Trajectory of Cognitive Function After Incident Heart Failure.
Feb 18, 2025DOI: 10.1161/circheartfailure.124.011837Journal ArticleAHA
Written by: Supriya ShoreHanyu LiMin ZhangRachael T WhitneyAlden L GrossAnkeet S BhattBrahmajee K NallamothuBruno GiordaniEmily M BriceñoJeremy B SussmanJose GutierrezKristine YaffeMichael E GriswoldMichelle C JohansenOscar L LopezRebecca F GottesmanStephen SidneySusan R HeckbertTatjana RundekTimothy M HughesW T LongstrethDeborah A Levine
Find these authors on Roon
AB
Dr. Ankeet Bhatt
General Cardiology · Kaiser Permanente
BN
Dr. Brahmajee Nallamothu
Interventional Cardiology · UMHS
INTRODUCTION
Understanding and reducing the cardiovascular contributions to cognitive decline and dementia is critical because dementia is common, costly, and morbid. Heart failure (HF) affects 6.7 million people worldwide and imposes a high burden of disabling symptoms on afflicted individuals, associated caregivers, healthcare systems, and society.1, 2 Management of HF is complex, relying heavily on a patient’s cognitive ability due to the need for significant care coordination, lifestyle changes, medication compliance, and to process multi-faceted information about available therapies.3–5 While cognitive dysfunction has been linked to several cardiovascular disorders like hypertension, coronary artery disease, and stroke,6–9 how cognition changes after incident HF remains poorly understood.Prior studies note a high prevalence of cognitive impairment in HF patients.10, 11 However, the magnitude of change in cognition at the time of incident HF diagnosis is unclear. It is also unclear whether HF survivors have a faster rate of cognitive decline (slope change) over the years following HF onset compared with their pre-HF rate of cognitive decline after accounting for initial decreases in cognition around the time of the incident diagnosis (intercept change). Studies describing changes in cognition after HF have been unable to measure actual changes associated with HF because they lacked repeated measures of cognition before and after HF onset.10–12 Additionally, studies have not simultaneously measured the initial decrease in cognition at the time of HF diagnosis and the change in the rate of cognitive decline over the years after HF. As the number of people with HF is expected to increase,13 identifying how cognition changes during the disease course has significant implications for clinical practice and healthcare policy.2We conducted a pooled study using six population-based cohorts with repeated cognitive testing, data on incident HF, and longitudinal follow-up to determine the changes in cognition after incident HF. We hypothesized that incident HF diagnosis is associated with an initial decrease in cognition and a faster rate of cognitive decline after accounting for pre-HF cognitive trajectories. If an association was found, we also aimed to assess if the association between HF and cognitive change varied by age, sex, and race.
METHODS
This cohort study included individual participant data from 6 well-characterized, US prospective cohorts and study years between 1971 and 2019: Atherosclerosis Risk in Communities Study (ARIC), Coronary Artery Risk Development in Young Adults Study (CARDIA), Cardiovascular Health Study (CHS), Framingham Offspring Study (FOS), Multi-Ethnic Study of Atherosclerosis (MESA) and Northern Manhattan Study (NOMAS). These cohorts were chosen because they have physician-adjudicated HF cases, repeated measures of cognition and blood pressure (BP), and geographic, age, racial, and ethnic diversity.14–18
The institutional review board at the University of Michigan approved this study. Participating institutions approved the cohort studies. All participants gave written informed consent. Data will be shared according to the procedures of the cohort studies. All analyses were pre-registered and approved by each cohort’s leadership teams prior to initiation.
