Methods
The ARIC (Atherosclerosis Risk In Communities) study is a prospective study of cardiovascular disease among individuals from 4 U.S. communities. To study the prevalence of heart failure hospitalizations, the ARIC study began surveillance of hospital discharge records for all residents of the 4 communities in 2005. Inclusion criteria for hospitalization review included age >55 years, home address within 1 of the 4 communities, and an International Classification of Diseases-Ninth Revision-Clinical Modification (ICD-9- CM) discharge diagnosis code for heart failure or a related condition or symptom (398.91, 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, 415.0, 416.9, 425.4, 428.x, 518.4, 786.0x). Discharge diagnosis codes could be in any position for inclusion.
Eligible hospitalizations were identified through review of medical records from hospitals serving the ARIC communities. Stratified random sampling was used to identify hospitalizations for initial abstraction by trained abstractors. Detailed chart abstraction was performed if there was evidence of symptoms that could be related to heart failure or physician documentation of heart failure as the reason for hospitalization. Hospitalizations with detailed chart abstractions were subsequently adjudicated by a committee of physicians. The committee classified hospitalizations into the following categories: acute decompensated heart failure, chronic stable heart failure, heart failure unlikely, or unclassifiable. As previously described, classification of acute decompensated heart failure was favored if there was evidence of worsening heart failure symptoms with augmentation of therapy, while chronic stable heart failure was selected if there was evidence of heart failure without change in symptoms.
The population for the present study was selected as a cohort of individuals with adjudicated heart failure, either acute decompensated heart failure or chronic stable heart failure, from 2005 to 2009. We excluded individuals who were transferred to another hospital or who died during hospitalization.
We compared the rate of compliance with quality of care measures for individuals with heart failure who were hospitalized with a principal diagnosis of heart failure and those hospitalized for another cause. Principal heart failure diagnosis was based on ICD-9-CM codes used by CMS and ACCF/AHA/AMA-PCPI listed in the primary position. Additionally, we repeated the analyses using 3 alternative definitions of heart failure as cause of hospitalization. The first was determined by the response to the following question by trained abstractors: was there evidence from physician notes that heart failure was the primary reason for hospitalization. This definition may best reflect the physician perception of the reason for admission and should not be influenced by hospital coders. The second alternate definition was the ARIC study adjudicated diagnosis of acute versus chronic heart failure. The third alternate definition used a combination of the ARIC study definition and ICD-9-CM coding. Similar to that used in some quality initiatives, this definition included hospitalizations with a principal heart failure diagnosis that was also adjudicated as acute heart failure.
The primary outcomes were the 2 CMS inpatient heart failure quality measures available in the ARIC study dataset: assessment of LV function and discharge prescription of an ACE inhibitor or ARB for individuals with LV systolic dysfunction. LV function assessment was determined based on chart evidence of assessment either prior to or during the hospitalization.
Rates of compliance for 3 additional discharge measures were evaluated: prescription of a beta-blocker for individuals with LV systolic dysfunction, prescription of an aldosterone antagonist for individuals with LV systolic dysfunction and creatinine ≤2.5 mg/dl in men and ≤2.0 mg/dl in women, and prescription of anticoagulation for individuals with atrial fibrillation. These guideline recommended therapies are not nationally reported measures but have been considered as emerging measures of care in heart failure. In the ARIC study, information regarding anticoagulation at discharge was only available for a random 20% sample of hospitalizations that had supplemental data abstracted. For comparison, we also evaluated the rate of statin use among patients with a history of coronary heart disease; we were unable to assess aspirin utilization as information on this medication was frequently missing.
Demographic characteristics, medical history, and clinical results were obtained through medical record abstraction. LV systolic dysfunction was considered present if either the physician reviewer indicated abnormal systolic function or the documented ejection fraction was <50%. Edema, systolic blood pressure, and weight were obtained at time of admission; we used spline terms for blood pressure up to and above 140 mm Hg. Discharge sodium and estimated glomerular filtration rate (eGFR) were based on the final sodium and creatinine laboratory values during hospital admissions, respectively; eGFR was based on the CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) equation and categorized.
Mortality data were obtained from the National Death Index and determined for 1 year following discharge. Mortality data were only available for individuals discharged during the period of 2005 to 2008.
Statistical Analysis
Baseline characteristics were compared between groups using chi-square and t tests. Poisson regression was used to estimate the prevalence ratio of compliance with quality measures for individuals with as compared to those without a principal heart failure diagnosis, after adjustment for a priori selected covariates of age, sex, race, insurance, systolic blood pressure, eGFR, and history of coronary heart disease, diabetes, asthma, chronic obstructive pulmonary disease (COPD), dialysis, stroke, and atrial fibrillation.
Logistic regression was used to determine the associations of a primary discharge diagnosis of heart failure and adherence to quality measures with mortality following discharge. Covariates in the models were those available in the ARIC study dataset that were similar to predictors of mortality in a prior study of hospitalized patients with heart failure and included age, systolic blood pressure, eGFR, sodium, presence of edema, statin at discharge, and history of coronary heart disease, diabetes, asthma, COPD, stroke, and depression plus the additional demographic covariates of race, sex, and insurance. We developed a primary model and then a second model with an interaction term of the quality measure and an index of whether heart failure was the principal discharge diagnosis. Logistic regression models were also developed for both individuals with a principal diagnosis of heart failure and those with another principal diagnosis.
Several sensitivity analyses were performed. First, we determined the association between each of the 3 alternative definitions of heart failure. Second, we repeated our primary analysis with limiting inclusion to CMS heart failure diagnosis in the primary or secondary position. This analysis was done as hospitalizations screened and potentially adjudicated in the ARIC study as heart failure included codes outside of the CMS coding definition of heart failure, including such diagnoses as rheumatic heart failure, cor pulmonale, and shortness of breath. Third, we estimated the adjusted prevalence ratio of the measures for individuals with as compared to those without a principal heart failure diagnosis by eGFR categories and COPD status.
All statistical analyses accounted for the sampling design of the ARIC surveillance study. Statistical significance was pre-specified at an alpha level of 0.05 (2-tailed). Analyses were performed using Stata version 12 (StataCorp, College Station, Texas).