Each data point represents 1 hospital, and the trend line represents the regression line of imaging rate vs rank of imaging use rate.
Each bar represents 1 hospital. CT indicates computed tomography; MRI, magnetic resonance imaging; PET, positron emission tomography.
eTable 1. ICD-9-CM Codes Selected for Inclusion
eTable 2. ICD-9-CM Diagnosis Codes Selected for Exclusion
eTable 3. ICD-9-CM Volume 3 Procedure Codes Selected for Exclusion
eTable 4. Percent of Patients With ICD-9-CM Discharge Diagnosis
eFigure. Patients Included in the Study Cohort
Safavi KC, Li S, Dharmarajan K, Venkatesh AK, Strait KM, Lin H, Lowe TJ, Fazel R, Nallamothu BK, Krumholz HM. Hospital Variation in the Use of Noninvasive Cardiac Imaging and Its Association With Downstream Testing, Interventions, and Outcomes. JAMA Intern Med. 2014;174(4):546-553. doi:10.1001/jamainternmed.2013.14407
Copyright 2014 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Current guidelines allow substantial discretion in use of noninvasive cardiac imaging for patients without acute myocardial infarction (AMI) who are being evaluated for ischemia. Imaging use may affect downstream testing and outcomes.
To characterize hospital variation in use of noninvasive cardiac imaging and the association of imaging use with downstream testing, interventions, and outcomes.
Design, Setting, and Participants
Cross-sectional study of hospitals using 2010 administrative data from Premier, Inc, including patients with suspected ischemia on initial evaluation who were seen in the emergency department, observation unit, or inpatient ward; received at least 1 cardiac biomarker test on day 0 or 1; and had a principal discharge diagnosis for a common cause of chest discomfort, a sign or symptom of cardiac ischemia, and/or a comorbidity associated with coronary disease. We excluded patients with AMI.
Main Outcomes and Measures
At each hospital, the proportion of patients who received noninvasive imaging to identify cardiac ischemia and the subsequent rates of admission, coronary angiography, and revascularization procedures.
We identified 549 078 patients at 224 hospitals. The median (interquartile range) hospital noninvasive imaging rate was 19.8% (10.9%-27.7%); range, 0.2% to 55.7%. Median hospital imaging rates by quartile were Q1, 6.0%; Q2, 15.9%; Q3, 23.5%; Q4, 34.8%. Compared with Q1, Q4 hospitals had higher rates of admission (Q1, 32.1% vs Q4, 40.0%), downstream coronary angiogram (Q1, 1.2% vs Q4, 4.9%), and revascularization procedures (Q1, 0.5% vs Q4, 1.9%). Hospitals in Q4 had a lower yield of revascularization for noninvasive imaging (Q1, 7.6% vs Q4, 5.4%) and for angiograms (Q1, 41.2% vs Q4, 38.8%). P <.001 for all comparisons. Readmission rates to the same hospital for AMI within 2 months were not different by quartiles (P = .51). Approximately 23% of variation in imaging use was attributable to the behavior of individual hospitals.
Conclusions and Relevance
Hospitals vary in their use of noninvasive cardiac imaging in patients with suspected ischemia who do not have AMI. Hospitals with higher imaging rates did not have substantially different rates of therapeutic interventions or lower readmission rates for AMI but were more likely to admit patients and perform angiography.
Chest pain is the second most common cause of emergency department visits in the United States and accounts for billions of dollars in annual hospital costs.1,2 Most patients who present with chest pain do not have acute myocardial infarction (AMI).3 During the initial evaluation of chest pain, clinicians have many options, including the use of stress testing with noninvasive cardiac imaging, to determine patients’ risk for a cardiovascular event. Unfortunately, clinical guidelines often do not provide strong recommendations about which testing strategies should be applied to which patients among the clinically heterogeneous group that presents with suspected ischemia.4,5
To our knowledge, there are no published studies that compare hospitals’ approaches to the evaluation of patients presenting with possible ischemic heart disease who do not have AMI. In particular, physicians face a decision about the use of noninvasive imaging in stress testing. Imaging studies result in substantial health care costs and often possible radiation exposure.6 To justify their use, imaging tests should therefore lead to better decisions and better patient outcomes.7 Hospitals have developed different diagnostic approaches, including the creation of units dedicated to assessing suspected myocardial ischemia with either mandatory or optional noninvasive cardiac imaging.7- 11
The implications of variation may be far-reaching. Patient-level studies have shown that frequent use of noninvasive cardiac imaging is associated with a greater likelihood of invasive and expensive downstream tests, such as catheterization for coronary angiography.12,13 Evidence of marked hospital-level variation in cardiac imaging practices would highlight a need to clarify the marginal benefit of more expensive strategies for patients with suspected ischemia.
