Identification of Hospital Cardiac Services for Acute Myocardial Infarction Using Individual Patient Discharge Data

Abstract

Background

The availability of hospital cardiac services may vary between hospitals and influence care processes and outcomes. However, data on available cardiac services are restricted to a limited number of services collected by the American Hospital Association (AHA) annual survey. We developed an alternative method to identify hospital services using individual patient discharge data for acute myocardial infarction (AMI) in the Premier Healthcare Database.

Methods and Results

Thirty‐five inpatient cardiac services relevant for AMI care were identified using American Heart Association/American College of Cardiology guidelines. Thirty‐one of these services could be defined using patient‐level administrative data codes, such as International Classification of Diseases, Ninth Revision, Clinical Modification and Current Procedural Terminology codes. A hospital was classified as providing a service if it had ≥5 instances for the service in the Premier database from 2009 to 2011. Using this system, the availability of these services among 432 Premier hospitals ranged from 100% (services such as chest X‐ray) to 1.2% (heart transplant service). To measure the accuracy of this method using administrative data, we calculated agreement between the AHA survey and Premier for a subset of 16 services defined by both sources. There was a high percentage of agreement (≥80%) for 11 of 16 (68.8%) services, moderate agreement for 3 of 16 (18.8%) services, and low agreement (≤50%) for 2 of 16 services (12.5%).

Conclusions

The availability of cardiac services for AMI care varies widely among hospitals. Using individual patient discharge data is a feasible method to identify these cardiac services, particularly for those services pertaining to inpatient care.

Introduction

In cardiovascular disease, the availability and utilization of individual services and procedures, such as percutaneous coronary intervention (PCI) and coronary artery bypass graft (CABG) surgery, vary widely between facilities.1, 2 However, a comprehensive assessment of a hospital's available cardiovascular services is presently lacking. Indeed, past assessments of cardiovascular services have been limited to basic surrogate measures, such as case volume, which have been shown to be associated with readmission rates, mortality, and costs.3 Examining the association between specific hospital measures of service capacity, such as the provision of in‐hospital procedures and secondary prevention measures, and outcomes is currently difficult, because we do not have methods to comprehensively identify services available at individual hospitals. In recognition of this, the American Heart Association/American College of Cardiology has stated that there is an “urgent need to develop measurement tools for the structure of [acute myocardial infarction] care, describe the reliability and validity of these tools, and link the results of these measurements to clinically relevant outcomes.”4

The American Hospital Association (AHA) annual survey is a source of national hospital structural characteristics that costs in excess of $5000 per year to purchase. Each year, over 6200 hospitals in the United States provide information about hospital characteristics and staffing in the AHA survey.5 Despite having a broad coverage of US hospitals, it is limited in its ability to fully characterize hospitals, because it is based on self‐reported data and only collects a limited range of cardiac services. Furthermore, studies that directly examine the reliability of the AHA annual and information technology (IT) surveys have been limited in quantity and variety of services.1, 6 A previous study has tested the reliability of the annual survey for identifying PCI services by utilizing individual patient administrative claims data from the Healthcare Cost and Utilization Project (HCUP). This study reported overall strong agreement for PCI capabilities between these 2 sources of data, with a kappa score of 0.70.1 This finding suggests that individual patient claims data can potentially be used to reliably identify services available at health care facilities. A claims‐based method is advantageous, because evidence of billing for a procedure or service is typically a reliable indicator that the service is provided by the facility. Moreover, such a method uses data that are already collected and have the potential to derive a service's utilization rate.

To our knowledge, administrative data have not been used to derive a wide range of acute myocardial infarction (AMI) services that encompass the emergency department (ED), in‐hospital, and secondary prevention services. Accordingly, we used individual patient hospital administrative data to determine the availability of a comprehensive range of cardiovascular services for the management of AMI among hospitals.