We included all individuals with at least one cognitive assessment during cohort follow-up and without a preceding history of HF, stroke, or dementia (Figure 1). We excluded individuals with HF, stroke, or dementia at or before the first cognitive assessment and those lacking complete data on covariates of interest. Follow-up was censored after an incident stroke if it occurred during follow-up, as stroke is associated with cognitive decline.6
HF was captured prospectively during cohort follow-up. HF diagnoses were adjudicated by expert physicians in CARDIA, CHS, FOS, MESA, and for participants in ARIC after 2005 using standardized protocols with medical record review (see Table 1 for details).19–23 For ARIC participants prior to 2005, HF diagnosis was classified as a hospitalization with an International Classification of Disease diagnosis code indicating HF in any position. For NOMAS, HF diagnosis was based on participant self-report,24 which also has high specificity.25 Measures of left ventricular ejection fraction are not available in all cohorts and were not used in our study.
The primary outcome of interest was global cognition (GC). Secondary outcomes were executive function and memory. Executive function comprises cognitive processes that include planning, decision making, problem solving, and attention. Validated cognitive tests, consistent with the Vascular Cognitive Impairment Harmonization Standards,26 were administered in the six cohorts by trained staff either in person or over telephone.27
To make inferences about cognitive domains instead of individual cognitive test items and to account for different cognitive tests administered across cohorts, we co-calibrated available cognitive test items into factors representing GC (a composite of executive function, memory, general mental status, language, motor and visuospatial ability), executive function, and memory using all available cognitive information.28 These harmonized measures for cognitive function facilitate assessment of cognitive domains rather than individual tests.7–9 In a pre-statistical harmonization phase, neuropsychologists identified 153 test items from 34 cognitive instruments across the cohorts and determined shared and unique items across cohorts. Neuropsychologists assigned all available cognitive test items into domains. Individual tests in each domain are listed in Table S1. Using item response theory methods, each test item was weighted based on its correlation with other items and empirically assigned a relative location along the latent cognitive trait corresponding to its estimated difficulty.29, 30 This method has been previously recommended.31 We computed factor scores for each domain, which were set to a t-score metric (mean [SD], 50[10]) at a participant’s first cognitive assessment; a 1-point difference thus represented a 0.1-SD difference in the distribution of cognition across the 6 cohorts.28 Higher cognitive scores indicate better cognitive function.
We included covariates that could confound the association between HF and cognition and were available in all cohorts before the first cognitive assessment.7, 22 We used covariate values measured immediately before or on the same day as the first cognitive assessment. All included covariates were harmonized by choosing common response categories for categorical variables and converting measurements to standard units for continuous variables, as done previously.7–9
Demographics included age, race/ethnicity, sex, education, and study cohort. Race/ethnicity was self-reported and categorized as Black, White, and Hispanic (any race) by the cohorts.28 All other reported races were excluded as they formed a small minority in the context of this larger study (n=953; 2.4%). Clinical variables included alcohol drinks weekly, active cigarette smoking, any physical activity, body mass index, waist circumference, history of myocardial infarction (MI), history of atrial fibrillation (AF), fasting glucose, low density cholesterol, glomerular filtration rate, use of antihypertensive medications and cumulative mean systolic blood pressure and baseline cognitive slope. We included BMI and waist circumference as covariates in our model as the combination offers a comprehensive measure of metabolic health, with BMI measuring overall body fat and waist circumference body fat distribution. Accordingly, the measures are complementary, and existing studies demonstrate that these covariates have independent associations with cognitive function.32 Expert physicians defined the presence of MI and AF at any point from cohort enrollment until the first cognitive assessment based on published guidelines, medical record review, and standardized protocols. Current antihypertensive medication use was assessed based on medication bottles and self-report. Cumulative mean systolic blood pressure was defined as a mean of all systolic blood pressures checked at in-person visits using standard protocols and equipment.8 All continuous variables were centered at the median.
We compared baseline characteristics between patients with and without incident HF diagnosis during follow-up using 2-sample t-tests for continuous variables and χ2 tests for categorical variables. Each outcome measuring cognition was assessed as a continuous variable. Incident HF diagnosis was treated as a time-dependent variable.
