the house or breathless on dressing or undressing? Participant responses to each of these items are dichotomous (yes/no).

General fatigue:

We used the Patient-Reported outcomes Measurement Information System (PROMIS)²³ fatigue scale items measured in ARIC at the first semi-annual follow-up call post visit 5 (2012-2014) to operationalize general fatigue in the latent class analysis. The PROMIS fatigue item bank is psychometrically validated in diverse populations to measure general fatigue.^24,25 The ARIC study collected data on 5 items from the PROMIS fatigue item bank asking: In the past 7 days 1) how often did you feel tired? 2) how often did you experience extreme exhaustion? 3) how often did you run out of energy? 4) how often were you too tired to think clearly? and 5) how often were you too tired to take a bath or shower? We dichotomized the responses to each of these items for use in the latent class analysis. A participant’s response to an item was assigned a 1 (yes) if they responded “sometimes”, “often”, or “always”. A participant’s response to an item was assigned a 0 (no) if they responded “rarely” or “never”. Thresholds for dichotomization are based on the Patient Reported Outcome Measures Information System (PROMIS) T-score mapping scheme which maps likert responses of scale items to population normalized scale score (T-score) according to the probability of selecting a particular response, given the overall scale score.²⁶ For the PROMIS fatigue item bank, the responses of “rarely” or “never” (and their equivalents) map to within 1 standard deviation (10 units) of a normalized T-score of 50, which indicates a “normal” score, or the absence of the symptom operationalized by the scale.

Demographic and clinical characteristics:

Demographic and clinical characteristics used to describe the sample overall and across fatigue subtype classes were selected 1) for their clinical relevance to HF physiology and/or outcomes, or 2) their known or hypothesized associations with fatigue in HF based on existing literature and a priori theory, or 3) their known association with fatigue in other chronic conditions.

Demographic characteristics in this analysis included self-reported sex (male or female); race (Black or White); and level of education (less than high school, high school/vocational school, college, graduate/professional school); and area deprivation index (ADI; Score of 1 - least socioeconomically disadvantaged - to 100 - most socioeconomically disadvantaged; using geocoded ARIC participant addresses at Visit 5 and linked to national and state ADI ranks^27-30). Clinical characteristics included age (years calculated from date of birth); body mass index (BMI; kg/m²); systolic blood pressure (mmHg); smoking status (self-reported current, former, or never smoker); pulmonary function (% predicted forced vital capacity and % predicted forced expiratory ratio, Spirometry following the American Thoracic Society quality criteria³¹ using a dry SensorMed 827 spriometer (Ohio Medical) and OMI analysis software); kidney function (eGFR serum creatinine and cystatin C (ml/min/1.73m²) using the Chronic Kidney Disease Epidemiology Collaboration equation.³²); NT-proBNP (pg/mL); high sensitivity cardiac Troponin T (ng/L); use of medications potentially related to fatigue or HF severity in the last 4 weeks (central nervous system (CNS) depressants, antihypertensives, loop diuretics, beta-blockers) based on 2004 medication billing codes; self-reported depressive symptoms (Center for Epidemiologic Studies Depression Scale (CES-D)³³, Score of 0-60, higher score indicates higher depressive symptoms); and prevalent comorbidities including diabetes (Hemoglobin A1C value of ≥ 6.5%³⁴, using medication for diabetes (2004 medication billing code), or self-reported diagnosis of diabetes by a physician), coronary heart disease (Coronary heart disease hospitalization prior to visit 5 or baseline diagnosis of coronary heart disease defined as adjudicated electrocardiogram (ECG) confirmed myocardial infarction (MI) or self-reported history of MI, heart or arterial surgery, coronary bypass, balloon angioplasty, or angioplasty of coronary arteries), and atrial fibrillation (12-lead ECG, hospital discharge records prior to visit 5³⁵). All demographic and clinical characteristics were determined at visit 5, except for education level, which was obtained at baseline enrollment.

