Temporal Trends in Heart Failure Incidence Among Medicare Beneficiaries Across Risk Factor Strata, 2011 to 2016

This national cohort study evaluates trends in heart failure incidence across risk factor strata, including hypertension, diabetes, obesity, acute myocardial infarction, and atrial fibrillation among Medicare beneficiaries.


Introduction
In the United States, Medicare beneficiaries represent 70% to 80% of all patients hospitalized with heart failure (HF) each year. 1 The Medicare population has also experienced substantial changes in the epidemiology of HF, with progressively fewer Medicare beneficiaries being diagnosed as having HF each year over the last decade after several decades of increasing incidence. [2][3][4][5][6][7][8] However, the epidemiological mechanisms underlying the observed decline in the incidence of HF are not well understood. This lack of understanding represents an important knowledge gap because the prevalence of HF among older individuals continues to increase despite the declining incidence, so the primary prevention of HF is a key strategy to reduce morbidity and mortality from HF in the United States.
A potential mechanism for the decline in the incidence of HF may be associated with a concomitant decrease in the prevalence of risk factors for HF. These include primary HF risk factors, such as hypertension, diabetes, and obesity, which may act through multiple distinct mechanisms to increase the risk of HF, and predisposing cardiovascular (CV) conditions, such as acute myocardial infarction (MI) and atrial fibrillation (AF), which can more directly lead to myocardial dysfunction and increase the risk of developing HF. 9 Another mechanism underlying the reduction in the incidence of HF may be better treatment of individuals with prevalent risk factors, [10][11][12][13][14][15] thereby mitigating the downstream development of HF.
Better understanding of the epidemiological mechanisms contributing to the observed decline in the incidence of HF is important to develop future public health interventions aimed at preventing HF.
Accordingly, in this study of a national cohort of Medicare beneficiaries, we evaluated the temporal trends in the incidence of HF across different strata of risk factor burden spanning inpatient and outpatient care settings.

Methods
This retrospective, national cohort study was reviewed by the institutional review board of the University of Texas Southwestern Medical Center, which waived the requirement for informed consent because the study represents secondary analyses of deidentified data. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Data Source
We used a national 5% sample of all fee-for-service Medicare beneficiaries with no prior HF followed up from 2011 to 2016, accessing data on all of the claims submitted to Medicare from inpatient and outpatient encounters, as well as physician or carrier claims. The data source has been described previously. 7 Briefly, for every 100 individuals enrolled in Medicare, 5 are entered into the Medicare 5% sample based on a predefined combination of the Medicare beneficiary identification number, a 9-digit unique beneficiary identifier. Once entered into the Medicare 5% data set, the beneficiary is included in the data every year, with 5% of new Medicare beneficiaries entering the data set every year. Beneficiaries leave the sample only when they die or when they are no longer eligible for Medicare benefits. The data include information on all sampled Medicare beneficiaries but include inpatient and outpatient claims only for fee-for-service Medicare beneficiaries enrolled in Medicare Part A and Part B.

Study Population
We created a cohort of Medicare beneficiaries 65 years and older who were continuously enrolled in both Part A and Part B of fee-for-service Medicare. Enrollment in Medicare Part A and Part B ensured that insurance claims submitted for all health care encounters for these patients were available. In this cohort, we classified patients as having prevalent HF if they had a health care encounter with a claim code for HF during the first 12 months of entering the cohort. encounter the health care system either during a hospitalization or in an outpatient setting over the course of 1 year and has been used in previous studies 4,7 using Medicare claims. Those without prevalent HF were followed up longitudinally within administrative claims to identify individuals who had new-onset or incident HF, which is defined in further detail below. We focused on the incidence of HF among this cohort in the contemporary era (2011-2016) after changes in data transaction standards for Medicare claims had been implemented widely. 16, 17 We used claims from 2009 and 2010 to identify prevalent HF for sensitivity analyses, as outlined in the Statistical Analysis section.