We sought to determine whether hospital use of noninvasive cardiac imaging was associated with patterns of downstream resource use, including inpatient hospitalization and catheterization for coronary angiography. Furthermore, we sought to determine whether hospitals that frequently used noninvasive cardiac imaging subsequently performed revascularization procedures at a substantially different rate. We also sought to determine whether hospitals that more frequently used noninvasive imaging had a different yield of revascularization for both noninvasive imaging testing and angiograms. Finally, we investigated whether more frequent use of cardiac imaging among hospitals was ultimately associated with fewer readmissions for AMI to the same hospital.
We conducted a cross-sectional study of acute care hospitals in the United States using a database maintained by Premier, Inc. The Premier database includes administrative, operational, and some clinical data from 2700 hospitals in the United States.14 Of the 2700 hospitals that submitted data, 372 hospitals signed an agreement to have their data used for the purposes of academic research and were eligible for inclusion in our study. The database contains a date-stamped log of all billed items at the individual patient level including medications and laboratory, diagnostic, and therapeutic services. For this study, Premier deidentified patient data in accordance with the Health Insurance Portability and Accountability Act, and random, unique patient and hospital identifiers were applied to each record to facilitate analysis. The Yale University Human Investigation Committee exempted this study protocol from review by the Office of Human Research Protections.
Quiz Ref IDWe included hospital visits that occurred in 2010, including those restricted to the emergency department, those in which the patient stayed in a bed coded as observation status, and those in which a patient was admitted to an inpatient bed. In order to include patients in whom cardiac ischemia was being considered as a primary diagnosis, we required patients to have received at least 1 cardiac biomarker test to assess for myocardial ischemia, defined as either a serum cardiac troponin or creatine kinase MB test on day 0 (day of the initial encounter) or day 1 of the hospital visit.
In addition, patients were included if the principal discharge diagnosis was among a selected group of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) discharge diagnoses (eTable 1 in Supplement). These codes were selected on the basis of a review by 3 of us (K.C.S., K.D., A.K.V.) to determine whether they met any of the following criteria: (1) a common cause of chest discomfort; (2) a symptom or sign that can be associated with cardiac ischemia; or (3) a diagnosis that is often comorbid with coronary artery disease. Patients with a principal discharge diagnosis of AMI were excluded.
We excluded patients with principal discharge diagnoses for acute illnesses, which may warrant cardiac biomarker testing but not additional ischemic evaluation, including aortic dissection and pulmonary embolism (eTable 2 in Supplement). We excluded visits if patients were younger than 35 years, transferred in from another hospital, or transferred out to another hospital. We also excluded patient visits in which cardiac biomarkers may have been ordered for preoperative or perioperative evaluation (eTable 3 in Supplement). We excluded hospitals that performed fewer than 5 revascularization procedures during the study period.
The primary outcome for each hospital was the proportion of eligible patients who received noninvasive cardiac imaging testing for ischemia. These tests included stress nuclear myocardial perfusion imaging, stress echocardiography, cardiac positron emission tomography, cardiac magnetic resonance imaging, and cardiac computed tomography with coronary angiography (CTCA) with or without calcium scoring.
To determine the association of noninvasive imaging use with downstream resource use, we calculated subsequent hospital-level proportions of inpatient hospitalization, coronary angiography, and revascularization (either percutaneous coronary intervention or coronary artery bypass graft surgery).
We considered cardiac imaging tests, catheterizations for coronary angiography, and revascularization procedures as outcomes if they occurred within the same month or in the month following the index hospital visit. Finally, we calculated readmissions for AMI as the proportion of patients who returned to the same hospital with a principal discharge diagnosis of AMI (ICD-9-CM codes 410.x, excluding 410.x2) within the same month or in the month following the index hospital visit.
