CABG indicates coronary artery bypass graft; MI, myocardial infarction; PCI, percutaneous coronary intervention.
Stress testing rates range from 22.1% to 70.6%, with a national mean of 44.5% (interquartile range, 30.9%-50.9%). Individual hospital referral region rates were converted into rate ratios based on the national mean. Hospital referral regions with less than 25 percutaneous coronary interventions in the 20% sample were excluded from the analysis (gray).
Results from a multivariate mixed-effects model adjusting for patient and physician characteristics and hospital referral regions. AOR indicates adjusted odds ratio; CI, confidence interval; and COPD, chronic obstructive pulmonary disease. Error bars indicate 95% CIs.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Lin GA, Dudley RA, Lucas FL, Malenka DJ, Vittinghoff E, Redberg RF. Frequency of Stress Testing to Document Ischemia Prior to Elective Percutaneous Coronary Intervention. JAMA. 2008;300(15):1765–1773. doi:10.1001/jama.300.15.1765
Author Affiliations: Divisions of General Internal Medicine (Dr Lin), Epidemiology and Biostatistics (Dr Vittinghoff), and Cardiology (Dr Redberg), and Institute for Health Policy Studies (Dr Dudley), University of California, San Francisco; Center for Outcomes Research and Evaluation, Maine Medical Center, Portland (Dr Lucas); and Division of Cardiology, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire (Dr Malenka).
Context Guidelines call for documenting ischemia in patients with stable coronary artery disease prior to elective percutaneous coronary intervention (PCI).
Objective To determine the frequency and predictors of stress testing prior to elective PCI in a Medicare population.
Design, Setting, and Patients Retrospective, observational cohort study using claims data from a 20% random sample of 2004 Medicare fee-for-service beneficiaries aged 65 years or older who had an elective PCI (N = 23 887).
Main Outcome Measures Percentage of patients who underwent stress testing within 90 days prior to elective PCI; variation in stress testing prior to PCI across 306 hospital referral regions; patient, physician, and hospital characteristics that predicted the appropriate use of stress testing prior to elective PCI.
Results In the United States, 44.5% (n = 10 629) of patients underwent stress testing within the 90 days prior to elective PCI. There was wide regional variation among the hospital referral regions with stress test rates ranging from 22.1% to 70.6% (national mean, 44.5%; interquartile range, 39.0%-50.9%). Female sex (adjusted odds ratio [AOR], 0.91; 95% confidence interval [CI], 0.86-0.97), age of 85 years or older (AOR, 0.83; 95% CI, 0.72-0.95), a history of congestive heart failure (AOR, 0.85; 95% CI, 0.79-0.92), and prior cardiac catheterization (AOR, 0.45; 95% CI, 0.38-0.54) were associated with a decreased likelihood of prior stress testing. A history of chest pain (AOR, 1.28; 95% CI, 1.09-1.54) and black race (AOR, 1.26; 95% CI, 1.09-1.46) increased the likelihood of stress testing prior to PCI. Patients treated by physicians performing 150 or more PCIs per year were less likely to have stress testing prior to PCI (AOR, 0.84; 95% CI, 0.77-0.93). No hospital characteristics were associated with receipt of stress testing.
Conclusion The majority of Medicare patients with stable coronary artery disease do not have documentation of ischemia by noninvasive testing prior to elective PCI.