Study Variables and Outcomes
The demographic characteristics of age, sex, and race/ethnicity were identified from the Medicare Denominator File. Sex and race/ethnicity categories are based on information provided by Medicare beneficiaries at the time of enrollment in Medicare. We used claims data over the 12-month period of entering the cohort to define the baseline characteristics and risk factor profile of individuals. We used inpatient, outpatient, and physician carrier claims to define 5 common risk factors that are identifiable in claims data. These include 3 conditions (hypertension, diabetes, and obesity) that serve as primary HF risk factors and potentially act through multiple pathways to elevate the risk of HF. An additional 2 conditions (acute MI and AF) were chosen as secondary predisposing CV conditions that may be associated with the presence of primary HF risk factors but also have a more proximate impact on the risk of developing HF. We used a single claim code for conditions across all sources to identify hypertension, diabetes, obesity, acute MI, and AF (eTable 1 in the Supplement).
We also defined other comorbidities using a similar approach. These included all comorbidities that are listed in eTable 2 in the Supplement.
The primary study outcome was incident HF, which was defined as at least 1 hospitalization with an inpatient HF claim or at least 2 outpatient or carrier HF claims among individuals without prevalent HF. For the latter, we required the outpatient claims to be in 2 separate calendar quarters to limit the selection of individuals who had HF as a potential differential diagnosis rather than a diagnosis of HF that requires longitudinal care. This approach has been used in previous studies. 4,7 In addition, among those with incident HF on follow-up, mortality within 30 days after the first diagnosis of HF was captured as the secondary study outcome.

Statistical Analysis
The demographic and clinical characteristics of Medicare beneficiaries at risk of HF at the beginning of each calendar year between 2011 and 2016 are described as the median (interquartile range) for continuous variables and as percentages for categorical variables. The incidence rate of HF for each year was also calculated for the overall cohort and across subgroups based on sex and race/ ethnicity. Annual trends in HF incidence rate were evaluated using linear regression models with year being the independent variable and yearly HF incidence being the dependent variable. In addition, trends in 30-day mortality rates of HF after the first diagnosis among those who developed incident HF during the study period between 2011 and 2016 were also calculated. To assess the association of comorbid conditions with the rate of incident HF, Poisson regression models were constructed with age, sex, race/ethnicity, socioeconomic status, and all 5 comorbid conditions as independent variables.
For each of the 5 comorbid conditions, HF incidence rate ratios between patients who had a comorbid condition and patients who did not have the condition were reported, adjusting for demographics and all other comorbid conditions. Separate models were created for each calendar year, and the annual trend of adjusted HF incidence rate ratios between patients with and without each of the comorbid conditions was also reported. To test the yearly trend, linear regression models

JAMA Network Open | Cardiology
Temporal Trends in Heart Failure Incidence Among Medicare Beneficiaries were constructed with the HF incidence rate ratio in each year as the dependent variable and with calendar year as the independent variable. Finally, we evaluated risk-adjusted trends in mortality. For this analysis, we constructed a logistic regression model with 30-day mortality as the outcome, covariates included in the Centers for Medicare & Medicaid Services model for mortality as independent variables, and calendar year as the exposure, consistent with the approach used in previous studies. [18][19][20][21] In sensitivity analyses, we defined prevalent HF based on a 2-year lookback period. Incidence trends were addressed only among those individuals who had 2 years of claims before a calendar year and no inpatient, outpatient, or professional claim for HF during this period.
The 2-sided level of statistical significance was set at P < .05 for all analyses. Statistical analysis was performed using R, version 3.6.0 (R Project for Statistical Computing).