Methods

Data Sources

We derived administrative data from the Premier Healthcare Database, which contains data on more than 585 million cumulative discharges (75 million cumulative hospital inpatient discharges and 510 million outpatient discharges). There are ≈850 hospitals in Premier's research database, which vary in geography, bed size, and teaching status. In addition to information available in the standard hospital discharge file, the Premier database contains a date‐stamped log of all billed items at the patient level, including diagnostic tests, medications, and therapeutic services. This database has been used in previous studies to determine outcomes (costs, length of stay, mortality, and readmission) and treatment use at hospitals.7, 8, 9, 10

In addition, we used data from the 2010 AHA annual survey. This data source is updated annually and contains over 1000 data fields that include self‐reported descriptions of hospital organizational structures and services.5

Study Design

We included Premier hospitals with at least 25 AMI inpatient cases in a 3‐year period (2009–2011). AMI patients were identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) codes 410.xx, excluding those with 410.x2 (AMI, subsequent episode‐of‐care), from January 1, 2009 through December 31, 2011.

To identify essential services, we first determined the services required to deliver AMI care by reviewing class I, IIa, and IIb American Heart Association/American College of Cardiology clinical guidelines for ST segment elevated myocardial infarction and Non‐ST‐elevation acute coronary syndromes.11 Any service that was deemed necessary to provide each guideline recommendation was included in the analysis. The resulting 35 services were then grouped into the following 4 service categories: acute care, diagnostic and procedural care, inpatient management, and secondary prevention. Any prehospital services were excluded from the analysis.

We then determined the percentage of Premier hospitals that provide each service by using a combination of administrative codes contained within the individual patient files recorded in the data: ICD‐9‐CM, current procedural terminology (CPT), Medicare revenue, and healthcare provider taxonomy codes. In addition, Premier chargemaster data, Premier attending physician specialty, and Premier department codes were included because of their specific use in the Premier database and opportunity to use additional codes outside of what is traditionally available through Medicare claims data. We used multiple sources of codes to maximize the possibility that services could be identified using these administrative data. Specifically, we defined each cardiac service using 1 or more of the 7 coding systems depending on the service. The seven coding systems used, their reasons for use, and an example can be found in Table 1. A complete listing of the specific codes used for each service category can be found in Tables S1 through S4.

Table 1.

Administrative Data Sources and Reasoning for Their Inclusion

Source of Data	Reasoning for Inclusion	Example
ICD‐9‐CM procedure codes	To identify inpatient procedures	37.22 Left heart cardiac catheterization
CPT codes	To identify outpatient procedures and services	97003 Occupational therapy evaluation
Medicare revenue codes (identifies the location where procedures and services were delivered during the billing episode)	To identify hospital units and departments involved in providing care	0450 Emergency room‐ general classification
Health care provider taxonomy code set (indicates the provider's specialty)	To identify specialist provider services available at the hospital	208G00000X Thoracic surgery (cardiothoracic vascular surgery)
Premier chargemaster codesa (codes for all hospital procedures, services, supplies, and drugs that are billed during the episode of care)	To identify hospital procedures and services	270270008900000 Catheter arteriogram
Premier attending physician specialty codesa	To identify specialist provider services available at the hospital	4012 CDS
Premier department codesa	To identify hospital departments	450 Emergency room

CDS indicates cardiovascular surgery; CPT, current procedural terminology; ICD‐9‐CM, International Classification of Diseases, Ninth Revision, Clinical Modification.

^aIndicates coding sources specific to the Premier Healthcare Database.

We considered a hospital as offering a service if it had 5 or more records for that particular service, as defined by the claims codes, during the 3‐year study period (2009–2011). We used a threshold of 5 records to minimize the risk of hospitals being attributed a service because of an occasional inadvertent or erroneous documentation. For services that can be provided in both inpatient and outpatient settings, we counted services in both settings to decide the service availability.

Furthermore, we compared the availability of hospital AMI services derived from the administrative data method with the AHA survey. In this subset analysis, we included hospitals captured in both the Premier database and the 2010 AHA survey and used the AMI health services that overlapped between these 2 data sources. Last, to ascertain the relative benefit of Premier specific codes over Medicare codes alone, we repeated our analysis using only codes that are typically contained in Medicare claims data (ICD‐9‐CM, CPT, Medicare revenue, and taxonomy codes).

This study and waiver for informed consent was approved by Yale University's Institutional Review Board.

