National and State Trends in Quality of Care for Acute Myocardial Infarction Between 1994-1995 and 1998-1999:  The Medicare Health Care Quality Improvement Program FREE

DESPITE STRONG scientific evidence demonstrating a survival benefit of a number of key therapies for acute myocardial infarction (AMI), many studies have documented their underuse.^1- 4 Substantial opportunity for improvement has been demonstrated in the quality of care that patients with AMI receive. Suboptimal quality of medical care has been increasingly recognized as a concern within the US health care system and has led to a presidential commission's recommendation that quality of care should be addressed as an important national priority.⁵ A recent report from the Institute of Medicine has further publicized this concern.⁶

Quality of care should be measured over time to determine whether treatment gaps are narrowing, but relatively little information about this has been published. Such information is critical to evaluate progress and help target future quality improvement efforts. Also, information about variation in the rate of improvement is needed to benchmark improvement strategies. For AMI, limited trend data have been reported from managed care plans focusing on use of β-blockers.⁷ An analysis of data from the National Registry of Myocardial Infarction examined trends in a number of AMI care processes, but this analysis was limited by not being population-based, not assessing whether patients had contraindications to certain therapies, and not focusing on variation in improvement.⁸ Other reports have focused on selected states⁹ or particular communities.^10- 12

Acute myocardial infarction has been a major focus of the Medicare Health Care Quality Improvement Program since 1992. This program is an ongoing initiative by the Centers for Medicare & Medicaid Services (CMS) (formerly the Health Care Financing Administration [HCFA], renamed in 2001), working with its contractor peer review organizations (PROs) for each state, to improve the quality of medical care for Medicare beneficiaries. The AMI efforts began as the Cooperative Cardiovascular Project pilot in 4 states and, based on initial success,⁹ was subsequently expanded nationally.

The data collected in this program are unique because they provide a national probability sample of AMI cases, focus on Medicare patients who bear much of the burden of AMI in the United States, and include substantial clinical detail which facilitates determination of whether patients had contraindications to a therapy. The data also provide a state-based analysis of the quality of care. State-level information is particularly useful because a number of the organizations that potentially can influence and facilitate quality improvement efforts are state based, including the PROs and the state chapters of a number of national professional societies or hospital associations.

A recent report described state performance on a number of AMI care processes.⁴ The present article extends these findings by examining the dynamic pattern of change over time (which is critical for evaluating progress), demonstrating feasibility of improvement, and determining the pace of diffusion of evidence-based therapies into practice, thereby providing information on which renewed strategies for improvement can be based.

We analyzed data collected for the Medicare Health Care Quality Improvement Program during 2 different periods: 1994-1995 and 1998-1999. The earlier data were collected as part of the Cooperative Cardiovascular Project. The later data were collected as part of the next phase of the CMS AMI efforts and titled the "National Acute Myocardial Infarction Project."^4,9 Both were national samples identified from Medicare claims data that generally consisted of fee-for-service claims. The administrative data were supplemented by in-depth clinical data abstracted from hospital medical records.

All AMI discharges were identified from the national claims history file, which is composed of Part A Medicare claims, by identifying records with a principal code 410.xx diagnosis (excluding 410.x2) from the International Classification of Diseases, Ninth Revision, Clinical Modification. Codes 41.x2 were excluded because they represent an admission for a subsequent episode of AMI care following an AMI within the previous 8 weeks.

The 1994-1995 sample totaled 234 754 records and has been previously described.³ Briefly, 100% of AMI discharges during an approximate 8-month period in 1994-1995 was sampled for most states. Exceptions to this included a 4-month sampling period for 4 states and an approximate 60% sample for 1 other state.

The 1998-1999 sample totaled 35 713 records and generally consisted of up to 750 AMI discharges per state (also the District of Columbia and Puerto Rico) in a 6-month period from April 1998 to March 1999. For some states, an additional 100 discharges were sampled. Systematic random sampling was used to select AMI discharges from each state after sorting by age, sex, race, and hospital provider number. If the number of records for a state was less than the number targeted for the sample, all AMI discharges in the relevant period were included. The 35 713 records obtained and abstracted represent 96% of the 37 376 records that were targeted for the sample.