Statistical Analyses:

We performed a latent class analysis (LCA) using the Mplus software (version 8)³⁶ with the general and exertional fatigue survey items as our latent class indicators. Our LCA used dichotomous ratings (0=no, 1=yes) on each of the 10 fatigue items (5 general fatigue items, 5 exertional fatigue items). LCA assumes that latent (underlying) classes, or groups, of individuals exist. It uses patterns of responses on categorical items to estimate two key parameters: 1) latent class probabilities, and 2) conditional probabilities. Latent class probabilities are the proportion of individuals in the sample that are estimated to be in each class. Conditional probabilities are the probabilities of endorsing an individual item, given membership in each class. ³⁷ We used these two parameters to estimate each participant’s 1) probability of membership in each latent class, given their pattern of responses to the fatigue items, and 2) most likely latent class membership.³⁸ Modal class membership was assigned to each participant using the highest probability of membership in a latent class. We determined the optimal number of latent classes based on bootstrapped likelihood ratio tests and BIC.³⁷ We checked the LCA assumption of item conditional independence using standardized bivariate residuals. The final model fit was evaluated using observed and model-based counts of individuals for each symptom pattern.³⁹ Latent class analysis was used because it groups people based on their patterns of responses to survey items, rather than grouping survey items together.

We performed descriptive statistics on the overall sample and by most likely class membership. Aggregate descriptive statistics describing the sample across classes were determined based on most likely fatigue class assignment. We tested for differences in demographic and clinical characteristics across most likely fatigue class using univariate multinomial latent class regression accounting for most likely class membership classification error using the 3-step method described by Asparouhov et. al.³⁸ Univariate regression analyses were conducted using complete case analysis. A significance level of 0.05 was established a priori. These analyses were performed using STATA (version 17)⁴⁰ and Mplus (version 8)³⁶ statistical analysis software.

Results

Overall Sample Characteristics:

Overall, the sample was 54% female (n=599), 38% Black (n=426). The mean age of participants was 77 ± 5.5 years. Most participants had a high school/vocational school education or less (n=741, 67%) and were former or current smokers (n=580, 63%). Mean body mass index (BMI) was 30 ± 7 kg/m², systolic blood pressure was 131 ± 20 mmHg, and eGFR was 58 ± 21 ml/min/1.73m². Depressive symptoms on average were low with a median score of 3 (1, 6). Table 1 shows a full summary of demographic and clinical characteristics of the overall sample.

Table 1.

Demographic and clinical characteristics of those with prevalent heart failure in the ARIC study at visit 5 (2011-2013).

	Overall
N	1,112
Age	77 ± 6
Sex
Female	599 (54%)
Male	513 (46%)
Race
Black	426 (38%)
White	686 (62%)
Education Level	(n=1110)
Less than High School	306 (28%)
High School/Vocational	435 (39%)
College	262 (24%)
Graduate/Professional	107 (10%)
Area Deprivation Index	47 (32 – 69)
Smoking Status	(n=917)
Current Smoker	61 (7%)
Former Smoker	519 (56%)
Never Smoker	337 (37%)
Systolic Blood Pressure (mmHg)	131 ± 20
Comorbidities
Coronary Heart Disease	484 (44%)
Atrial Fibrillation	236 (21%)
Diabetes	520 (47%, n=1096)
Medications	(n=1103)
Antihypertensives	1049 (94%)
Beta blockers	685 (62%)
Loop Diuretics	360 (34%)
CNS Depressants	106 (10%)
Kidney Function (sCr & cystatin C eGFR, ml/min/1.73m2)	58 ± 21
Cardiac Biomarkers median (Q1, Q3)
NT-proBNP (>= 450 pg/mL)	275 (116, 720)
hs-Troponin T (>= 14 ng/L)	15 (10, 26)
Pulmonary Function
% predicted FVC	90 ± 26
% predicted FVC/FEV1	95 ± 17
Depressive Symptoms median (Q1, Q3)	3 (1, 6)
Body Mass Index (kg/m2)	30 ± 7