A linear mixed-effects model was used to estimate the association between HF and cognitive change, both around the time of incident HF diagnosis and in the following years, adjusted for the covariates and baseline (pre-HF) cognitive slope. It included random effects for intercept and slope to accommodate the correlation of cognitive measures within participants over time and to allow participant-specific rates of cognitive change. We included interaction terms between time and age, sex, and race/ethnicity based on prior research.7, 8 The first cognitive assessment is considered as time 0 for this study.
To estimate the association between incident HF diagnosis and cognition, the model included a time-varying incident HF variable (changed from 0 to 1 on the date of the first incident HF diagnosis). In addition, a spline term for years after HF was added to determine if incident HF was associated with a faster rate of cognitive decline in the years following the diagnosis. For individuals with HF, this variable was 0 before their first HF diagnosis. The coefficient for this variable represents the association of incident HF and the rate of post-HF decline in cognitive function. This study considers the first cognitive assessment after incident HF diagnosis the initial change around incident HF diagnosis. We found no evidence of non-linear effects of cognition on time and the effect of HF on time.
Statistical significance for all analyses was set as P <0.05 (2-sided). All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) and R (version 4.2.1).
In sensitivity analyses, to account for attrition bias, we repeated analyses with individuals who had ≥2 cognitive assessments. We also repeated analyses within cohorts to assess for heterogeneity in associations between HF and cognitive decline.
In secondary analyses, to explore whether the association of incident HF diagnosis with cognitive decline differed based on the age of HF onset, we added the following interaction terms of incident HF with age as fixed effects: incident HFage at HF onset and time elapsed since incident HF diagnosisage at HF onset. Similar analyses were performed to assess for effect modification by sex and race.
We performed additional sensitivity analyses using MI and AF as time-varying covariates to assess the impact of incident MI and AF instead of treating them as fixed covariates. These additional analyses were proposed to account for any additional changes in exposure to these variables after the first cognitive assessment, as they have been previously associated with a cognitive decline and are on the causal pathway with HF and cognitive decline.9, 33
Results
Our study cohort included 29,614 participants with a mean age of 61±10 years at first cognitive assessment, 55% were female, 70 % were White, 22% were Black, and 8% were Hispanic. Overall, 1,407 (5%) had an incident HF diagnosis over a median follow-up duration of 6.6 (Q1-Q3: 5.0–19.8) years. All participants had GC assessed. However, executive function and memory assessments were introduced later and performed on 26,417 (89%) and 19,176 (65%) participants without prior HF diagnosis. Median (Q1-Q3) number of assessments were 3(2–5) for GC, 2(2–4) for executive function and 2(2–3) for memory.
Table 2 presents the baseline characteristics of study participants by incident HF diagnosis status. Participants who developed HF were older (68 vs. 61 years) at first cognitive assessment, less likely to identify as Black (18% vs. 22%), and had lower levels of education. Participants who developed HF had a higher incidence of MI (11% vs. 4%) and AF (5% vs. 2%). Compared with participants who did not develop HF, participants with HF had a lower glomerular filtration rate, mean (SD) [68 (17) vs. 75 (19) mL/min/1.73m2], and a higher cumulative mean systolic blood pressure [148 (21) vs. 137 (19) mmHg] despite higher rates of antihypertensive medication use (49% vs. 31%). Participants with incident HF diagnosis, compared with participants without HF, had a greater median (Q1-Q3) number of assessments for GC [8 (4–11) vs. 3 (2–5)], executive function [6 (4–9) vs. 2 (2–3)] and memory [4(3–4) vs. 2(2–3)]. Participants who developed HF had lower scores for GC (−1. 6 points; 95% CI −2.0 to −1.1) and executive function (−2.3 points; 95% CI −2.9 to −1.8) but similar scores for memory (0.0 points; 95% CI −1.6 to +0.6) at the study visit immediately prior to incident HF diagnosis compared to those who did not develop HF.
