Risk Adjustment in the Medicare Atrial Fibrillation Population
Brian F. Gage, M.D., M.Sc.
- What is the association between cardiovascular prognostic factors and mortality in Medicare beneficiaries with atrial fibrillation?
- What is the association between cardiovascular prognostic factors imputed from Medicare part A data and mortality in Medicare beneficiaries with atrial fibrillation?
- What is the association between the Deyo-modified Charlson index andmortality in Medicare beneficiaries with atrial fibrillation?
- To quantify the association between common cardiovascular prognostic factors (eg. hypertension and diabetes) and mortality in patients who have atrial fibrillation.
- To determine how well cardiovascular prognostic factors imputed from administrative data predict mortality.
- To quantify the accuracy of using the Deyo-modified Charlson index for mortality risk adjustment and to compare it to alternative schemes (e.g. A. Elixhauser, 1998).
The study will use an existing database, the National Registry of Atrial Fibrillation (NRAF). With funding from the Agency for Healthcare Research and Quality (AHRQ) and in collaboration with the Medicare Peer Review Organization (PROs), we assembled the NRAF dataset, a registry of Medicare patients who have chronic atrial fibrillation and a high mortality rate (approximately 19% in year 1). The NRAF dataset contains both administrative and chart-review data on 3932 Medicare beneficiaries.
This project focuses on mortality in elderly patients with heart disease. Drs. Kerzner, Gage, Freedland, and Rich have focused on two common types of heart disease that are increasing in prevalence because of the aging of the American population: congestive heart failure and atrial fibrillation.
Congestive heart failure affects at least 5 million Americans. Prior research has focused on heart failure in patients who have systolic dysfunction (inability of the heart to squeeze normally). However, among the elderly, diastolic dysfunction (inability of the heart to relax normally) is more common than systolic dysfunction and both type of heart failure increase mortality and decrease quality of life.
The investigators have correlated how gender, age, heart function, laboratory tests, and medications affect the risk of death in patients who have heart failure. They found that predictors of mortality were different in the two subpopulations: among elderly patients with diastolic dysfunction a common test of renal function (BUN) was the only independent predictor of death; among elderly patients with systolic dysfunction the independent predictors of death were advanced age, prior heart attack, high-dose diuretics (water pills), or more severe systolic dysfunction on echocardiography. Patients with systolic dysfunction had a significantly greater risk of death. Thus, although both types of heart failure have similar signs (e.g. ankle swelling) and symptoms (e.g. shortness of breath), their prognosis and factors affecting prognosis differed.
How comorbid conditions affect the risk of death in patients who have atrial fibrillation (AF) and other heart conditions is unclear. AF causes an irregular heart rate and affects 2-3 million Americans. Prior research has determined that patients with AF have a high mortality rate. The cause of this increased mortality is partly from an increased rate of stroke in AF and partly from the company that AF keeps-advanced age, high blood pressure, diabetes, and heart disease. Determining which factors are most strongly associated with mortality may allow clinicians to determine treatment priorities.
The goal of this project is to determine how to control for the affect of co-morbid conditions when quantifying the risk of death from AF and other heart diseases. To quantify the effect of co-morbid conditions, investigators analyze large data sets. The most available large data sets are administrative data, such as Medicare Part A Records (MedPAR) that contain thousands of records, each with lists of co-morbid conditions noted as ICD-9 (International Classification of Disease-9) codes. The goal of this project is to learn how to use these data to quantify the affect of co-morbid conditions on mortality in Medicare beneficiaries who have AF and other heart disease.
6 Month Report:
We have finished our study of how systolic and diastolic dysfunction affect mortality in patients with heart failure. The Am Heart J has accepted the manuscript describing our results. Below is an abstract of that manuscript (Kerzner R, Gage BF, Freedland KE, Rich MW. Predictors of Mortality in Patients with Heart Failure and Preserved Left Ventricular Systolic Function), which was presented at American Geriatrics Society in 2002.
Although half of elderly patients with heart failure have preserved left ventricular ejection fraction (LVEF), little is known about predictors of mortality in this group.
We reviewed the charts of 400 patients hospitalized at an academic medical center in 1999 with a principal discharge diagnosis of heart failure. Patients were divided into four groups based on age > 75 or < 75 years and either preserved or reduced LVEF. Vital status was ascertained as of October 2001.
373 patients (mean age 69.1 years, 56.0% female, 47.5% non-white) underwent echocardiography to assess LVEF. Of these, 216 patients were < 75 years of age [81 with preserved LVEF (Group 1) and 135 with reduced LVEF (Group 2)], and 157 were > 75 years of age [81 with preserved LVEF (Group 3) and 76 with reduced LVEF (Group 4)]. After a mean follow-up of 25 months, independent predictors of mortality in the four groups differed substantially:
Group 1 - male gender, prescription of a calcium channel blocker, and diuretic dose at discharge;
Group 2 - blood urea nitrogen, lower hemoglobin level, and not being prescribed a beta-blocker at discharge;
Group 3 - blood urea nitrogen; and
Group 4 - older age, history of myocardial infarction, severity of reduced LVEF, and diuretic dose.
In patients with heart failure, predictors of mortality vary by age and by the presence of preserved or reduced LVEF. Traditional predictors of mortality in patients with reduced LVEF may not apply to elderly patients with preserved LVEF.
Now that we have finished the above manuscript, we will focus on how to control for the affect of co-morbid conditions when quantifying the risk of death from AF and other heart diseases. This knowledge will help clinicians to understand the interaction between AF and comorbidities, and to target specific comorbid conditions to reduce risk of death. We will focus on two popular comorbidity measurements that use administrative databases-the Charlson-Deyo scheme1 and the Elixhauser scheme.2 Neither scheme was developed and tested in populations with heart disease. Thus, their applicability to the AF population is unknown. Purposes of the study are: (1) to evaluate predictive accuracy of two existing comorbidity schemes to each other and to a disease-specific co-morbidity index, CHADS2, that has been used to predict stroke in the AF population.3
We have made two important pieces of progress towards the AF analysis. First, we have obtained permission from the Center of Medicare and Medicaid Services to obtain administrative data from approximately 15,000 patients who have AF. The primary analysis of these data is supported by a grant from the American Heart Association, but funding from the Longer Life Foundation allows for a specific analysis of mortality. Second, we have developed the macros in SAS that will allow us to implement the Charlson-Deyo and the Elixhauser schemes and to compare their predictive accuracy on the basis of calibration4 and discrimination.5
We have submitted a grant entitled, Pharmacogenetics, Biomarkers, and Antithrombotic Therapy to the Heart-Lung-Blood Institute at the NIH to continue our work in the area of AF. If funded, the proposed 4-year study will analyze genetic and non-genetic laboratory markers of fatal and non-fatal adverse events in patients with AF and others who take warfarin.
1. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 1992;45:613-19.
2. Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care 1998;36:8-27.
3. Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Validation of Clinical Classification Schemes for Predicting Stroke: Results from the National Registry of Atrial Fibrillation. JAMA 2001;285:2864-70.
4. Graf E, Schmoor C, Sauerbrei W, Schumacher M. Assessment and comparison of prognostic classification schemes for survival data. Stat Med 1999;18:2529-45.
5. Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15:361-87.
This grant focused on mortality in elderly patients with heart disease. We studied patients with heart failure (Project 1) and atrial fibrillation (Project 2) ... Read the full Final Report.