Categorical variables are represented as n (%); continuous variables are represented as mean (standard deviation) unless noted as median (Q1=quartile 1, Q3=quartile 3). Abbreviations: CNS = central nervous system; NT-proBNP= N-terminal prohormone of brain natriuretic peptide; hs-troponin T = high sensitivity cardiac troponin T; ADI = area deprivation index; BMI = body mass index; sCr = serum creatinine; eGFR = estimated glomerular filtration rate, mmHg = millimeters of mercury; Q1 = quartile one; Q3 = quartile three; FVC = forced vital capacity; FVC/FEV1 = Forced vital capacity to Forced expiratory volume in 1 second ratio.

Latent Class Analysis:

We fit LCA models with 10 dichotomous fatigue items with 2-5 classes. Standardized bivariate residuals showed associations between 2 items in the PROMIS fatigue (general fatigue) scale (items 4 and 5). The odds ratio for the association between item 4 and 5 was 7.53 (95% CI: 4.14, 13.70). We modeled this association in the LCA. Using bootstrapped likelihood ratio tests and BIC, a three-class model fit significantly better than a two-class model (BIC= 8593.75, −2LL difference: 352.3, p<0.001), and a four-class model fit significantly better than a three-class model (BIC = 8496.619, −2LL difference: 347.7, p<0.001). Based on bootstrapped likelihood ratio tests, the five-class model did fit significantly better than a four-class model (−2LL difference: −73.8, p <0.001), however, the four-class model had the lowest BIC (4-class = 8496.619 vs 5-class = 8533.543). Based on this, a four-class model was selected. The entropy of the four-class model was 0.72 which indicates moderately high confidence in most likely class membership classification. The model fit statistics for the 2, 3, 4, 5, and 6-class models are shown in Supplemental Table S1.

Supplemental Table S2 shows the estimated class prevalence and conditional probabilities for each fatigue item on the PROMIS fatigue scale and the MRC breathlessness scale. Figure 1 shows the item conditional probabilities by class. Class 1 is expected to include the smallest proportion (18%) of the sample and had high conditional probabilities on almost all general and exertional fatigue items. We interpreted this class as the “high general/high exertional fatigue” subtype. Class 2 is expected to include 27% of the sample and had moderate to high conditional probabilities on the general fatigue items, and low conditional probabilities on the exertional fatigue items. We interpreted this class as the “high general/low exertional fatigue” subtype. Class 3 is expected to include 20% of the sample and had moderate conditional probabilities on the majority of general and exertional fatigue items. We interpreted this class as the “moderate general/moderate exertional fatigue” subtype. Class 4 is expected to include the largest proportion of the sample (35%) and had low conditional probabilities on all general and exertional fatigue items. We interpreted this class as the “low/no general & exertional fatigue” subtype.

Figure 1. Item conditional probabilities of fatigue survey items by latent class.

Solid lines indicate item conditional probabilities. Dotted lines indicate 95% confidence intervals for these estimates. PROMIS indicates an item from the Patient-Reported Outcomes Measurement Information System (PROMIS) fatigue scale which we used to measure general fatigue. MRC indicates an item from the Medical Research Council (MRC) Breathlessness scale which we used to measure exertional fatigue. Wording of the scale questions for both the PROMIS fatigue scale and the MRC Breathlessness scale are described in supplemental table 1.

Demographic and Clinical Characteristics by Fatigue Subtype

Table 1 shows the demographic and clinical characteristics of all participants at visit 5 with prevalent heart failure. Supplemental Table S3 shows the estimated sample size of each fatigue subtype based on the most likely class membership, as well as aggregate demographic and clinical characteristics for each subtype. Table 2 shows univariate comparisons of demographic and clinical characteristics for each subtype compared to the reference subtype (class 4 – the low/no general & exertional fatigue subtype). Figure 2 depicts the demographic and clinical characteristic profiles of each fatigue subtype based on the univariate odds ratios for characteristics found to be statistically significant for at least one subtype in Table 2. These radar graphs display the univariate odds ratios of subtype membership by characteristic compared to the low/no general & exertional fatigue subtype (reference subtype).