Study Assessment

Availability (frequency) of each service at the included Premier hospitals was calculated by dividing the number of hospitals defined as providing the service by the total number of included Premier hospitals. Additionally, for each individual hospital, we determined the number of available services in each of the 4 service categories (acute care, diagnostic and procedural care, inpatient management, and secondary prevention services).

For the subset analysis between Premier and the AHA survey, we determined the proportion of hospitals that showed agreement between these 2 data sources. This percentage of agreement examined concordant responses and was defined as instances in which both administrative data and the AHA survey identified a hospital as providing a service, in addition to instances in which both sources identified a hospital as not providing the service. We calculated agreement between the AHA survey and all 7 coding systems, as well as agreement between AHA survey and non‐Premier codes (ICD‐9‐CM, CPT, Medicare revenue, and taxonomy codes). When possible, the percentage of agreement was also quantified using the kappa statistic and McNemar's test.

Statistical Analysis

We presented categorical data as frequencies and percentages. Hospitals were identified as having a particular service or not having the service. Percentage agreement was calculated for the subset of services by assessing the percentage of hospitals that had concordant responses through Premier administrative data and the AHA survey out of the total number of hospitals. For services that did not have a zero cell for agreement or disagreement, percentage of agreement was also quantified with Cohen's kappa coefficient, with a kappa statistic of 1.0 indicating perfect agreement. Furthermore, the P value for McNemar's test for symmetry was computed to test whether the paired evaluations of a service by administrative data and the AHA survey for hospitals are agreeable. The null hypothesis in the McNemar's test is that the 2 evaluations are agreeable. A significant value (P<0.05) indicates that the AHA survey and Premier's evaluation of a service are statistically different.

Results

Data Sources

After applying our exclusion criteria, we included a total of 432 Premier hospitals for the analysis. Additionally, for the hospital service subset analysis between Premier and AHA, we had a hospital sample size that ranged from 345 to 370 hospitals, depending on the service.

Study Design

Of the 35 originally defined cardiac services necessary for AMI care, 31 could be defined using the administrative data sources (Table 2). The 4 services, and their respective categories, that could not be defined using our coding sources were as follows: primary PCI (diagnostic and procedural care); inpatient heart failure services (secondary prevention); outpatient heart failure services (secondary prevention); and readmissions prevention programs (secondary prevention). A subset of 16 services could be compared between the AHA survey and the Premier database.

Table 2.

AMI Services and Sources Used for Their Definition

Service Recommended for AMI Care	Source of Administrative Data
	ICD‐9‐CM Codes	CPT Codes	Medicare Revenue Codes	Taxonomy Codes	Premier Chargemaster Codes	Premier Attending Physician Specialty Codes	Premier Department Codes
Acute care
Cardiac biomarkers					X
Chest pain unit		X	X		X
Dedicated emergency department			X				X
Emergency department specialist				X		X
Pathology services			X
Thrombolysis					X
Diagnostic and procedural care
Cardiac nuclear perfusion imaging		X			X
Chest X‐ray	X	X			X
CABG	X	X			X
Coronary CT angiogram		X			X
Diagnostic coronary angiography	X	X			X
EP ablation	X	X			X
EP testing	X	X			X
Exercise stress testing	X	X			X
General CT	X	X			X
PCI	X	X			X
PPM/AICD implantation	X	X			X
Stress echocardiography		X			X
Transthoracic echocardiography	X	X			X
Primary PCI	Could not be identified using administrative data
Inpatient management
Coronary care unit			X	X	X
Coronary step‐down unit			X		X
Inpatient cardiac surgical service				X		X
Inpatient cardiology service				X		X
Intensive care unit	X	X	X		X
Inpatient internal medicine service				X		X
Transplant unit	X	X	X		X
Secondary prevention
Inpatient cardiac rehabilitation (phase I)	X		X		X
Outpatient cardiac rehabilitation (phase II or III)		X	X		X
Pharmacist		X	X		X
Physiotherapy/occupational therapy	X	X			X
Social worker	X	X			X
Inpatient heart failure services	Could not be identified using administrative data
Outpatient heart failure services	Could not be identified using administrative data
Readmissions prevention program	Could not be identified using administrative data

ACID indicates automatic implantable cardioverter defibrillator; AMI, acute myocardial infarction; CABG, coronary artery bypass grafting; CPT, Current Procedural Terminology; CT, computed tomography; EF, electrophysiology; ICD‐9‐CM, International Classification of Diseases, Ninth Revision, Clinical Modification; PCI, percutaneous coronary intervention; PPM, permanent pacemaker.

Study Assessment

The availability of the 31 services among the Premier hospitals varied between 100% (for services such as chest X‐ray) to 1.2% (transplant unit; Figure 1). For the analysis of agreement between the AHA survey and administrative data sources, a high percentage of agreement (≥80%) was noted for 11 of 16 (68.8%) services (Table 3). Moderate agreement was noted for 3 of 16 services (18.8%), and low agreement (<50%) was noted for 2 of 16 (12.5%) services. All acute care and diagnostic and procedural services had a high degree of agreement between administrative data and the AHA survey. The 2 low‐agreement services were physiotherapy/occupational therapy (41.4%) and presence of a social worker (6.7%), both of which were classified as secondary prevention services.

Figure 1.

Availability of cardiac services among Premier hospitals identified using administrative data. Thirty‐one cardiac services, which were divided into 4 categories (acute care, diagnostic and procedural care, inpatient management, and secondary prevention), were able to be defined using administrative data. Of the 432 Premier hospitals included in the analysis, the availability of services in hospitals ranged from 100% for several services to 1.2% for transplant units.

Table 3.

Percentage of Agreement and Disagreement Between Premier Administrative Data and AHA Annual Survey

Services Recommended for AMI Care	N (Hospitals With AHA Survey Data)	% Agree or Disagree				% Agreement	Cohen's Kappa (95% CI)	McNemar's Test for Symmetry P Valuea
		Agree: Yes on AHA Survey/Yes on Administrative Data (%)	Disagree: Yes on AHA Survey/No on Administrative Data (%)	Disagree: No on AHA Survey/Yes on Administrative Data (%)	Agree: No on AHA Survey/No on Administrative Data (%)
Acute care
Dedicated ED	345	99.7	0.3	0.0	0.0	99.7	N/A	N/A
Diagnostic and procedural
Cardiac nuclear perfusion imaging	345	94.8	5.2	0.0	0.0	94.8	N/A	N/A
General CT	345	98.8	0.3	0.9	0.0	98.8	N/A	0.32
Diagnostic coronary angiography	345	76.5	2.0	2.0	19.4	95.9	0.88 (0.82, 0.94)	1.00
PCI	345	64.6	0.9	4.3	30.1	94.8	0.88 (0.83, 0.93)	0.01
EP testing	345	40.6	15.4	1.7	42.3	82.9	0.66 (0.59, 0.74)	<0.01
Inpatient management
Coronary care unit	345	34.8	20.6	7.5	37.1	71.9	0.45 (0.36, 0.54)	<0.01
Inpatient cardiac surgical service	345	6.4	45.5	0.3	47.8	54.2	0.11 (0.06, 0.16)	<0.01
Inpatient cardiology service	345	68.4	14.2	5.8	11.6	80.0	0.42 (0.30, 0.53)	<0.01
Intensive care unit	345	97.4	0.3	2.0	0.3	97.7	0.19 (−0.14, 0.52)	<0.01
Inpatient internal medicine service	320	88.1	1.9	9.1	0.9	89.1	0.11 (−0.40, 0.25)	<0.01
Transplant unit	345	0.0	2.6	0.0	97.4	97.4	N/A	N/A
Secondary prevention
Inpatient cardiac rehabilitation (phase 1)	345	51.0	25.8	1.7	21.4	72.5	0.43 (0.25, 0.52)	<0.01
Pharmacist	370	98.6	0.0	1.4	0.0	98.6	N/A	N/A
Physiotherapy/Occupational therapy	345	41.4	0.0	58.6	0.0	41.4	N/A	N/A
Social worker	345	3.8	92.8	0.6	2.9	6.7	−0.01 (−0.03, 0.01)	<0.01

CT indicates computed tomography; ED, emergency department; EP, electrophysiology; PCI, percutaneous coronary intervention.

^aStatistically significant at P<0.05. AHA indicates American Hospital Association; AMI, acute myocardial infarction.

For 9 of 16 services, the more‐common type of disagreement stemmed from hospitals identified as having the service on the AHA survey, but not through administrative data (yes on AHA survey/no based on administrative data), compared to hospitals identified as having the service through administrative data but not the AHA survey (no on AHA survey/yes using administrative data; Table 3).

We examined each hospital's proportion of available services according to the 4 service categories (Figure 2). Across the 432 hospitals included in the analysis, the total number of available services ranged from 6 to 29 (of the 31 total services). The number of acute care services remained relatively stable across all hospitals. The proportion of inpatient management and secondary prevention services increased slightly as the number of total services increased for hospitals. The most significant contributor to the gradient in the number of services available at individual hospitals was attributed to diagnostic and procedural services (range, 1–13).

Figure 2.

Number of services available in each service category in Premier hospitals. For each of the 432 hospitals, the total number of available services was determined according to the type of service category (acute care, diagnostic and procedural care, inpatient management, and secondary prevention). The total available services ranged from 6 to 29 for each hospital, with diagnostic and procedural care services contributing most significantly to the wide range of available services across hospitals.

Agreement of Administrative Data With the AHA Survey

Cohen's kappa coefficient could be calculated for 10 of the 16 services and ranged from −0.01 to 0.88 (Table 3). Of the 6 services that did not have a kappa coefficient calculated, 5 of the services had a high degree of agreement (>94.8%), and 1 service (physiotherapy/occupational therapy) had a low degree of agreement (41.4%). The P value for McNemar's test for symmetry could be determined for 11 of the services and was statistically significant at P<0.05 for 9 of these services (Table 3). Even though this indicates statistical disagreement between the sources for those services, coupled with the results with the kappa coefficient and percentage agreement, we see that, in practice, there is relatively good agreement.

When services were defined using codes typically found in Medicare claims alone, excluding Premier specific codes, we found that agreement remained the same for some services, but agreement sensitivity decreased for 7 of 16 services (Table S5). Additionally, when we examined agreement by AMI discharge volume (≤400 AMI discharges from 2009 to 2011 vs >400 discharges), we did not find statistically significant differences in agreement between smaller‐ and larger‐volume hospitals (data not shown).

Discussion

In this study, we show that for a majority of the services, discharge data can be a reasonable proxy for direct survey. We found a high degree of agreement (≥80%) between administrative data and the AHA survey for many services, but also identified some services for which there was common discord. Nevertheless, the method using administrative data is a feasible way to profile most inpatient hospital cardiac services among hospitals and detect variation in service availability between hospitals. The ability to use administrative data is valuable, because it is accessible, can capture a wide range of services, is nationally representative data, and can be used to calculate service utilization rates in hospitals.

In previous studies, the Premier Healthcare Database has been used to identify utilization rates of individual health services, such as patterns of intensive care unit utilization among heart failure patients12 and computed tomography (CT) angiogram and perfusion utilization among acute ischemic stroke patients.13 Additionally, Concannon et al.1 used claims data from the HCUP's Inpatient Database to identify PCI services at hospitals and compare claims‐based utilization rates with those of the AHA survey. In contrast to these previous studies, we extended the current literature by developing a systematic method to comprehensively identify a wide range of cardiac health services using administrative data. To our knowledge, no previous studies have classified a comprehensive set of services for AMI care using this approach. Our study also found a higher degree of agreement for PCI when compared with the results from Concannon et al.1 (κ=0.88 vs 0.70). Furthermore, the method that we developed can be potentially extended to other conditions and databases.

The AHA survey is the most widely used and is currently one of the only nationally representative sources of hospital infrastructure data. However, it has several limitations. Although it is relied upon by health care systems nationwide as the source of truth for structural questions about hospitals, data in the survey are self‐reported by hospitals, only capture a subset of specialized cardiac services (eg, cardiac intensive care, cardiac surgery, cardiac electrophysiology, magnetic resonance imaging, and CT scanner), are costly to purchase, and are not validated.

Although we found an overall high degree of agreement between administrative data and the AHA survey, agreement was not consistent across all services. It is possible that available services in the AHA survey are over‐reported by hospitals. Additionally, we defined service availability as hospitals with ≥5 service codes during our time period. Thus, if hospitals have the capacity for the service, but a low utilization rate, they would be incorrectly classified as not providing the service. Finally, differences between the AHA survey's definition for data elements, the interpretation of survey questions by the hospital personnel completing the survey, and our method using administrative data could explain our discordances. For example, social work services had a low degree of agreement (6.7%), which may be attributed to a different definition and interpretation from the AHA survey (includes counseling for social, emotional, and environmental problems), compared to the health and behavior therapy services we were able to utilize from administrative data.

Our findings indicate that administrative data can be used for identifying the availability of services at hospitals. This method gives us the ability to identify cardiovascular services at hospitals, which can drive further research to explore the relationship between structures, outcomes, and processes. We designed our study to use data from the Premier Healthcare Database instead of Medicare claims data because they have extra information outside of more traditionally used ICD‐9‐CM and CPT codes found in Medicare data. These additional codes available in Premier include the Premier chargemaster, attending physician specialty, and department codes. Our finding that agreement sensitivity decreases for 7 of 16 services when only non‐Premier codes were used indicates that supplementing more widely available coding sources with Premier codes may help the accuracy of identifying some cardiac services.

This study has a few limitations. First, our sources of data could not define primary PCI, inpatient heart failure services, outpatient heart failure services, or readmissions prevention programs. Three of these 4 services were secondary prevention services. Among the secondary prevention services that could be identified using the data, the agreement between administrative data and the AHA survey was low. However, for other service categories, we had a high degree of agreement between the AHA survey and administrative data. Another limitation is that Premier is a performance improvement database, which means that the results may not be generalizable to all hospitals if medical coding is more accurate at these hospitals that strive to improve quality. Despite this limitation, it is important to note that the Premier Healthcare Database is a large database that includes a wide range of hospitals. Finally, this method does not allow us to discern whether services with low degrees of agreement are attributed to an error in the administrative data or the AHA survey.

In conclusion, this method of using individual administrative discharge data to define hospital services is feasible for a majority of cardiovascular services. This provides the ability for future research to use readily available data to link structural characteristics with processes and outcomes. Future research should test this methodology to see whether this can be extended to hospital services for other diseases.

Conclusion

We developed a method to identify hospital cardiac services using administrative data available in the Premier Healthcare Database. When we compared the availability of services between our administrative data method and data from the AHA survey, we found a high or moderate degree of agreement for 14 of the 16 included services. Our analysis showed that it is feasible to use administrative data to identify AMI services, especially inpatient services.

Sources of Funding

This work was supported by grant DF10‐301 from the Patrick and Catherine Weldon Donaghue Medical Research Foundation in West Hartford, Connecticut, and by grant UL1 RR024139‐06S1 from the National Center for Advancing Translational Sciences in Bethesda, Maryland. Dr Krumholz is supported by grant U01 HL105270‐05 (Center for Cardiovascular Outcomes Research at Yale University) from the National Heart, Lung and Blood Institute in Bethesda, Maryland. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Donaghue Foundation or the National Institutes of Health.

Disclosures

Drs Krumholz and Li work under contract with the Centers for Medicare & Medicaid Services to develop and maintain performance measures. Dr Krumholz is the recipient of research agreements from Medtronic and from Johnson & Johnson, through Yale University, to develop methods of clinical trial data sharing and chairs a cardiac scientific advisory board for United Healthcare. Dr Ranasinghe is supported by an Early Career Fellowship cofunded by the National Health and Medical Research Council and the National Heart Foundation of Australia.

Supporting information

Table S1. Codes Used to Identify Acute Care Services

Table S2. Codes Used to Identify Diagnostic and Procedural Services

Table S3. Codes Used to Identify Inpatient Management Services

Table S4. Codes Used to Identify Secondary Prevention Services

Table S5. Percentage Agreement and Disagreement Between Administrative Data (Non‐Premier Codes) and AHA Annual Survey

(J Am Heart Assoc. 2016;5:e003680 doi: 10.1161/JAHA.116.003680)