Results
Overall, there were 1 799 027 unique Medicare beneficiaries at risk of developing incident HF at the beginning of each year during the study period. The median age of this cohort was 73 years

Incidence of HF Across Risk Factor Strata
The prevalence of each risk factor (hypertension, diabetes, obesity, acute MI, and AF) increased over time among those with and without incident HF on follow-up (eFigure 1 in the Supplement).
Throughout the study period, the incidence of HF was higher among individuals with vs without each of the risk factors for HF (Figure 2). Of all the risk factors, the incidence of HF was highest among individuals with acute MI and individuals with AF. A-C, The incidence rate per 1000 Medicare beneficiaries is shown. P values reflect trends in the incidence rate per 1000 beneficiaries over time.

JAMA Network Open | Cardiology
Temporal Trends in Heart Failure Incidence Among Medicare Beneficiaries During the study period, the incidence of HF declined across all risk factor-based subgroups, with a larger absolute decline among those with vs without each risk factor (Figure 2). For the primary HF risk factors, compared with the subgroup without a specific risk factor, there was a relative excess decline of 12% among patients with prevalent hypertension, 3% among patients with diabetes, and 16% among patients with obesity. In contrast, among individuals with predisposing CV conditions, there was a relative increase in HF incidence among individuals with acute MI (26% vs no acute MI) and individuals with AF (22% vs no AF) (Table). In adjusted analyses accounting for age, sex, race/ ethnicity, socioeconomic status, and other risk factors, there was a statistically significant decrease in the incidence ratio among individuals with diabetes (adjusted incidence, 1.48; 95% CI, 1.45-1.51 to 1.37; 95% CI, 1.34-1.40) and individuals with obesity (adjusted incidence, 1.59; 95% CI, 1.55-1.62 to    (Figure 3).

Trends in HF Mortality Over Time
The unadjusted 30-day mortality rate after incident HF increased in the overall study cohort from 16.2 cases per 1000 Medicare beneficiaries to 17.2 cases per 1000 Medicare beneficiaries (eFigure 2 and eFigure 4 in the Supplement). In risk-adjusted analyses, the odds of 30-day mortality decreased over the study period (the risk-adjusted odds ratio for mean annual change relative to 2011 was 0.82; 95% CI, 0.73-0.93; P = .001) (Figure 4).

Sensitivity Analyses
The use of a 2-year lookback period of claims to exclude prevalent HF was consistent with the primary analyses that used a 1-year period of claims. In the sensitivity analyses, we found a similar decline in incident HF, as well as similar trends among specific risk factor groups (eFigure 3 and eFigure 4 in the Supplement). A-E, Poisson regression models were constructed to assess the association of each risk factor with heart failure incidence for each year, with adjustment for age, sex, race/ethnicity, socioeconomic status, and all 5 risk factors. P values reflect trends in the incidence rate ratio over time. Error bars indicate 95% CIs.

Discussion
In a national cohort of almost 1.8 million fee-for-service Medicare beneficiaries 65 years and older followed up longitudinally for 6 years, we made several observations. The overall incidence of HF continued to decline between 2011 and 2016, regardless of sex or race/ethnicity. Concomitantly, among individuals with and without HF, rates of key underlying risk factors increased, suggesting that the decline in HF incidence is not because of an overall lower burden of HF risk factors. There was substantial heterogeneity in HF incidence across individual risk factors. Among patients with primary HF risk factors, including hypertension, diabetes, and obesity, the risk of developing new-onset HF compared with individuals without these risk factors decreased over time. In contrast, the risk of developing HF increased in individuals with predisposing CV conditions, such as acute MI and AF, compared with individuals without these conditions. In addition, we observed a concerning trend of increasing 30-day mortality rates among Medicare beneficiaries with HF throughout the study period. Taken together, these data suggest that the decline in HF incidence observed among Medicare beneficiaries is more likely associated with better treatment of patients with prevalent primary HF risk factors rather than early recognition of HF. Earlier identification and management of HF would be less consistent with the observed increase in 30-day mortality rates, a striking trend that warrants further investigation.
We found a statistically significant decline in incident HF during the study period. Our international studies, notably from Scotland, 5 Australia, 23 Canada, 24 Sweden, 25 Denmark, 8 Norway, 26 and the United Kingdom, 27 which all reported a decline in incident HF. eTable 3 in the Supplement summarizes these studies. Of note, the large study by Christiansen and colleagues 8 reported a trend by age, with a decreasing incidence of HF among individuals older than 50 years but an increase among individuals 50 years and younger. Therefore, our findings among Medicare beneficiaries are consistent with these studies. The observed trends in incident and prevalent HF appear to be associated with the decline in HF mortality among individuals with incident HF over the last 2 decades. 28 The declining incidence of HF among individuals with prevalent hypertension, diabetes, and obesity highlights the potential contribution of optimal management of these risk factors in the prevention of HF. Previous observational studies and randomized clinical trials have demonstrated Error bars indicate 95% CIs. that optimal control of these conditions is associated with lower risk of HF. [29][30][31] Furthermore, data from Medicare Part D have shown an increase in the use of cardioprotective pharmacotherapies in patients with type 2 diabetes. 32 Previous studies 33,34 have also demonstrated an overall improvement in adherence to antihypertensive therapies among Medicare beneficiaries. It is plausible that such improvements in health care practices may have contributed to optimal management of these comorbidities, contributing to a decline in the incidence of HF.