In the United States, percutaneous coronary intervention (PCI) has become a common treatment strategy for patients with stable coronary artery disease (CAD) and such patients now account for the majority of PCIs performed.1,2 However, multiple studies have established that some important outcomes for patients with stable CAD (death and risk of future myocardial infarction) do not differ between patients treated with PCI plus optimal medical therapy and patients treated with optimal medical therapy alone.3-10 The addition of PCI does offer quicker relief of angina than medical therapy alone but also carries an increased risk of repeat revascularization, late-stent thrombosis, and a decreased quality of life if performed in patients with minimal symptoms.3,11-19
There also is evidence that PCI reduces ischemia to a greater extent than medical therapy.20,21 Reduction of ischemia has been suggested by some cardiologists as a rationale for performing PCI even in asymptomatic patients.22-24 Given these considerations, guidelines for PCI published jointly by the American College of Cardiology, the American Heart Association, and the Society for Cardiovascular Angiography and Intervention state that for patients with stable angina, any vessels to be dilated must be shown to be “associated with a moderate to severe degree of ischemia on noninvasive testing.”25 Previous studies have shown that patients who receive PCI in accordance with these guidelines have better outcomes.26 In addition, stress testing prior to cardiac catheterization and angioplasty has been associated with lower overall diagnostic costs, shorter hospital stays, and lower rates of revascularization, without adverse effects on cardiac death or myocardial infarction.27,28
With the increasingly widespread use of PCI in patients with stable CAD, it is important that the procedure is being performed in patients for whom there is reasonable expectation of benefit (ie, patients with documented ischemia and/or symptoms). However, the frequency with which ischemia is documented prior to PCI is unknown in the current era of PCI with stenting and this creates uncertainty about whether PCI is being performed in patients in whom benefits outweigh risks. Because professional guidelines call for objective documentation of ischemia prior to elective PCI, we hypothesized that the majority of patients undergoing elective PCI would have had a stress test because this is the most common method of documenting ischemia. Therefore, we undertook a study of Medicare beneficiaries throughout the United States undergoing elective PCI to determine the frequency with which stress testing was performed prior to PCI and to examine the predictors of receipt of stress testing prior to PCI.
The study was approved by the institutional review boards at the University of California, San Francisco, and Maine Medical Center.
To identify patients undergoing elective PCI, data on the use of noninvasive stress testing prior to PCI were obtained for a 20% random sample of Medicare beneficiaries. Selection was based on the last digit of 0 or 5 in a unique patient identifier assigned by Medicare and recorded in the Centers for Medicare & Medicaid Services' Medicare Provider Analysis and Review (MEDPAR), physician and supplier, and denominator files between January 1, 2003, and December 31, 2004. The MEDPAR files (Part A) and physician and supplier files (Part B) contain data on services provided to Medicare fee-for-service beneficiaries, including hospitalizations, physician claims, and outpatient services. The files contain unique identifiers for individual patients, physicians, and hospitals; date and place of service; and diagnoses and procedures as defined by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes and the physicians' Current Procedural Terminology (CPT) codes. The denominator files have enrollment and demographic data for every Medicare beneficiary for the entire year.
To examine possible relationships between use of prior noninvasive testing and characteristics of physicians, the unique physician identifying number from each claim was matched to the American Medical Association's Physician Masterfile. This database contains information on all physicians in the United States, including age, sex, year of graduation, geographic location, specialty, and board certification.
To examine possible relationships between the use of prior noninvasive testing and characteristics of hospitals, the unique hospital identification number from each claim was matched to the American Hospital Association's annual survey database from 2004. The survey contains information on hospital organizational structure, facilities and services, geographic location, accreditation, and teaching status.
Using Part B claims from 2004, the CPT codes for PCI (92980-92982, 92984, 92995-92996) were used to identify patients undergoing PCI. The population was then limited to patients who had a denominator record, were eligible because of age, were eligible for both Part A and Part B, had at least 12 months of eligibility prior to the index PCI, and were enrolled in the Medicare fee-for-service program. The first PCI was selected per patient (index PCI). We then linked MEDPAR and Part B data to create a file containing patients who had a PCI in 2004 and had a complete year of claims available prior to the index PCI.
Patients were excluded if they had a primary diagnosis of acute myocardial infarction (ICD-9-CM codes 410.XX) or unstable angina (ICD-9-CM codes 411.1, 411.81, 411.89). Patients who had PCI, coronary artery bypass graft surgery, valve surgery, acute myocardial infarction, or unstable angina within 1 year prior to the index PCI also were excluded. We further excluded patients if they were admitted from an emergency department, transferred from another institution, or had emergency procedures.
Each included patient was assigned to 1 of 306 hospital referral regions based on referral patterns for tertiary care and the patient's ZIP code of residence, and as defined in the Dartmouth Atlas of Health Care.29
The CPT codes were used for stress echocardiography (93350), exercise treadmill or pharmacological stress test (93015, 93016-93018), and myocardial nuclear imaging (78460-78461, 78464-78466, 78468-78469, 78472-78473, 78478, 78480-78481, 78483, 78491-78492, 78494, 78496) to identify stress tests for each patient. For imaging stress tests, patients must have had both a stress component and an associated imaging component within a 1-day window.