Trained abstractors at 2 HCFA-contracted clinical data abstraction centers (CDACs) performed medical record abstraction. Interrater reliability of medical record abstraction was monitored using random reabstraction of samples of records. Discrepancies between abstractors were identified and examined, and retraining of staff was performed as needed based on the results. Extensive written abstraction guidelines provided instruction for standardized data collection. Minor clarifications to these instructions were incorporated over time, based on the issues identified during interrater comparisons.

When assessing the reliability associated with measuring process-of-care performance for hospitalized patients with AMI, we first examined the reliability of the individual variables used to generate the quality indicators. For the 1994-1995 data, interrater reliability was assessed between abstractors working at the same CDAC. Variable agreement (agreement rate between 2 abstractors assessing the same variable) averaged greater than 90%.¹³ Second, we examined the indicator components, namely, the numerators and denominators that also make up the individual variables. Quality indicator reliability (agreement rate between 2 abstractors assessing patient eligibility for and receipt of quality indicator) was 93% to 99% (κ = 0.41-0.76).¹⁴

For the 1998-1999 data, abstraction reliability was monitored within each CDAC and between the 2 CDACs. For the latter, a sample of 640 records was compared. Variable agreement generally averaged more than 90%. Quality indicator reliability using the method reported by Huff¹⁴ was as follows (agreement; κ value): early administration of aspirin (95.3%; 0.60); aspirin prescribed at discharge (96.7%; 0.65); early administration of β-blocker (96.3%; 0.71); β-blocker prescribed at discharge (97.3%; 0.55); angiotensin-converting enzyme (ACE) inhibitor prescribed at discharge (99.1%; 0.82); smoking cessation counseling (93.9%; 0.49); receipt of acute reperfusion therapy (97.8%; 0.66).

Only records with confirmed AMI were included in the analysis. For the 1994-1995 database, this was defined as a principal diagnosis of AMI and chart documentation of a creatine kinase–muscle-brain isoenzyme (CK-MB) fraction greater than 0.05, a lactate dehydrogenase (LDH) level more than 1.5 times normal with LDH₁ levels greater than those of LDH₂, or 2 of the following 3 criteria: chest pain, a 2-fold elevation of the creatine kinase level, or a new AMI found on the official electrocardiogram report. For the 1998-1999 database, elevation in troponin I or T level was also considered the equivalent of elevated enzymes and was defined as an AMI. To see if this differing definition substantially affected results, key analyses were repeated after excluding patients whose AMI was confirmed only by an elevated troponin level.

Quality indicators were defined using data elements common to both data sets (Table 1). Patients were counted as eligible for early therapies if they had confirmed AMI, were not transferred from another hospital or emergency department, were not transferred to another hospital or emergency department on day 1, and did not die on day 1. Patients were counted as eligible for discharge therapies if they had confirmed AMI, were not transferred to another acute care hospital, and did not die during the hospitalization. Additional criteria were used for specific indicators. After identifying confirmed patients with AMI who were eligible for the therapy, a subset of patients who were "ideal" candidates for the therapy was defined by excluding patients with absolute or relative contraindications from the denominator.⁹ Some indicators were treated slightly differently than in previous reports^3- 4 so that data elements common to both data sets could be used.

Sample weights were assigned to each record based on the inverse of the selection probability. National performance on quality indicators and patient characteristics during the 2 periods were compared using regression analysis of the weighted data. The software used for these analyses was SAS version 8.1 (SAS Institute Inc, Cary, NC). Because of the small numbers of cases included in the time from hospital arrival to initiation of thrombolytic therapy or percutaneous transluminal coronary angioplasty, particularly in 1998-1999, unweighted data were used for these analyses, and comparisons were performed using the Kruskal-Wallis test. State-level performance on quality indicators during the 2 periods was compared using χ² analysis. Correlation between the change in performance over time and the baseline performance on quality indicators or between the changes in performance on pairs of indicators was assessed with linear regression using Microsoft Powerpoint 97 SR-2 trend features for scatterplots (Microsoft Corporation, Redmond, Wash) and Epi Info version 6.04b (Centers for Disease Control and Prevention, Atlanta, Ga).