The median time (Q1-Q3) between the last pre-HF cognitive assessment and incident HF diagnosis was 2.6 (1.0–6.2) years for GC, 2.2 (1.0–6.6) years for executive function, and 6.1 (5.1–15.1) years for memory. Among participants who had an incident HF diagnosis, the proportion of patients with GC, executive function, and memory assessments were 57.5% (n= 809), 48.9% (n=393), and 29.7% (n=89), respectively. The median time (Q1-Q3) between incident HF diagnosis and first post-HF cognitive assessment was 1.0 (0.9–2.2) years for GC, 1.0 (0.9–2.1) years for executive function, and 5.0 (4.3–5.4) years for memory.
An incident diagnosis of HF was associated with initial decreases in GC (−1.1 points; 95% CI −1.4 to −0.8) and executive function (−0.6 points; 95% CI −1.0 to −0.3) after adjusting for pre-HF slope and covariates (Table 3). An incident diagnosis of HF was also associated with a faster decline in GC (−0.1 points per year; 95% CI −0.2 to −0.1) and executive function (−0.2 points per year; 95% CI −0.2 to −0.1) after controlling for pre-HF slope and the initial decrease in cognition around the time of the HF diagnosis. While the change in memory with incident HF diagnosis did not reach statistical significance, it showed a similar trend with an initial decline of −0.5 points (95% CI −1.4 to +0.3) and a slope of −0.1 points per year (95% CI −0.3 to 0.0). After selecting the final, parsimonious model, we calculated participant-specific (conditional) predicted values for cognitive scores for a 70-year-old White male conditional on his being diagnosed with incident HF during the follow-up period (at year 9; Figure 2). The average values of covariates at baseline were college degree or higher education, non-smoker, no alcohol use, no history of AF, no history of MI, some physical activity, no anti-hypertensive medication use, body mass index of 27.1 kg/m2, waist circumference of 95.2 cm, fasting blood glucose of 97 mg/dL, low density lipoprotein cholesterol of 124 mg/dL, glomerular filtration rate 73 ml/min/m2, and cumulative mean systolic blood pressure of 136 mmHg.
Table 4 shows the results of our interaction analyses for change in cognitive function with incident HF diagnosis by age of HF onset, sex, and race. All analyses were adjusted for pre-specified covariates detailed above and baseline (pre-HF) cognitive slope. Older age at HF onset was associated with a larger initial decrease in GC with incident HF diagnosis (−1.5 points per 10-year older age of HF onset; 95% CI −2.0 to −1.0). However, there was no statistically significant difference in change in GC slope after incident HF diagnosis by age (0.0 points per 10 year older age of HF onset; 95% CI −0.0 to +0.1). While older age at HF onset did not modify change in executive function around incident HF diagnosis (−0.3 points per 10-year older age of HF onset; 95% CI −1.2 to + 0.5), the change in executive function slope after incident HF diagnosis was slower with increasing age of HF onset (+0.4 points per year per 10-year older age of HF onset; 95% CI +0.2 to +0.6). We found no evidence that age modified the association between HF and memory at either incident diagnosis(p=0.06) or after (p=0.29).
We observed racial and ethnic differences in cognitive trajectory after HF diagnosis when holding sex constant. Compared with White participants, Black participants had a smaller initial decrease in GC and executive function around the time of incident HF diagnosis (+0.8 points; 95% CI 0.0 to + 1.6 & +1.6 points; 95% CI +0.7 to +2.4 respectively) but not memory (+0.6 points; 95% CI −1.4 to +2.5). However, long-term cognitive declines after incident HF diagnosis did not show a statistically significant difference between Black and White participants for any cognitive outcome (p=0.3 for GC, p=1.0 for executive function, and p=0.5 for memory). We found no statistically significant difference in the cognitive changes associated with incident HF diagnosis between Hispanic (any race) and non-Hispanic White participants.