Statistical analyses were performed using SAS version 8 (SAS Institute Inc, Cary, North Carolina), STATA version 9.0 (StataCorp, College Station, Texas), and MlwiN (Center for Multilevel Modeling, Bristol, England) statistical software. We calculated the percentage of all patients who underwent a stress test in the 90 days prior to PCI, stratified by hospital referral regions.
Hospital referral regions with less than 25 PCIs were excluded for the purpose of calculating regional (ie, specific to the hospital referral region) stress testing rates because there were too few procedures to accurately determine those rates. Exclusion of hospital referral regions was done for the analysis of geographic variation only; all patients and hospital referral regions were included for the main analyses.
The relationship between patient, physician, and hospital characteristics and the probability of receiving stress testing prior to PCI was then examined. Patient characteristics examined included age, sex, race, and comorbidities. Race was included as a predictor variable because previous studies have documented racial differences in the receipt of stress testing.30 The ICD-9-CM codes were used to identify diagnoses typically included in the Charlson comorbidity score, including prior myocardial infarction, peripheral vascular disease, chronic obstructive pulmonary disease, renal disease, severe diabetes mellitus, severe liver disease, congestive heart failure, rheumatic disease, cerebrovascular disease, peptic ulcer disease, metastatic cancer, dementia, and AIDS.31 The ICD-9-CM codes also were used to identify diagnoses of angina, chest pain, CAD, hyperlipidemia, hypertension, and prior cardiac catheterization for each patient. The comorbidities and diagnoses to be included in the final model were determined a priori based on their clinical relevance to the likelihood of stress testing.
For physician characteristics, we analyzed physician age, sex, specialty (cardiology, interventional cardiology, internal medicine, or other), and board certification status; years since graduation was omitted due to its colinearity with physician age. The volume of PCIs performed by each physician, multiplied by a factor of 5, also was calculated to estimate total Medicare volume for that physician in 2004, and physicians were categorized into volume quartiles (≤59, 60-94, 95-149, ≥150).
For hospital characteristics, we analyzed variables that may affect rates of stress testing, including hospital association with an outpatient surgery center, having single photon emission computed tomography (CT), electron beam CT, multislice spiral CT, or cardiac catheterization facilities, and having interventional cardiology or adult cardiac surgery services. Hospital teaching status was defined as having been accredited by either the Council of Teaching Hospitals or the Accreditation Council for Graduate Medical Education. Hospital volume of PCI performed in 2004, multiplied by a factor of 5, also was calculated to estimate total Medicare volume for that hospital, and hospitals were categorized into volume quartiles (<285, 285-485, 486-849, ≥850). We also examined whether there was a correlation between PCI volume and likelihood of stress testing within hospital referral regions by calculating a Pearson correlation coefficient.
To determine independent predictors of receiving stress testing, a hierarchical logistic-normal regression model was used with nested random effects to adjust for the effects of hospital referral region, hospital, and physician. The model was estimated using penalized quasi likelihood, as implemented in MlwiN software, and yielded adjusted odds ratios (AORs) with 95% confidence intervals (CIs) relating to the likelihood of stress test receipt. The variance of the random effects was tested for statistical significance, and the influence of the underlying system-level factors was summarized by the median relative difference in the odds of stress testing between 2 randomly selected hospital referral regions, hospitals, or physicians; under the normality assumption, this summary statistic is a simple function of the estimated variance of the random effect.32 The minimum OR detectable with 80% power in a 2-sided test with an α level of .05 was 1.48 for a rare binary covariate of prevalence 1% and reached a minimum of 1.08 for an exposure of prevalence 50%. Interactions between patient age, sex, and race were assessed, as well as interactions between physician volume and hospital volume.