Acute myocardial infarction was confirmed in 204 501 cases (87%) in the 1994-1995 data set and 31 331 cases (88%) in the 1998-1999 data set. The numbers of cases eligible for early or discharge therapies are given in Table 2. Characteristics of patients with confirmed AMI are summarized in Table 3. Compared with AMI cases in 1994-1995, cases in 1998-1999 more frequently occurred among patients who were older, were admitted from or discharged to a long-term care facility, and/or had a history of hypertension, prior revascularization, stroke, or other comorbidities such as lung disease, dementia, or renal insufficiency. They less frequently had ST-elevation myocardial infarction and had lower peak creatine kinase levels.

When the quality indicators were determined among patients without contraindications to the therapy ("ideal" candidates), results were as follows: nationally, the amount of absolute improvement was highest for the prescription of β-blockers (early, 17.4 percentage points; at discharge, 20.5 percentage points), moderate for the use of ACE inhibitors (8.0 percentage points) or aspirin (early, 6.6 percentage points; at discharge, 5.6 percentage points), not significantly changed for receipt of acute reperfusion therapy (1.5 percentage points), and negative for smoking cessation counseling (−3.6 percentage points) (Table 4 and Figure 1).

Among patients who received acute reperfusion therapy, the median time from hospital arrival to initiation of thrombolytic therapy improved (decreased) by 7.0 minutes, and the percentage of patients who had therapy initiated within 30 minutes increased from 25.2% to 33.7% (P<.001) (Figure 2). The median time from hospital arrival to initiation of primary angioplasty improved (decreased) by 12.0 minutes, and the percentage of patients who had therapy initiated within 90 minutes increased from 28.2% to 30.6%, but the change was not statistically significant (P = .46). Among the broader group of patients "eligible" for therapy, the change over time in the quality indicators was similar to that found for the ideal candidates except for ACE inhibitor at discharge, which improved 4.3 percentage points among eligible cases and 8.0 percentage points among ideal cases (Table 4).

Further stratification of the quality indicator results based on state is provided in Table 5. The distribution of discharges included in the state-based analysis was as follows: the median number of confirmed AMI cases in a state in 1994-1995 was 2366 (range, 135-15 114), and in 1998-1999 it was 642 (range, 130-763). The median number of ideal cases included in the denominator of an indicator (1994-1995; 1998-1999) was as follows: early administration of aspirin (1244; 301); aspirin prescribed at discharge (787; 211); early administration of β-blocker (704; 153); β-blocker prescribed at discharge (387; 78); ACE inhibitor prescribed at discharge (252; 78); smoking cessation counseling (318; 84); and receipt of acute reperfusion therapy (202; 32).

States varied substantially in their change over time (range): early administration of aspirin (−0.5% to 17.9%); aspirin prescribed at discharge (−1.2% to 14.4%); early administration of β-blocker (0.5%-39.1%); β-blocker prescribed at discharge (2.2%-51.2%); ACE inhibitor prescribed at discharge (−8.3% to 21.8%); smoking cessation counseling (−29.2% to 27.4%); receipt of acute reperfusion therapy (−23.1% to 40.0%).

When we examined whether the amount of improvement was explained by the baseline rate in 1994-1995, we found that a higher baseline level was associated with less improvement. This was statistically significant (P<.05) for early administration of aspirin (r² = 0.49), early administration of β-blocker (r² = 0.22), β-blocker at discharge (r² = 0.20), receipt of acute reperfusion therapy (r² = 0.13), smoking cessation counseling (r² = 0.11), and ACE inhibitor at discharge (r² = 0.09), but not for aspirin at discharge (r² = 0.06). After excluding the 2 nonstates (District of Columbia and Puerto Rico), which appeared to be outliers on a few indicators, we again found statistically significant correlation for 6 of the 7 indicators, including aspirin at discharge, but not receipt of acute reperfusion therapy.