Table 2.

Comparing demographic and clinical characteristics across fatigue subtypes.

	High General/High Exertional Fatigue	High General/Low Exertional Fatigue	Moderate General/ Moderate Exertional Fatigue	Low/No Fatigue
Estimated sample size	n=175	n=275	n=218	n=397
Age	1.01 (0.96, 1.00)	1.05 (1.01, 1.10)	1.02 (0.97, 1.00)	REF
Sex	5.81 (3.20, 10.53)	2.00 (1.27, 3.17)	0.88 (0.48, 1.61)
Race	3.36 (2.07, 5.47)	0.79 (0.47, 1.32)	0.41 (0.17, 1.00)
Education:
Less than High School	5.49 (3.31, 9.12)	1.76 (0.97, 3.16)	1.22 (0.53, 2.83)
High School/Vocational	0.79 (0.49, 1.28)	1.08 (0.68, 1.72)	1.38 (0.79, 2.40)
College	0.40 (0.21, 0.76)	0.85 (0.52, 1.39)	0.68 (0.36, 1.28)
Graduate/Professional	0.06 (0.002, 1.61)	0.46 (0.22, 0.97)	0.76 (0.37, 1.58)
Area Deprivation Index (OR for one SD increase in ADI)	2.03 (1.58, 2.62)	1.04 (0.83, 1.32)	0.78 (0.58, 1.06)
Smoking Status:
Current Smoker	0.96 (0.32, 2.88)	0.28 (0.03, 2.42)	6.84 (0.73, 64.1)
Former Smoker	0.60 (0.30, 1.22)	0.76 (0.37, 1.55)	3.95 (0.49, 32.0)
Never Smoker	1.09 (0.54, 2.22)	0.90 (0.42, 1.92)	2.67 (0.31, 23.2)
Systolic blood pressure	1.01 (0.99, 1.02)	1.01 (1.00, 1.02)	0.99 (0.98, 1.01)
Body Mass Index (OR for each 1 unit increase in BMI)	1.20 (1.13, 1.27)	1.00 (0.95, 1.09)	1.09 (1.02, 1.16)
Comorbidities:
Diabetes	4.40 (2.61, 7.41)	1.20 (0.74, 1.94)	1.61 (0.92, 2.84)
Atrial fibrillation	1.48 (0.82, 2.47)	0.97 (0.52, 1.80)	1.93 (1.04, 3.60)
Coronary heart disease	0.82 (0.51, 1.30)	0.80 (0.50, 1.28)	1.96 (1.10, 3.51)
Medications:
Beta Blockers	0.97 (0.61, 1.54)	0.86 (0.54, 1.36)	3.23 (1.41, 7.39)
Loop Diuretics	2.85 (1.77, 4.58)	1.00 (0.57, 1.78)	1.84 (1.02, 3.34)
CNS Depressants	2.75 (1.55, 4.88)	1.56 (0.78, 3.10)	0.88 (0.30, 2.64)
Kidney function(sCr & cystatin C eGFR, ml/min/1.73m2)	0.98 (0.97, 1.00)	0.99 (0.98, 1.00)	0.98 (0.97, 1.00)
Cardiac Biomarkers
NT-proBNP (>= 450 pg/mL)	1.24 (0.75, 2.05)	1.13 (0.69, 1.86)	1.71 (0.99, 3.03)
hs-Troponin T (>= 14 ng/L)	1.74 (1.07, 2.83)	1.04 (0.66, 1.65)	2.85 (1.51, 5.37)
Pulmonary function:
% predicted FVC	0.98 (0.97, 1.00)	1.01 (1.00, 1.01)	0.99 (0.98, 1.00)
% predicted FVC/FEV1	0.99 (0.98, 1.01)	1.01 (1.00, 1.03)	0.99 (0.98, 1.01)
Depressive Symptoms	1.77 (1.53, 2.10)	1.35 (1.20, 1.52)	1.26 (1.12, 1.42)