We divided hospitals into quartiles based on the proportion of patients who received noninvasive imaging tests for cardiac ischemia. We compared quartiles of hospitals using χ2 tests. We compared the proportion of patients receiving specific imaging modalities across quartiles. We determined the association of noninvasive imaging with downstream resource utilization across quartiles by calculating the number of patients in each quartile who were admitted to an inpatient bed during their index hospital encounter, received catheterization for a coronary angiogram after undergoing noninvasive imaging, and received a revascularization procedure. We determined the association of imaging with outcomes by using χ2 tests to compare, across quartiles, the proportion of patients who were readmitted with AMI to the same hospital as that of their index encounter during the same month or the month following the index hospital visit. For all analyses, we considered P < .05 to be statistically significant.
We also fit 3 hierarchical logistic regression models to investigate the degree to which patient case mix and hospital structural factors accounted for the variation in use of noninvasive imaging. We first fit an unconditional model with only a hospital random intercept to determine whether any hospital-level variation existed. In the second model, we added patient characteristics, including age and sex, to investigate the extent to which differences in cardiac imaging use across hospitals were accounted for by patient case mix. In the third model, we added hospital characteristics to determine whether the variation among hospitals could be accounted for by known hospital characteristics such as region, teaching status, or bed size. For each of the models, we calculated the between-hospital variance and reported a C statistic as a measure of model discrimination. We then derived the interclass correlation coefficient (ICC) to measure the proportion of variance that was attributable to the between-hospital variation and the median odds ratio to quantify the between-hospital variation in the use of imaging.
We included 549 078 patients treated at 224 hospitals in the United States (eFigure 1 in Supplement). The median (interquartile range [IQR]) number of hospital beds was 360 (250-462). Hospitals tended to be located in the South (41 ), serve an urban population (88%), and identify as nonteaching (66%).
The noninvasive cardiac imaging rates for hospitals ranged from 0.2% to 55.7% (median [IQR], 19.8% [10.9%-27.7%]) (Figure 1). When hospitals were divided into quartiles based on imaging rates, median rates of imaging were Q1, 6.0%; Q2, 15.9%; Q3, 23.5%; Q4, 34.8%.
Hospitals in higher quartiles of noninvasive imaging use had similar characteristics in terms of bed size (P = .13), availability of observation beds (P = .87), urban or rural location (P = .93), and teaching status (P = .44) (Table 1). Hospitals in higher imaging quartiles were more likely to be located in the Midwest and Northeast (P < .001) (Table 1).
Quiz Ref IDMyocardial perfusion imaging and stress echocardiography were the most commonly used imaging modalities (Figure 2). A total of 113 602 imaging tests were performed in our study cohort. Of these, 80.4% were myocardial perfusion images (91 336), 16.6% were echocardiograms (18 858), and 1.2% were CTCAs (1365). There were no statistically significant differences between hospitals in different imaging quartiles in terms of the rates at which they used specific modalities when performing noninvasive cardiac imaging (Table 2).
Hospitals in higher imaging use quartiles admitted a larger proportion of patients to an inpatient bed (Q1, 32.1% vs Q4, 40.0%; P < .001) (Table 3).
Hospital angiogram rates ranged from 0% to 16.9% (median [IQR], 2.5% [1.4%-4.1%]). There was only 1 hospital with an angiogram rate greater than 10%. Hospitals in higher imaging use quartiles more frequently performed downstream coronary angiography (Q1, 1.2% vs Q4, 4.9%; P < .001) (Table 3).
The rate of revascularization procedures ranged from 0% to 9.5% (median [IQR], 0.8% [0.4%-1.4%]). Hospitals in higher imaging use quartiles performed slightly more revascularization procedures compared with those in lower imaging quartiles (Q1, 0.5% vs Q4, 1.9%; P < .001) (Table 3). Hospitals in the top imaging quartile had a lower revascularization yield per imaging study compared with hospitals in the lowest quartile (Q1, 7.6% vs Q4, 5.4%; P < .001) (Table 3). Similarly, the yield in revascularizations per angiogram was also lower at hospitals in the highest imaging quartile compared with hospitals in the lowest quartile (Q1, 41.2% vs Q4, 38.8%; P < .001) (Table 3).
All quartiles had a similar proportion of patients readmitted for AMI within the same or the subsequent month (0.3% for all quartiles; P = .51) (Table 3).