JAMA Network Open | Cardiology
In contrast to the decline in primary HF risk factors, we observed a statistically significant studies are needed to assess whether greater uptake of these therapeutic strategies over time may alter the trajectory of HF incidence among these patient populations.
We also found that the unadjusted 30-day mortality rate after a new diagnosis of HF increased during the study period. However, this increase in mortality was explained by older age and increasing risk factor burden at diagnosis, with a small decrease in risk-adjusted mortality over this period. Of note, mortality trends in this study included only Medicare beneficiaries with incident HF, who represent a small proportion of the overall population of Medicare beneficiaries with HF.
Therefore, these findings would not generalize to the larger trends in HF mortality among Medicare beneficiaries. However, the patterns found in our study of increasing risk factor burden have been observed in the larger Medicare population after hospitalization for HF. 20

Limitations
This study has some limitations. First, we used administrative claims to identify conditions of interest and patient outcomes and could not independently confirm these findings. However, as much as possible, we used claim codes that were used in prior studies to identify the major conditions included in the study, including HF, hypertension, diabetes, obesity, acute MI, and AF. 54,55 Second, the study was sensitive to changes in coding intensity over time. Although there may have been

JAMA Network Open | Cardiology
Temporal Trends in Heart Failure Incidence Among Medicare Beneficiaries pressure to code more conditions over time, the observed associations across conditions argue against this consideration being a major mechanism for the observed decline in HF incidence. Third, we could not definitively exclude prevalent HF among the study population. However, the results of the sensitivity analyses that evaluated a 2-year lookback period to define prevalent HF were consistent with the primary analyses. We argue that it is probably uncommon that an insured patient with HF would not seek any inpatient or outpatient care for comorbid HF. Fourth, we were unable to classify HF into subtypes because information on ejection fraction or left ventricular systolic dysfunction was not available. Fifth, although the study found a relative decrease in the incidence of HF among patients with hypertension, diabetes, and obesity, we cannot infer that this was associated with improvement in management, which was not evaluated in this study. This study only highlights that trends in HF incidence among major risk factor groups were concurrent with improvement in risk factor management during the last decade. 56 Sixth, the study's findings do not generalize to populations or individuals other than fee-for-service Medicare beneficiaries, and trends in younger populations, particularly those with high-risk conditions, were not a focus of this study.

Conclusions
An observed decline in HF incidence among Medicare beneficiaries does not reflect a lower risk factor burden but rather a decrease in incident HF among those with associated risk factors, suggesting improved primary HF risk factor management. The relative increase in HF incidence among those with predisposing CV conditions, such as acute MI and AF, highlights potential targets for future HF prevention strategies.