Because our Medicare cohort included only patients aged 65 years or older, we analyzed the rate of stress testing prior to elective PCI in 1630 commercially insured patients who underwent PCI in a 1-year period from October 1, 2005, through September 30, 2006, as a comparison group. The data were drawn from an analytical database fed by the claims and administrative data systems, and included all enrollment information, all claims (from both preferred provider organizations and health maintenance organizations), and all submitted encounters (from health maintenance organizations). We used the same inclusion and exclusion criteria and the same algorithm to determine stress testing rates.
A total of 87 343 patients undergoing PCI in 2004 were identified from 107 162 Medicare line-item claims. Of these, 23 887 met the eligibility criteria and became the study cohort (Figure 1). In the United States, 44.5% (n = 10 629) of patients underwent stress testing within the 90 days prior to elective PCI. Patients undergoing elective PCI were mostly male (60.3%) and mostly white (92.7%). In unadjusted analyses, older patients had lower overall rates of prior stress testing, as did patients with angina, prior CAD, prior cardiac catheterization, chronic obstructive pulmonary disease, and congestive heart failure (Table 1). Patients of physicians who performed a higher volume of PCI had slightly lower rates of stress testing (Table 2). Patients who had their PCI performed at hospitals that were associated with a freestanding outpatient surgery center that owned single photon emission CT or multislice CT scanners had slightly higher rates of stress testing (Table 2).
Figure 2 shows significant geographic variation in the rate of stress testing by hospital referral region, with rates ranging from a low of 22.1% to a high of 70.6% (interquartile range, 39.0%-50.9%). The rate of stress testing did not correlate with the volume of PCI performed in the hospital referral region (Pearson correlation coefficient, −0.043); ie, a hospital referral region with a high volume of PCI did not necessarily have a high percentage of stress testing and vice versa.
For the multivariate model, patients with missing predictor data were excluded so that a total of 20 399 patients were included in a hierarchical logistic model to determine the predictors of undergoing stress testing prior to PCI. Patients who had a cardiac catheterization previously were least likely to undergo stress testing prior to elective PCI (AOR, 0.45; 95% CI, 0.38-0.54). Female sex (AOR, 0.91; 95% CI, 0.86-0.97), age of 85 years or older (AOR, 0.83; 95% CI, 0.72-0.95), and having comorbidities such as rheumatic disease (AOR, 0.80; 95% CI, 0.69-0.94), chronic obstructive pulmonary disease (AOR, 0.83; 95% CI, 0.78-0.89), congestive heart failure (AOR, 0.85; 95% CI, 0.79-0.92), and CAD (AOR, 0.90; 95% CI, 0.81-1.00) also were independent negative predictors of stress testing prior to PCI (Figure 3). Conversely, patient characteristics associated with a higher likelihood of a stress test prior to PCI were being black (AOR, 1.26; 95% CI, 1.09-1.46) and having a diagnosis of chest pain (AOR, 1.28; 95% CI, 1.09-1.54).
Physician age was a significant predictor of the likelihood of receiving a stress test. Patients of physicians aged 50 to 69 years were less likely to receive a stress test prior to PCI than those of physicians younger than 40 years old, while patients of the oldest physicians (age ≥70 years) were more likely to undergo stress testing prior to PCI than those of the younger physicians (age <40 years) (Figure 3). In addition, there was an inverse relationship between physician PCI volume and receipt of stress test. Patients of physicians who performed a low (60-94), medium (95-149), and high (≥150) volume of PCIs were less likely to undergo stress testing within 90 days prior to PCI (AOR, 0.91 [95% CI, 0.84-0.99]; 0.88 [95% CI, 0.81-0.96], and 0.84 [95% CI, 0.77-0.93], respectively) compared with patients of physicians who performed less than 60 PCIs. No hospital factors were independent predictors of stress test receipt, and no interactions having both statistical and clinical significance were found.
The influence of system-level factors (ie, physician, hospital, and hospital referral region) on receiving a stress test prior to PCI was assessed. After controlling for all other factors included in the model, the median relative difference between physicians in odds of stress testing (ie, the likelihood patients from different physicians underwent stress testing prior to PCI) was 40%. Similarly, the median difference in the odds of stress testing between 2 hospital referral regions was 22% and between hospitals was 18%. Thus, although many of system-level factors were not significant predictors by themselves, collectively these factors played roles as large as or larger than patient characteristics in predicting whether stress testing prior to PCI was performed.