When we examined whether improvement on one indicator was correlated with improvement on other indicators at the state level, we found statistically significant correlations between β-blocker administered early and at discharge (r² = 0.37); early administration of aspirin and receipt of acute reperfusion therapy (r² = 0.15); early administration of β-blocker and receipt of acute reperfusion therapy (r² = 0.12); and β-blocker prescription at discharge and receipt of acute reperfusion therapy (r² = 0.10). No significant correlations occurred between the other 17 pairs of indicators. After the District of Columbia and Puerto Rico were excluded, the only statistically significant correlation was between β-blocker administered early and prescribed at discharge. For the 1998-1999 data, when patients whose AMI was confirmed by an elevated troponin level were excluded from the analysis (n = 2539), results were not substantially affected.

These data from the Medicare Health Care Quality Improvement Program provide the first national state-based assessment of the change in quality of care for AMI. Our analysis focuses on a 4-year period during which a number of national and state-level events or interventions occurred that improved AMI care, and therefore provides important feedback on the progress to date and direction for future efforts. From 1994-1995 to 1998-1999, major improvements were achieved nationally in the use of β-blockers, while more moderate improvement occurred in the use of ACE inhibitors and aspirin for Medicare patients with AMI. No improvement occurred in the receipt of acute reperfusion therapy. However, among persons receiving it, the time from hospital arrival to initiation of thrombolytic therapy or primary angioplasty improved. The rate of documented smoking cessation counseling worsened slightly. These data extend the results from a recently published national and state profile⁴ by demonstrating that improvement has been achieved, including major gains in some indicators and states, but also that ample opportunities for improvement remain.

Several factors may be contributing to the differing amounts of improvement that occurred for different indicators. Most of the indicators that improved involved administration or prescription of a medication (aspirin early or at discharge, β-blocker early or at discharge, or ACE inhibitor at discharge). This type of treatment process is largely dependent on the action of a single physician and might be improved by interventions that change the physician's behavior (eg, reminder systems, educational outreach visits, data feedback, and the proactive participation of opinion leaders).¹⁵

Acute reperfusion therapy involves timely assessment of eligibility and either administration of a medication (thrombolytic agent) or access to a catheterization laboratory and cardiologist for primary percutaneous transluminal coronary angioplasty. Decision making about this therapy may be complicated by perceived greater risks in the older elderly population, although patients older than 80 years or with certain comorbidities are excluded from the ideal subgroup. Also, at institutions where both reperfusion options are available, the decision-making process may be more complex, may involve more than 1 physician, and has been reported to increase the time to administration.¹⁶ Improving timeliness involves a number of steps beyond physician decision making, including rapid attainment of an initial electrocardiogram, immediate availability of a thrombolytic agent or preparation of the catheterization laboratory, and availability of appropriate personnel for each step.

Smoking cessation counseling mainly involves discussion between the physician or other health professional and the patient and possible prescription of pharmacologic therapy. Improvement efforts could focus on increasing counseling by physicians or nonphysician staff such as specialized counselors. However, in the acute-care setting, other treatment procedures might take priority over smoking cessation counseling, which would then have to be undertaken during an ever-shortening inpatient stay.

Although some types of quality improvement efforts would be expected to address all these different care processes concomitantly (eg, identification of root causes for suboptimal care and development of a multifaceted intervention to address them), these data suggest that, in practice, some types of processes are more readily improved than others. Lack of improvement in the percentage of patients receiving reperfusion therapy or smoking cessation counseling during their hospital stay may be due to greater challenges in implementing improvement efforts in these areas than in drug administration or other treatments.

Acute myocardial infarction cases in 1998-1999 occurred among patients who were slightly older than the 1994-1995 patients, more likely to have transferred from or been discharged to a long-term care facility, and more likely to have certain comorbidities or have undergone previous coronary revascularization, yet the later patients had less frequent ST elevation and lower peak creatine kinase levels. An analysis of data from the National Registry of Myocardial Infarction⁸ covering this same period also noted increases in some comorbidities such as diabetes, stroke, hypertension, and a history of revascularization. Because the indicator calculations in the present study used to determine ideal candidates for therapy excluded patients with contraindications, it is unlikely that the change in indicator rates over time is due to differing frequencies of patient characteristics in the 2 periods. This is an important contribution of our study.