Odds ratios (OR) are represented as OR (95% CI) and are from multinomial latent class regression accounting for most likely class membership classification error. Sample size is estimated based on most likely subtype membership. Abbreviations: OR = odds ratio; CNS = central nervous system; NT-proBNP= N-terminal prohormone of brain natriuretic peptide; hs-troponin T = high sensitivity cardiac troponin T; ADI = area deprivation index; BMI = body mass index; sCr = serum creatinine; eGFR = estimated glomerular filtration rate, mmHg = millimeters of mercury; FVC = forced vital capacity; FVC/FEV1 = Forced vital capacity to Forced expiratory volume in 1 second ratio.

Figure 2. Demographic and clinical characteristic profiles of general and exertional fatigue subtypes compared to the low/no general & exertional fatigue subtype (class 4).

Numbers indicate odds ratios of fatigue subtype membership compared to the reference subtype (low/no general & exertional fatigue subtype). Odds ratios go from smallest to largest from the center to the outermost circle. Red circles indicate variables for which odds ratios were statistically significant (95% CI’s do not contain 1.00). Odds ratios for continuous variables (age, BMI, depressive symptoms, and hs-Troponin T) indicate odds ratio for one unit increase in the variable. Odds ratios for all other variables indicate increased odds for endorsement of that characteristic. These odds ratios are based on most likely subtype membership, accounting for the probability of misclassification. Abbreviations: CNS = central nervous system; hs-Troponin T = high sensitivity troponin T; CHD = coronary heart disease; afib = atrial fibrillation; BMI = body mass index; ADI = area deprivation index; HS = high school.

High General & Exertional Fatigue Subtype – Cardiometabolic Risk & Sociodemographic Factors

Several sociodemographic factors (female sex, Black race, a less than high school education level, and increased area deprivation index), cardiometabolic risk factors (prevalent diabetes, higher BMI, increased hs-Troponin T), medications (CNS depressants, loop diuretics), and increased depressive symptoms were associated with greater odds of being in the high general/high exertional fatigue subtype vs. the low/no general & exertional fatigue subtype. Among the covariates, Female sex (OR = 5.8), Black race (OR = 3.4), a less than high school education level (OR = 5.5), and prevalent diabetes (OR = 4.4) showed the strongest relationships with membership in this subtype.

High General & Low Exertional Fatigue Subtype – Older Women with Depressive Symptoms

Female sex was also associated with increased odds of membership in the high general/low exertional fatigue subtype compared to the low/no general & exertional fatigue subtype (OR = 2.0), although this association was not as strong as it was for the high general/high exertional fatigue subtype. Other variables significantly associated with membership in the high general/low exertional fatigue subtype were increased age and worse depressive symptoms.

Moderate General & Moderate Exertional Fatigue Subtype – Arrhythmia & Atherosclerosis

Prevalent coronary heart disease, atrial fibrillation, and increased hs-troponin T levels were associated with increased odds of membership in the the moderate general/moderateexertional fatigue subtype. Other variables significantly associated with membership in this subtype were the use of beta blockers, and loop diuretics, as well as higher depressive symptoms compared to the low/no general & exertional fatigue subtype. Increased hs-troponin T (OR = 2.9) and the use of beta blockers (OR = 3.2) showed the strongest relationship with membership in this subtype.

Discussion

Using latent class analysis, we identified four unique subtypes of general and exertional fatigue in those with prevalent HF in the ARIC study including: 1) high general/high exertional fatigue, 2) high general/low exertional fatigue, 3) moderate general/exertional fatigue, and 4) low/no general & exertional fatigue subtypes. Interestingly, exertional fatigue did not appear to occur alone in this sample and was only seen in combination with general fatigue. Most classes had similar levels of general and exertional fatigue; however, one discordant group was identified - the group with high general fatigue but low exertional fatigue. There is need to further understand the clinical and prognostic significance of variable degrees of exertional fatigue among patients with high general fatigue. Nonetheless, our findings underscore the importance of considering both general and exertional components of fatigue to fully characterize symptomatology among HF patients. We additionally found significant clinical and demographic differences across fatigue subtypes.

To our knowledge, this is the first study to date to perform a latent class analysis on both general and exertional fatigue survey items to identify unique fatigue subtypes in those with prevalent HF. This work, defining unique subtypes of fatigue, may provide enhanced understanding of the symptoms experienced in persons with HF and related clinical and demographic correlates. Further, this study adds to our limited understanding of the factors linked to general fatigue, and the co-occurrence of general and exertional fatigue. It may also improve our ability to assess and identify the most distressing and prognostically important symptoms and extract potential targets for intervention with the goal of improving overall quality of life.

Female sex was the strongest predictor of membership in fatigue subtypes with high levels of general fatigue, which included the high general/high exertional fatigue subtype and the high general/low exertional fatigue subtype. Strong associations between female sex and higher prevalence/severity of general fatigue have been shown in previous work. ^{5,41,14,42,43,44,45} However, the mechanisms driving this association - whether biological, psychosocial, or both - are still unclear and provide a unique opportunity for future study.

Key social determinants of health were associated with the high general/high exertional fatigue subtype including Black race, area deprivation index, and lower education level. Black individuals in the US are disproportionately affected by HF and its adverse outcomes.⁴⁶ Area deprivation index (ADI) is reflective of a diversity of unmet social needs that may influence access to resources needed for disease management and health promoting behaviors. In other studies, worse ADI has been linked to an increased risk of adverse outcomes in cardiovascular disease.^47,48 A lower level of education has been linked to lower health literacy⁴⁹ and poorer HF self-care^8,50, which can influence HF management and overall health over the life course. Our analysis adds to the literature by highlighting symptom manifestation and burden disparities related to these key social determinants of health.

The use of beta blockers was associated with moderate/low severity of both general and exertional fatigue, while loop diuretics were most prevalent in the high severity fatigue subtypes. Previous studies have shown conflicting associations between HF medications and prevalence/severity of fatigue^8,13,14 and the associations seen in our analysis may be a reflection of participants’ HF severity rather than effects of the medications themselves.

A notable clinical finding of our analysis was the association of multiple measures of cardiometabolic disease with the highest severity fatigue subtype. Those with higher BMI and diabetes were more likely to be in the high general/high exertional fatigue class. This may suggest that the pathophysiologic pathways linked to cardiometabolic disease may be related to HF severity factors that influence the symptom experience. Further, this is a unique finding that has not been found previously in symptom cluster analyses. Increased BMI and diabetes have been identified as key characteristics of a high adverse cardiovascular outcome risk phenogroup of HFpEF in a recent analysis of the Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Agonist (TOPCAT) study.⁵¹ This high general/high exertional fatigue subtype may represent a unique subgroup of the HF population, like that seen in the TOPCAT study, who could potentially be at high risk for adverse clinical and patient reported outcomes.⁵¹

Atherosclerotic disease and atrial fibrillation were associated with membership in the moderate general/moderate exertional fatigue subtype. This is an interesting finding suggesting that co-occurrence of general and exertional fatigue at differing severity may be associated with different risk factors. The mechanisms of the relationship between fatigue type and coronary heart disease are still unclear. However, fatigue is a common symptom of atrial fibrillation.⁵²

It is important to note that increased depressive symptoms were associated with membership in all fatigue classes, when compared to the low/no general & exertional fatigue class. General fatigue in HF has been shown in numerous observational studies to be associated with depression^4,5,7,53,54, and depression is frequently implicated in impaired HF self-care.^50,55,56

These results have important clinical implications. First, our results suggest that there are different subtypes of general and exertional fatigue that persons with HF experience. These subtypes, if reproduceable in diverse HF samples, may provide clinical utility in identifying HF patients with a greater likelihood of increased morbidity, or higher psychosocial needs that need to be addressed. By considering the patient’s subjective experience of fatigue (including general and exertional components), we may be able to better foster shared decision making related to treatments and interventions to improve the HF patient’s quality of life as a whole. Second, these subtypes of fatigue are associated with key demographic and clinical characteristics that, with further investigation, may provide mechanistic insight into the drivers of fatigue in this population and elucidate novel targets for tailored intervention based on fatigue subtype.

Strengths/limitations

Strengths

The identification of latent classes (or subtypes) of general and exertional fatigue is advantageous because it groups individuals based on their pattern of responses to the items on the fatigue scales. We used data from the ARIC study, which is a large and representative, community-based cohort study with rigorous risk characterization and adjudication protocols for HF. We also had a large sample size, improving the statistical power of our inferences.

Limitations

There is always the potential for residual confounding due to the observational nature of this study. This analysis does not include data on cardiac structural and functional factors that may influence the fatigue experience, which is planned for a future analysis. We operationalized exertional fatigue using the MRC breathlessness scale which measures exertional dyspnea. It is possible that this scale, the only exertional fatigue scale available in ARIC, does not fully capture the construct of exertional fatigue. There is also a lack of some important variables in ARIC that are theoretically related to the constructs of general and exertional fatigue (e.g. sleep quality). Additionally, in ARIC, race is highly correlated with study site, with the majority of black participants being from the Jackson, MS population center, limiting our ability to separate race from geographic setting. Finally, future studies should examine the association of these fatigue subtypes with clinical outcomes in HF patients, accounting for covariates and markers of HF severity.

Conclusions

Among ambulatory patients with prevalent HF in a community-based study, we used latent class analysis to identify 4 unique subtypes (classes) of general and exertional fatigue. Furthermore, we found significant differences in several demographic and clinical characteristics across those classes including sex, race, education level, BMI, diabetes, atrial fibrillation, coronary heart disease, key HF medications, and depressive symptoms. Understanding the different subtypes of fatigue in HF may improve our ability to elucidate pathways underlying the symptom experience in HF patients, and to individually tailor assessment and treatment of symptoms in HF patients with the goal of improving overall quality of life.

What is Known:

• Fatigue in heart failure is prevalent and distressing for patients.

• Fatigue has general and exertional components, but current literature primarily examines fatigue as a single construct.

• We have a poor understanding of the factors associated with general and exertional fatigue, and their co-occurrence.

What this Study Adds:

• There are unique subtypes of general and exertional fatigue in a sample of ambulatory adults with prevalent heart failure in the ARIC study.

• Individuals with heart failure within different subtypes of general and exertional fatigue have distinct sociodemographic and clinical profiles.

• These results describe the spectrum of fatigue symptomatology among ambulatory heart failure patients and can help elucidate pathways underlying differential fatigue experiences in heart failure.

Non-standard Abbreviations and Acronyms

HF

Heart failure

ARIC

Atherosclerosis Risk in Communities

PROMIS

Patient Reported Outcomes Measurement Information System

MRC

Medical Research Council

ADI

Area deprivation index

CHD

Coronary heart disease

Afib

Atrial fibrillation

CNS

Central nervous system

LCA

Latent class analysis

eGFR

Estimated glomerular filtration rate

hs-troponin T

High sensitivity cardiac troponin T

NT-proBNP

N-terminal prohormone of brain natriuretic peptide

BMI

Body mass index

FVC

Forced vital capacity

FVC/FEV1

Forced vital capacity to Forced expiratory volume in 1 second ratio

sCr

Serum creatinine

OR

Odds ratio

SD

Standard deviation

Q1

Quartile one

Q3

Quartile three

MI

Myocardial infarction

ECG

Electrocardiogram