Finally, the rate of stress testing prior to elective PCI was examined in a comparison cohort of commercially insured patients and it was found that 34.4% of patients underwent stress testing prior to elective PCI during a 1-year period.
We found that less than half of Medicare patients who underwent elective PCI had a stress test in the 90 days prior to the procedure, and that this proportion varied widely by region and with physician characteristics (even after adjusting for patient factors that may drive use of PCI). Variation in use driven by nonclinical factors begs the question of what proportion of PCI in our sample is actually appropriate. In fact, our data may overestimate the overall rate of stress testing prior to elective PCI. We and Topol et al28 found even lower rates (34.4% and 29%, respectively) in commercially insured populations. Without documentation of ischemia, PCIs may have been performed in patients who would not have benefited from the procedure.
We found significant geographic variation in stress testing, providing another example in which where one lives determines the likelihood and intensity of testing and intervention.33-35 Because stress testing is widely available, it is not clear why such variation should exist, and the likelihood of stress testing does not appear to be related to the volume of PCI performed in a region. Geographic variation in practice patterns has been documented for many procedures, including PCI,29,36 and our data further support that local practice patterns may supersede clinical guidelines and evidence from clinical trials in the decision-making process for patients with stable CAD. Such variation may result in overuse or underuse of PCI in certain populations, leading to inefficient and potentially ineffective care.
We found that patient characteristics affected the rate of stress testing independently of geographic effects. Patients having had prior cardiac catheterization were the least likely to undergo stress testing prior to PCI. Older patients and those with comorbidities also were less likely to undergo stress testing prior to PCI. Although these factors do increase the probability that obstructive CAD is present, evidence of ischemia (not just visualization of anatomy) is crucial in determining if the use of PCI is appropriate.
Women were less likely than men to undergo stress testing prior to PCI, which is consistent with previous studies.30 This may be due in part to the fact that women are more likely to have symptoms that are atypical or are attributed to noncardiac diseases, and also may reflect the lower sensitivity and specificity of stress testing in women, leading physicians to rely less on stress testing to diagnose cardiac disease in women.37-40 On the other hand, we found that patients with a diagnosis of chest pain and blacks were more likely to undergo stress testing prior to PCI. This is in contrast to an earlier study documenting lower population rates of stress testing in blacks compared with whites.30 However, regardless of the direction of the association, these findings support the view that sex and race influence the decision to refer for stress testing and PCI. Referral for testing on the basis of perceived risk, rather than objective evidence of ischemia, also has been demonstrated in referral for cardiac catheterization in patients with non–ST-segment elevation myocardial infarction.41
We found that physician characteristics such as younger age and higher volume of PCI performed predicted lower rates of stress testing prior to PCI. There also was substantial unexplained variation among physicians. This suggests that physician decision making regarding PCI was influenced less by presence of ischemia, as PCI guidelines suggest, and more by physicians' own biases and community practice patterns.22,42,43 While many of the physicians in our sample may have had the patient referred to them specifically for a PCI and may not have done the initial evaluation of the patient, our data would have captured a stress test ordered by any physician. In addition, the variation in the rate of stress testing by hospital referral region did not correlate with the hospital referral region volume of PCI, further suggesting that there may not have been systematic adherence to guidelines. Physician adherence to guidelines has been documented to be suboptimal in many situations,44,45 even though adherence to guidelines has been demonstrated to improve quality of care, particularly in patients with CAD.46-48 Thus, the geographic variation and low overall rates of stress testing could be partially ameliorated by promoting better adherence to clinical practice guidelines.
We found that the influence of system-level factors (ie, physician, hospital, and geographic characteristics) on stress testing is as strong as or stronger than the most important patient-level effects assessed, even if the specific characteristics measured at the physician or hospital level were not significant. In fact, the effect of choice of physician (median relative OR, 1.40) was greater than any single patient characteristic, underscoring the magnitude of physician influence on the rate of stress testing prior to PCI. The fact that all system-level effects were significant indicates that the rate of stress testing may be affected by factors such as the availability of catheterization laboratories or the density of specialists in a region.49 The specific system-level characteristics that are most influential need to be better defined to understand how to design health system–level interventions to address the low rate of stress testing in Medicare patients.