The ICCs (95% CI) for the unconditional (unadjusted) model, the model adjusted for patient age and sex, and the model further adjusted for hospital characteristics were 23.2 (19.5-26.5), 23.3 (19.6-26.6), and 16.1 (8.6-22.4), respectively (Table 4), indicating that approximately 23% of the variation in rates of noninvasive cardiac imaging was attributable to between-hospital variation and that this variation was not significantly affected by hospital patient mix. The median odds ratio was 2.6 for the unconditional model and for the age and sex–adjusted model and 2.1 for the model with hospital characteristics, indicating that a randomly selected patient receiving noninvasive cardiac imaging at a particular hospital would have approximately 2-fold higher odds of receiving an imaging test than an identical patient at a different randomly selected hospital.
Our study indicates substantial hospital-level variation in the use of noninvasive imaging for cardiac ischemia in patients without AMI who are undergoing further evaluation for coronary artery disease. Quiz Ref IDHigher rates of imaging use at the hospital level were associated with greater use of downstream coronary angiography. However, we observed that hospitals that more frequently engaged in testing did not have substantially higher rates of revascularization; hospitals in the highest imaging use quartile had the lowest yields with regard to revascularization. Furthermore, high imaging use hospitals did not achieve lower rates of same-hospital readmission with AMI. These findings suggest that more frequent use of noninvasive cardiac imaging at a large, diverse group of hospitals in the United States was associated with greater rates of inpatient admission and use of invasive downstream diagnostic tests without evidence of a substantial effect on the use of therapeutic interventions or short-term outcomes.
We could not directly determine whether imaging was appropriate for an individual patient. Our aim was to characterize hospital-level patterns of imaging use rather than to determine the appropriateness of individual provider decisions, and it is unlikely that patient case mix would account for the wide variation in imaging rates among hospitals that we observed. Furthermore, high imaging use hospitals had a lower yield to angiogram and coronary revascularization and had the same readmission outcomes compared with low imaging use hospitals, suggesting the need to clarify the benefit to patients of higher rates of imaging use.
Our results extend the literature that demonstrates variation in the use of diagnostic modalities in the evaluation of common acute clinical presentations.15- 17 Studies have shown particularly high variation in clinical scenarios in which guidelines are not well established, including neuroimaging in the evaluation of dizziness, headache, trauma, and epilepsy. For subgroups of patients in which guidelines are well established, such as pediatric neuroimaging, variation is considerably less. In comparison, clinical guidelines do not clearly identify which patients among the heterogeneous group presenting with suspected myocardial ischemia should receive cardiac imaging.
Moreover, our findings are similar to those of studies examining the relationship between increased use of a diagnostic modality and patient outcomes. For example, despite the increasing rates of imaging with computed tomography for stroke and pulmonary embolism, rates of diagnosis have not changed.15,17 One potential explanation for unchanged outcomes despite higher rates of cardiac imaging is that hospitals that use imaging more frequently are doing so in patients for whom the benefit is not clear. Again, without a strong evidence base to inform guideline development, the choice of which patients are likely to benefit from imaging may not be readily apparent.
Despite uncertainty in the value of testing, single-center studies have demonstrated that patients frequently receive noninvasive cardiac imaging for ischemia, regardless of pretest probability of coronary artery disease.18 However, when imaging is used in a broader patient population, false-positive results will increase.19 These results can increase referral for angiograms that subsequently reveal clinically insignificant lesions. Our findings are consistent with this relationship because hospitals that performed noninvasive imaging in the highest proportion of their patients had higher rates of angiography but the lowest yields in subsequent revascularization. These findings indicate the need to clarify which patients being evaluated for cardiac ischemia would most benefit from additional imaging studies.20,21
The yield of revascularization for noninvasive imaging and for angiography was derived by grouping all patients from hospitals within the same imaging quartile together and calculating the mean yield. Thus, we believe that the yield associated with each quartile was not sensitive to an individual hospital’s or provider’s tendency to perform angiography or revascularization. The goal of our study was not to characterize the practice of an individual hospital or provider as being more or less aggressive or higher or lower in quality.
In an attempt to understand the factors that may be responsible for variation in imaging practices, we conducted a hierarchical logistical regression analysis. Quiz Ref IDWe found that between-hospital differences in patient case mix did not account for the majority of the variation. Rather, our study demonstrates that the likelihood of a given testing strategy being used for an individual patient depends largely on the particular hospital at which the patient receives care.11,18 However, additional work is necessary to understand the organizational, cultural, and financial aspects of hospital practice that may influence imaging decisions.11 Previous literature suggests that many hospitals may use specific protocols to evaluate patients with suspected ischemia, including the standard use of cardiac imaging to rule out coronary disease.8- 10 Other studies have demonstrated increases in utilization and spending in physician practices that have purchased expensive imaging equipment.22 It is plausible that hospitals may experience similar financial motivations.