Females experienced a larger initial decrease in GC after incident HF diagnosis (−0.7 points; 95% CI −1.3 to −0.2), but the GC slope did not show statistically significant difference by sex after incident HF diagnosis (0.0 points per year; 95% CI −0.1 to +0.1). The changes in executive function and memory after incident HF diagnosis did not differ by sex (p=0.2 and 0.6 for initial change and p=0.8 and 0.7 for slope change, respectively).
Results were similar in analyses including participants with ≥ 2 cognitive assessments (n=24,090 for GC, n=20,533 for executive function, and n=14,649 for memory; Table S2). Table S3 shows the association between HF and GC across cohorts. The directionality of initial change in cognition around the time of incident HF diagnosis and change in slope of cognition after incident HF diagnosis was similar across cohorts except in two cohorts, including NOMAS and CARDIA. These exceptions are likely due to loss of power as NOMAS is the smallest cohort, and the incidence of HF was lowest in CARDIA, comprised of young adults. Results were similar in sensitivity analysis, including history of MI and AF as time-varying covariates (Table S4). Incident MI and AF were associated with larger initial cognitive decline in HF patients compared to those without.
Discussion
In a pooled analysis of 6 large, US prospective cohort studies that performed serial cognitive assessments on 29,614 participants, we observed that incident HF diagnosis was associated with initial decreases in GC and executive function around the time of the diagnosis as well as faster declines in GC and executive function over the years after the diagnosis adjusting for pre-HF cognitive trajectories and known determinants of cognitive function. We did not find significant changes in memory associated with incident HF diagnosis. Results persisted after further adjustment for MI and AF. In clinical terms, our findings suggest that individuals diagnosed with incident HF experience global cognitive aging equivalent to 10 years within 7 years after HF diagnosis, compared to those without HF. We also observed larger initial decreases in GC around the time of incident HF diagnosis associated with older age, female sex, and self-identified White race.
Existing studies report a high prevalence of cognitive impairment in HF. Fewer studies have looked at the longitudinal trajectory in cognition following HF diagnosis, and some report stable neuropsychological testing in contrast to our study.34, 35 However, these studies are much smaller (n=77–279 HF patients) with shorter duration of follow-up (n=6–24 months) and use of smaller neuropsychological batteries that limit power.10 Previous studies have classified clinically meaningful cognitive decline as a decrease in cognitive function of ≥ 0.5 standard deviations from baseline cognitive scores.36, 37 If the observed HF-related changes in cognitive function in our study are confirmed and causal, cohort-representative adults who develop HF could experience a 0.5 SD decline in global cognition scores 4.6 years earlier, and in executive function 3.8 years earlier, than those without HF following HF diagnosis.
In addition to showing an association between incident HF diagnosis and persistent cognitive decline after, our findings also suggest larger initial decreases in cognition with incident HF diagnosis in older, White, and female adults. The larger initial decrease in cognition with older age likely reflects the loss of cognitive reserve and identifies subgroups at higher risk for cognitive impairment who might need closer monitoring and more support in managing HF. The larger initial decreases in GC and executive function after HF seen in White participants likely reflect the higher pre-HF cognitive performance of this subpopulation38 with a greater room to decrease. However, sex differences in cognitive functioning are complex. Females likely have a faster cognitive decline with HF due to differences in sex hormones, greater microvascular disease, and inflammation compared to males.7, 39, 40
Incident HF could be associated with cognitive decline through several mechanisms. We examined the best-established mechanisms, vascular risk factors, and comorbidities such as hypertension, MI, and AF, but they failed to explain our findings. Other hypotheses include HF causing systemic inflammation and endothelial dysfunction, which have been implicated in the causal pathway for cognitive impairment.41, 42 In addition, HF may cause cognitive decline due to decreased cerebral perfusion.43 We likely did not see a statistically significant difference in memory as we were underpowered, as reflected by the wide confidence intervals in our estimates. Additional studies need to evaluate mechanisms behind the cognitive decline associated with HF to identify modifiable targets.