Our study has several limitations. Our analysis is based on administrative data in Medicare patients, and lacks clinical data not documented by the clinicians or coded in the billing process (eg, <5% of patients had angina or chest pain coded). Also, as with all retrospective, observational studies using administrative databases, our results are limited by uncertainties in patient selection and unmeasured confounding variables that may affect a patient's likelihood of undergoing stress testing. Although we would not expect that 100% of elective PCIs would be preceded by stress testing, there were regions with stress testing rates of 70%, suggesting that the overall stress testing prior to PCI rate of 44.5% may be too low. Because our cohort only included Medicare patients, our results may not necessarily be generalizable to younger patients or those with a different insurance status.
We chose a 90-day time window to assess the rate of stress testing prior to PCI, which may miss patients who undergo PCI only after optimal medical therapy alone fails to improve symptoms. However, extending our surveillance prior to PCI to 180 days did not substantially increase the rate of stress testing prior to PCI (49.1% at 180 days). With previous history limited to 1 year prior to the procedure, we may not have fully ascertained the number of patients with prior PCI, coronary artery bypass graft surgery, or diagnosis of CAD. In addition, we could not assess the use of newer noninvasive imaging modalities such as electron beam CT, multislice CT angiography, or cardiac magnetic resonance angiography because Medicare did not cover those tests in 2004 for the diagnosis of cardiac disease. These tests visualize atherosclerosis and thus may lead to cardiac catheterization and subsequent PCI based on anatomical data, without documentation of ischemia by stress testing. Further study needs to be done to establish what role these tests may play in the use of PCI.
Percutaneous coronary intervention is an important treatment option for stable CAD, and its use has increased by more than 300% over the past decade.1 Guidelines for PCI call for documenting ischemia prior to PCI in the vast majority of patients with stable CAD; however, our data suggest that this is not being done consistently. Assessing whether PCI is being performed in appropriately selected patients is crucial to providing high-quality, patient-centered medical care in light of evidence that patients in regions providing high-intensity care do not have better (and sometimes have worse) outcomes than those in regions providing low-intensity care.50
In addition, because Medicare spends $10 000 to $15 000 per PCI and PCI has accounted for at least 10% of the increase in Medicare spending since the mid-1990s, it is important to document that patients are receiving PCI for appropriate indications to ensure the optimal use of Medicare resources. Our findings highlight an opportunity for improvement in the care of patients with stable CAD and suggest that current proposals to restructure Medicare payment to reward hospitals and physicians who adhere to guidelines would improve the safety and delivery of health care to Medicare beneficiaries while decreasing Medicare expenditures on costly and inappropriate procedures.
Corresponding Author: Rita F. Redberg, MD, MSc, 505 Parnassus Ave, Ste M-1180, School of Medicine, Division of Cardiology, San Francisco, CA 94143 (email@example.com).
Author Contributions: Dr Lin 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: Lin, Dudley, Lucas, Malenka, Redberg.
Acquisition of data: Lin, Dudley, Lucas.
Analysis and interpretation of data: Lin, Dudley, Lucas, Malenka, Vittinghoff, Redberg.
Drafting of the manuscript: Lin, Dudley, Malenka, Vittinghoff.
Critical revision of the manuscript for important intellectual content: Lin, Dudley, Lucas, Malenka, Redberg.
Statistical analysis: Lin, Dudley, Lucas, Malenka, Vittinghoff.
Obtained funding: Dudley, Redberg.
Administrative, technical, or material support: Dudley, Lucas, Malenka.
Study supervision: Dudley, Lucas, Malenka, Redberg.
Financial Disclosures: None reported.
Funding/Support: This study was funded by the Blue Shield of California Foundation. A portion of Dr Dudley's work was supported by an Investigator Award in Health Policy from the Robert Wood Johnson Foundation.
Role of the Sponsor: The Blue Shield of California Foundation and the Robert Wood Johnson Foundation had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.
Create a personal account or sign in to: