Declining mortality following acute myocardial infarction in the Department of Veterans Affairs Health Care System

Background Mortality from acute myocardial infarction (AMI) is declining worldwide. We sought to determine if mortality in the Veterans Health Administration (VHA) has also been declining. Methods We calculated 30-day mortality rates between 2004 and 2006 using data from the VHA External Peer Review Program (EPRP), which entails detailed abstraction of records of all patients with AMI. To compare trends within VHA with other systems of care, we estimated relative mortality rates between 2000 and 2005 for all males 65 years and older with a primary diagnosis of AMI using administrative data from the VHA Patient Treatment File and the Medicare Provider Analysis and Review (MedPAR) files. Results Using EPRP data on 11,609 patients, we observed a statistically significant decline in adjusted 30-day mortality following AMI in VHA from 16.3% in 2004 to 13.9% in 2006, a relative decrease of 15% and a decrease in the odds of dying of 10% per year (p = .011). Similar declines were found for in-hospital and 90-day mortality. Based on administrative data on 27,494 VHA patients age 65 years and older and 789,400 Medicare patients, 30-day mortality following AMI declined from 16.0% during 2000-2001 to 15.7% during 2004-June 2005 in VHA and from 16.7% to 15.5% in private sector hospitals. After adjusting for patient characteristics and hospital effects, the overall relative odds of death were similar for VHA and Medicare (odds ratio 1.02, 95% C.I. 0.96-1.08). Conclusion Mortality following AMI within VHA has declined significantly since 2003 at a rate that parallels that in Medicare-funded hospitals.


Background
The Department of Veterans Affairs (VA) operates the largest integrated health care system in the U.S. including 154 medical centers in all 50 states and Puerto Rico. Over the past 12 years the Veterans Health Administration (VHA) has undergone a remarkable transformation accompanied by remarkable improvements in quality. [1] In a study of nearly 600 patients receiving care from VHA in 12 communities between 1997 and 2000, their health care was generally of higher quality on 294 measures than that received by nearly 1000 patients enrolled in commercial managed care plans in the same communities. [2] More recently, VHA's performance on a wide range of measures that are monitored for patients enrolled in Medicare, Medicaid, and commercial managed care plans (Health Employer Data Set -HEDIS) outpaced that of health care plans in these settings. [3] On most of the ORYX measures publicly reported by The Joint Commission, VHA's most recent average national scores equaled or exceeded mean scores for non-VA facilities. [4] These achievements reflected by performance measures are paralleled by data indicating that adjusted mortality was lower over a fouryear period for patients receiving care in VHA compared with those enrolled in the Medicare Advantage program. [5] These efforts have led to VHA's recognition as one of the highest quality health care systems in the nation [6][7][8][9], although others have questioned their validity. [10] In the area of cardiac care, studies performed a decade ago found that mortality following acute myocardial infarction (AMI) among patients treated in VHA facilities was similar to that of patients whose care was funded by Medicare in non-VA hospitals. [11,12] Subsequently, in an external study commissioned by VA that used administrative data from patients hospitalized with AMI between 1996 and 1999, Landrum and colleagues found that survival was significantly lower at one year than for patients whose care was funded by Medicare. [13] Despite concerns about that study's methodology, VA used the study as an impetus for an ambitious 10-point cardiac care initiative intended to improve acute cardiac care. [14] A subsequent study, however, showed that the higher mortality within VHA reported by Landrum, et al was likely an artifact related to a much higher proportion of patients who suffered an AMI after being admitted to the hospital for another medical problem in the VA system than in the Medicare cohorts. [15] Given these controversies, we sought to reevaluate trends in mortality since the VA cardiac care initiative was undertaken. Moreover, because mortality related to cardiac disease has been declining worldwide, we also were interested in comparing trends within VHA with the Medicare population as an indication of secular trends. [16]

Methods
We undertook two separate sets of analyses: one to calculate annual mortality rates in VHA following AMI and a second to compare mortality rates among patients treated for AMI in VHA hospitals and those in hospitals funded by Medicare (Table 1).

Trends in Mortality in VHA
To examine trends in mortality following AMI within VHA we used information from the VHA External Peer Review Program (EPRP), which is a quality monitoring and improvement system for a variety of medical conditions and procedures, including AMI. All patients who were age 65 years and older and who had been assigned International Classification of Diseases, 9th Revision: Clinical Trained abstractors under contract to the EPRP program reviewed these records to validate the coded diagnoses and collect extensive clinical information. We used information from EPRP and VA workload files to describe demographic and personal characteristics (e.g., age, sex, ethnicity, body mass index and distance from the veteran's residence to the medical center), receipt of reperfusion therapy, and use of cardiac procedures. [17,18] Ethnicity was classified as African-American, Hispanic, White, other, or unknown. Clinical data, such as medical history, cardiac medications administered, time from symptom onset to hospital admission, initial heart rate, systolic and diastolic blood pressures, and initial symptoms, were obtained from the medical record. Patients included in the analysis were males age 65 or older who presented to a VHA facility with an AMI. Patients who developed evidence of AMI during hospitalization were excluded.
The initial electrocardiographic diagnosis of ST-segment elevation AMI included patients with ST-segment elevation of 1 mV or higher in 2 or more contiguous leads and/ or left bundle branch block. The remainder of patients had non-ST-segment elevation AMI. Both the initial and highest troponin values were recorded; positive tests were determined according to criteria for the particular assay used. In both groups, more than 99% of the AMIs were confirmed by troponin level elevations.

Ascertainment of Vital Status
The primary endpoint was all-cause mortality within 30 days of admission. Secondary endpoints were in-hospital, 90-day, and

Analysis
We applied a logistic regression model for 30-day mortality that was developed and validated expressly for VA patients using earlier versions of EPRP data. [15] The model includes history of cancer, history of dementia, history of heart failure, history of stroke within 5 years, history of lipid disorder or prescriptions for lipid medications, age, elevated initial troponin, creatinine, heart rate, systolic blood pressure, chest and/or shoulder pain, and presentation at night. Analyses were run with and without a cluster correction for medical center.

Comparisons of Trends in VHA with Medicare
We used comparable administrative data to estimate the relative mortality rates among patients admitted with AMI to VHA medical centers versus patients whose care was covered by Medicare at non-VA hospitals. Data were derived from two primary sources: the Medicare Provider Analysis and Review (MedPAR) data files; and the VA Patient Treatment File. The MedPAR files contain administrative data on all Medicare fee-for-service hospitalizations, including: demographic information; admission source (e.g., transfer from another hospital, emergency room); ZIP code; primary and secondary diagnoses and procedures, as defined by ICD-9-CM codes; hospital admission and discharge dates; disposition at the time of hospital discharge; and a six-digit unique hospital identifier.  Table 4 (defined by ICD-9-CM secondary diagnosis codes using previously defined algorithms for administrative data developed by Elixhauser, et al [22] and Hannan, et al [23]; and type of AMI (categorized as anterior or lateral, inferior or posterior, subendocardial, or other unspecified location (defined by the fourth digit of the primary ICD-9-CM code)).   Table 3 and Figure 1). Similar declines were found for inhospital and 90-day mortality.

Comparisons of Trends in VHA with Medicare
Declines in mortality following AMI within VHA hospitals were compared with those experienced in private sector hospitals using administrative data from VHA and CMS.

Discussion
Using both administrative data and detailed clinical information abstracted from medical records, we found that mortality following AMI among patients admitted to VA medical centers significantly declined between 2004 and 2006. This decline approximates the decrease that has occurred in hospitals outside of the VHA system during the same timeframe. These results are consistent with reports from a decade ago indicating parity in AMI mortality between VHA and Medicare. Our findings appear to lay to rest concerns that were raised more recently that mortality among patients treated for AMI in VHA appeared significantly higher than for patients whose care was funded by Medicare. In that study, which was actually commissioned and funded by VHA, administrative data from 1997 and 1998 were analyzed to evaluate mortality for patients admitted with AMI. A subsequent study, however, indicated that findings of high post-AMI mortality in VA patients were biased, because patients who suffered an AMI after being admitted to the hospital for another problem were far more  N = 11,609).
prevalent in the VHA than in Medicare cohorts. Using updated data and methods in previous work, we have found that in-hospital AMI accounts for 11% of all AMIs within the VHA system and the mortality for this problem is 28%, nearly fourfold higher than for patients who are admitted initially for an evolving AMI. [15] When all patients with in-hospital AMI were excluded from consideration, 30-day mortality following AMI was found to be 11.9%, which is lower than observed in the present analysis but includes patients younger than age 65.
Although the decrease in mortality within VA was commensurate with that occurring in the rest of the country, it may, in part, have been related to an aggressive effort undertaken within VHA over the past four years. Following public release of the data by Landrum, et al, VHA embarked upon a system-wide cardiac care initiative that encompassed educational programs for patients and providers, expansion of performance measurement and quality improvement systems, creation of administrative structures to coordinate care for AMI, and enhancement of infrastructure such as cardiac catheterization laboratories. [14] There are, however, other explanations as well.
The most likely is that VHA has maintained very high levels of performance with regard to prescribing cardioprotective medications post-AMI. Data from 2006 showed that eligible patients in VHA received aspirin, statins, and β-blockers at discharge at least 95% of the time, and that 87% of those with left-ventricular ejection fractions below 40% received angiotensin converting enzyme inhibitors. These figures exceeded performance of all hospitals reporting to the Joint Commission. [4] Moreover, investigators have found that VA patients tend to be substantially more adherent to these medication regimens over time, presumably because their cost of medications is so much lower than that of most patients enrolled in Medicare. [24] There have been concerns that patients in VA were less likely to undergo indicated cardiac interventions after AMI. [25] In 2005, 39% of eligible patients with AMI received thrombolytic therapy within 30 minutes (39%), which is comparable to all hospitals reporting to the Joint Commission. However, VA currently reports a lower proportion of patients who undergo percutaneous coronary interventions within 90 minutes when indicated (48% vs. 78%). Although this latter difference is clinically important, it would be likely to translate into only very small differences in overall survival following AMI. Less than one-quarter of patients admitted to VA hospitals with AMI have ST-segment elevation, and of these, only a relatively small fraction present early enough to be eligible for emergent PCI. Compared with noninvasive treatment, emergent PCI improves short-term mortality by approximately 2% in absolute terms. [26,27] Moreover, Ford, et al estimate that pharmacologic interventions and rehabilitation during and after hospitalization are responsible for 93% of the reduction in mortality following AMI while only 7% is attributable to revascularization. [28] More aggressive early treatment of AMI thus is likely responsible for only a relatively small proportion of the observed declines in mortality in either VA or Medicare patients.
It could be argued that some of the observed decline in mortality is simply due to increasing use of "super-sensitive" troponin assays. These assays could be responsible for increasing detection of patients with minimal myocardial damage whose prognosis is good. However, the data are inconsistent with this hypothesis. Existing evidence suggests that elevated troponin, even in the absence of classic indicators of AMI, confers a substantially increased risk of short-term mortality so that including these patients would not necessarily bias mortality downward. [29] Furthermore, increasing the detection of NSTEMI AMIs would be expected to increase the proportion of AMIs that are NSTEMI and to increase the overall number of AMIs. In the data reported here, the proportion of AMIs that were NSTEMI increased only slightly from 73% to  76% while the absolute number of AMIs actually declined.
To determine whether the observed declines could be explained by increasing inclusion of patients with small, low-risk infarcts diagnosed by slight elevations of serum troponin, we performed a regression of troponin level on discharge date. Approximately 200 assay type/hospital groups were excluded because there were less than three measurements, leaving approximately 9000 of 11,000 troponin values for analysis. Although we obtained a regression coefficient of about -0.07 per year (p < 0.0001) for initial and peak troponins, the R 2 was quite small (0.003). And finally, we performed an informal analysis of 9000 troponin values in an attempt to ascertain whether there was a trend toward lower values over time. Although the analysis was complicated by the fact that patients varied widely with respect to the number of tests performed and their timing, date of discharge accounted for only 0.3% of the variance in mean initial and peak troponin levels.
This study has several strengths that include eligibility of all patients diagnosed with AMI in both VA and Medicare populations over a period of several years and careful adjustment of mortality within VHA using two completely different approaches. In addition, we ascertained mortality using methods that have been shown to be highly reliable and valid. There are, however, significant potential limitations that deserve comment. First, the EPRP data used to assess mortality following AMI in VHA were available only for calendar years 2004 through 2006. Second, we were able to compare mortality in VHA and Medicare using only administrative data because CMS does not collect detailed clinical data on all patients with AMI. Although there may have been unrecognized differences in the way these data were collected, we believe that measurement of vital status is consistent for the two populations. Third, there is ample evidence that reliance solely on administrative data may fail to adequately adjust for differences among patients treated in different systems of care. [30] We think this is unlikely to have seriously affected our analyses given that both the crude and fully adjusted estimates of mortality were similar in the VA and Medicare groups. Fourth, because the public release of Medicare data is delayed, we were only able to make these comparisons through 2005. Finally and most importantly, the inherent differences in the VA EPRP data and the VHA and CMS administrative data sources led to slightly different denominators and estimates of outcome. The lower number of patients with AMI in the EPRP analysis is related to the fact that in a substantial minority of patients who have a discharge diagnosis of AMI, careful review of the medical record does not substantiate that diagnosis. The differences in adjusted outcomes result, in part, from the ability to adjust more completely for the considerable burden of coexisting, chronic illness among VA patients.

Conclusion
In summary, we found that mortality following AMI has declined over the past four years in VA and that this decline has been commensurate with that occurring in the Medicare population.