The indicator results among eligible candidates for therapy demonstrate that physicians often make the clinical judgment to treat patients who have relative contraindications. Thus, even though the denominators of the ideal subgroup are relatively small, the therapies have broader applicability than it would first appear. In fact, changes in the quality indicators between 1994-1995 and 1998-1999 among the eligible group were generally similar to those for the ideal group, though to a lesser extent for the use of ACE inhibitors. Although the eligible group would be expected to be more affected by changes in patient population over time, these population changes either were not sufficient to affect the trend or were counterbalanced by an increasing propensity by physicians to treat patients with certain relative contraindications.

The change in indicator rates over time varied by state. Some of this variation is explained by the baseline value in 1994-1995, with a higher baseline being associated with less improvement over time. One possible explanation for this is that improvement becomes more challenging as the baseline approaches 100% because the easiest improvements have already been achieved. Also, the opportunity for improvement decreases as the baseline approaches the ceiling. However, these factors alone are unlikely to fully account for the observed association because indicators with relatively low national baselines (eg, ACE inhibitor at discharge, 63%) provide ample room for improvement in all states.

A variety of factors are likely playing a role in the variation in improvement among states, including intensity and type of quality improvement efforts within the state. The CMS's charge to the Medicare PROs during this period included performance of a quality improvement project related to AMI in each state. The PROs worked collaboratively with health professionals, hospitals, health plans, professional organizations, and others to facilitate voluntary quality improvement efforts. The scope and type of projects varied among states, which could be one of the factors contributing to the observed variation in the change in indicator rates.

While most successful quality improvement projects will identify and address barriers to optimal care in the particular hospital, health plan, or local area, the state-level results provide a standardized means of monitoring trends over time and a statewide average with which local results can be compared. Also, they are useful to state-based organizations for planning strategies for improvement. All states improved on at least some of the quality indicators, demonstrating that improvement is possible despite geographic differences in population characteristics or health delivery systems. Continued efforts by interested partners to improve statewide performance and bring the lower-performing states closer to the higher-performing ones could decrease preventable post-AMI morbidity and mortality.

Interestingly, improvement on one indicator did not necessarily correlate with improvement on other indicators at the state-level. We would have expected that if an organized improvement effort were implemented focusing on all AMI care, this would have resulted in global improvement and high correlation among multiple indicators. Instead, among the 50 states, we found a significant correlation only between 1 pair of indicators (β-blocker prescribed early and at discharge). This suggests that improvement thus far has been sporadic or categorical, affecting only individual aspects of care, rather than systemic, which would simultaneously affect multiple aspects of care.

A number of events or interventions occurred during the period between 1994-1995 and 1998-1999 that may have contributed to the observed improvement in most indicators. These include the national Cooperative Cardiovascular Project launched by HCFA (with efforts by the PROs beginning in 1996), the publication of clinical guidelines for AMI care by the American College of Cardiology and American Heart Association in 1996,¹⁷ and dissemination of information from clinical trials and guidelines through various media or educational programs by professional societies. Other ongoing efforts included the commercially sponsored National Registry of Myocardial Infarction initiated in 1990 and the National Heart Attack Alert Program initiated by the National Heart, Lung and Blood Institute in 1991.¹⁸ Two efforts specifically focused on β-blockers included the adoption of β-blocker treatment after a heart attack as a Health Plan Employer Data and Information Set (HEDIS) measure for managed care, with initial reporting year 1996,¹⁹ and publication of a β-blocker alert by the American Medical Association in October 1998,²⁰ though this occurred in the midst of the remeasurement period, April 1998 through March 1999.

The pace of diffusion into practice of evidence-based therapies is not yet optimal. For example, although the substantial improvement in the use of β-blockers is an important success, it is notable that much of the evidence for their use in secondary prevention was established in the early 1980s.^21- 23 More than a decade and a half later, almost 30% of ideal candidates for treatment with β-blockers are still not receiving it at hospital discharge. Thus, the diffusion into practice has been quite delayed. If the current rate of improvement continues, 21% over 4 years, it will take another 6 years to achieve prescription of β-blockers at discharge for all ideal patients, although it is unlikely that improvement will continue in this linear fashion. The prescription of ACE inhibitors at discharge for patients with systolic dysfunction improved 8% nationally over 4 years among ideal candidates for therapy, or about 2 percentage points per year. If this rate continues, it will take another 15 years to achieve use of ACE inhibitors for all ideal patients. In the meantime, many patients will miss the opportunity for decreased morbidity and mortality that these therapies provide.