Although we included a diverse cohort of hospitals in this study, there may be some questions about the generalizability of our sample. Our cohort of patients was highly specific for individuals who are evaluated for cardiac ischemia. Because the Premier database indicates whether a patient received a cardiac troponin test, we avoided the selection bias and misclassification of patients that could result from solely relying on principal discharge diagnoses to identify those with suspected ischemia. We included a non–chest pain code if the code represented a common cause of chest pain, a symptom commonly associated with chest pain, and/or a comorbidity that may cause concern for occlusive coronary disease. Thus, we believe that our cohort represents the full spectrum of patients in whom cardiac ischemia may be considered as a primary diagnosis.
Our study has several limitations. We were not able to track whether a hospital referred a patient for cardiac imaging to a different hospital or private physician office. However, it is unlikely that hospitals with the capability to perform cardiac imaging (all of the hospitals in our study) would refer patients to another hospital for an imaging study. Moreover, patients may seek care at multiple hospitals, and we were not able to capture downstream testing and revascularization if it occurred at a hospital different from that of the index encounter. In addition, although our hospital cohort included more than 220 institutions with diverse characteristics, all voluntarily participate in a consortium that gathers and shares data with the aim of improving practice. Quiz Ref IDWe could also only capture readmissions for AMI if the patient was readmitted to the same hospital at which he or she was initially evaluated. Although this may mean that readmission rates were in fact higher than what we report, it is unlikely that the relative readmission rates between hospitals would differ. Furthermore, it is possible that the threshold that individual health care providers use to decide whether to order tests of cardiac biomarkers may have indirectly affected the rate of cardiac imaging use by altering the types of patients included in the cohort at each hospital. We believe that this possibility is unlikely given that the relative mix of discharge diagnoses was similar across the hospital quartiles (eTable 4 in Supplement).
In a health care system that faces increasing resource constraints, it is critical to identify opportunities to reduce resource use that is not associated with improved patient outcomes. We observed substantial variation in the use of cardiac imaging in patients who present for evaluation of ischemia but much smaller differences in the rates of revascularization and no difference in readmission outcomes. An important determinant of whether these expensive tests were used was the hospital at which the patient received care.
Accepted for Publication: November 29, 2013.
Corresponding Author: Harlan M. Krumholz, MD, SM, Section of Cardiovascular Medicine, Department of Internal Medicine, Yale University School of Medicine, 1 Church St, Ste 200, New Haven, CT 06510 (firstname.lastname@example.org).
Published Online: February 10, 2014. doi:10.1001/jamainternmed.2013.14407.
Author Contributions: Dr Li had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Safavi, Li, Dharmarajan, Venkatesh, Krumholz.
Acquisition of data: Krumholz.
Analysis and interpretation of data: All authors.
Drafting of the manuscript: Safavi, Li, Venkatesh, Lowe.
Critical revision of the manuscript for important intellectual content: Li, Dharmarajan, Venkatesh, Strait, Lin, Fazel, Nallamothu, Krumholz.
Statistical analysis: Safavi, Li, Strait, Lin, Lowe.
Obtained funding: Krumholz.
Administrative, technical, and material support: Krumholz.
Study supervision: Dharmarajan, Venkatesh, Krumholz.
Conflict of Interest Disclosures: Dr Krumholz is the recipient of a research grant from Medtronic, through Yale University, to develop methods of clinical trial data sharing and chairs a cardiac scientific advisory board for UnitedHealth.
Funding/Support: This work was supported by grant DF10-301 from the Patrick and Catherine Weldon Donaghue Medical Research Foundation in West Hartford, Connecticut; grant UL1 RR024139-06S1 from the National Center for Advancing Translational Sciences in Bethesda, Maryland; and grant U01 HL105270-04 (Center for Cardiovascular Outcomes Research at Yale University) from the National Heart, Lung, and Blood Institute in Bethesda, Maryland. At the time this study was conducted, Dr Dharmarajan was supported by grant HL007854 from the National Institutes of Health; he was also supported as a Centers of Excellence Scholar in Geriatric Medicine at Yale by the John A. Hartford Foundation and the American Federation for Aging Research.
Role of the Sponsors: The sponsors had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.