Our study’s findings have several implications for clinical management and healthcare policy. Reducing HF-associated burden is a priority.5 Successfully achieving this goal relies heavily on a patient’s cognitive ability with the need to follow a specific diet, monitor symptoms, adhere to medications – with an average HF patient taking at least 10 different medications – and coordinate care across multiple sub-specialties.44 In addition, progressive cognitive decline is associated with depression,45 dementia,46 and functional decline,47 causing greater caregiver burden and institutionalization risk. Cognitive monitoring for older adults with HF could help identify individuals at risk for these adverse events. Such screening may also facilitate interventions such as avoiding medications contributing to cognitive dysfunction and using non-pharmacological measures such as exercise training and cognitive rehabilitation that may halt or slow cognitive decline.48, 49 HF treatment for older adults may need to be individualized to consider associated cognitive decline. This would involve home care programs to monitor these patients to prevent hospitalizations closely, interventions to help with medication adherence, and modifying decision aids for HF treatments for patients with cognitive impairment.
Our study has several strengths. First, we used harmonized measures of cognition performed serially in participants, allowing us to control for pre-HF cognitive trajectory and describe the entire trajectory of cognitive function for HF patients. Pooling cohort studies enabled us to capture diverse populations with well-described socio-demographic characteristics and to generalize more broadly than earlier studies that were more limited in scope. Finally, HF diagnosis in our study used clinical methods and did not rely on administrative or claims measures.
Our findings should be interpreted in the context of several limitations. HF typically has an insidious onset, and diagnosis may be delayed until the progression of symptoms or a severe stage. Enrolled participants are likely to have delayed their cohort visit until they felt better following a diagnosis of HF that may have delayed capturing early changes in cognition. Assuming that participants’ postmortem cognitive data are missing at random might lead to immortal cohort bias. Similarly, individuals with more severe cognitive impairment or worse disease severity are more likely to drop out with fewer cognitive assessments. However, these factors would likely bias our findings toward underestimating cognitive decline. In addition, we used linear mixed effects models that are robust to sparse data, using all available cognitive measurements prior to HF diagnosis with additional sensitivity analyses in individuals with 2 or more cognitive assessments that demonstrated consistent results. We were unable to control for possible confounders such as depression that are associated with both HF50 and cognitive decline49 or markers of HF severity such as NYHA class. We were also unable to separate race and ethnicity as we were limited by cohort-defined categories. Third, NOMAS relied on patient self-report to identify HF, though self-reported HF has high accuracy.25 Since participant-reported HF is highly specific but not as sensitive as clinically validated HF diagnosis, data from NOMAS could bias our findings towards the null. We are also unable to account for practice effects with repeated cognitive tests since we used a composite factor score of multiple tests. However, this is likely to bias our results towards the null. Finally, we are unable to account for heterogeneity in the definition of HF across cohorts. Similarly, despite using a harmonized measure of cognition, which has been previously validated, residual heterogeneity may persist, especially for the memory domain, and limits our ability to assess changes in components of each cognitive domain. However, we repeated analyses within cohorts in our sensitivity analysis to assess for heterogeneity in associations between HF and cognitive decline and these results were similar to results noted in the pooled cohort.
In conclusion, in a pooled analysis of individual participant data from 6 large, prospective cohort studies, participants with incident HF diagnosis, compared with participants without HF, had initial decreases in GC and executive function and faster, persistent declines in GC and executive function after accounting for pre-HF cognitive trajectories and comorbidities like stroke. Larger initial decreases in cognition after incident HF diagnosis were noted in older adults, White, and female participants. Future studies should assess the mechanisms of cognitive decline after incident HF, if specific therapies mitigate the risk for cognitive decline associated with HF, and whether assessment of cognitive function should be considered as part of HF management.