Several limitations should be considered when interpreting the data in the present report. First, the data were obtained using retrospective medical record review and thus only reflect care that was documented. Conditions or therapies that were not recorded were missed. This may be especially pertinent to the smoking cessation indicator and could result in underestimation of the amount of smoking cessation counseling provided to patients.

Second, although our methods include many aspects of medical record abstraction that might be considered "gold standard," including centralized abstraction, detailed ascertainment of clinical variables, and an ongoing data quality control process, certain challenges must be recognized that may have led to misclassification. For example, some patients' eligibility for acute reperfusion therapy was noted to be misclassified in targeted reviews of a small sample of records (HCFA, unpublished data, 1999). Athough the CDAC abstraction process incorporates approximately 24 variables to collect information from a single electrocardiogram (including measurement of ST elevation in each lead and key findings from the narrative interpretation), a variety of nuances can lead to misclassification. For example, an ST elevation might not be clinically relevant because it is old or related to intraventricular conduction delay or left ventricular hypertrophy. Interestingly, despite this limitation, the less than 2% increase nationally that we found for receipt of acute reperfusion therapy is consistent with the findings from the National Registry of Myocardial Infarction,⁸ although the increase was statistically significant in the latter study.

Misclassification of some patients as ideal candidates for early administration of β-blocker was noted where bradycardia or hypotension not present at the time of arrival at the hospital occurred soon after arrival. In general, because medical records are free-form narratives rather than standardized forms, it is not feasible to abstract every permutation of a clinical condition or contraindication. Also, abstraction reliability is high, but imperfect. The major effect of all these factors is to decrease the attainable target to somewhat lower than 100%, but this is unlikely to substantially affect the trend over time.

Third, we measured only a subset of AMI care and excluded the substantial number of patients transferred into or out of the hospital of interest. This might have introduced substantial selection bias. The measured care processes may or may not be a good proxy for the entirety of AMI care. However, they have been associated with reduced mortality in clinical trials²⁴ and clinical practice.^9,13,25- 26

Within this context, these national and state-level trends provide important feedback about progress and can be used by interested partners to help plan next steps and goals. During this 4-year period, important advances were made. However, diffusion of evidence-based therapies into practice can be a slow and uneven process, and the data do not support complacency. The observed variation in improvement highlights the need to better understand successful strategies. Starting in 1999, CMS's contracts with the PROs have included an increased emphasis on achieving statewide improvements in quality of care for AMI (as well as 5 other national clinical topics: heart failure, stroke, diabetes, pneumonia, and breast cancer) through collaborative efforts with the health care community.^4,27 Urgent efforts are needed to optimize translation and address preventable morbidity and mortality.

Corresponding author and reprints: JoAnne Micale Foody, MD, 330 Cedar St, FMP 315B, Yale University School of Medicine, New Haven, CT 06520 (e-mail: joanne.foody@yale.edu).

Accepted for publication August 30, 2002.

All funding for this project was provided by the CMS, Baltimore, Md, formerly the HCFA. Some of this work was performed under contract 500-99-CT02, titled "Utilization and Quality Control Peer Review Organization for the State of Connecticut," sponsored by the HCFA.

We thank the many current or former CMS staff members who contributed to this project, including Timothy Cuerdon, PhD; Thomas A. Marciniak, MD; Cynthia G. Wark, RN, MSN; and Edward F. Ellerbeck, MD, MPH. We also thank David R. Arday, MD, MPH, for contributions to the recent national sampling effort; James Michael, MS, for technical assistance; and Anita J. Bhatia, PhD, MPH, for technical assistance in the statistical analyses. In addition, we benefited from statistical discussions with Arlene Ash, PhD, Boston University Medical School. We thank the numerous individuals in the quality improvement organizations, the health care community, and professional societies who have contributed to the development and implementation of the Cooperative Cardiovascular Project and the National Acute Myocardial Infarction Project.

The opinions herein are those of the authors and not necessarily those of the CMS.