 |
|
Variation in Routine Electrocardiogram Use in Academic Primary Care Practice
Randall S. Stafford, MD, PhD;
Bismruta Misra, MPH
Arch Intern Med. 2001;161:2351-2355.
ABSTRACT
| |
Background Lack of practical consensus regarding routine electrocardiogram (ECG)
ordering in primary care led us to hypothesize that nonclinical variations
in ordering would exist among primary care providers.
Methods We used 2 computerized billing systems to measure ECG ordering at visits
to providers in 10 internal medicine group practices affiliated with a large,
urban teaching hospital from October 1, 1996, to September 30, 1997. To focus
on screening or routine ECGs, patients with known cardiac disease or suggestive
symptoms were excluded, as were providers with fewer than 200 annual patient
visits. Included were 69 921 patients making 190 238 visits to 125
primary care providers. Adjusted rates of ECG ordering accounted for patient
age, sex, and 5 key diagnoses. Logistic regression evaluated additional predictors
of ECG ordering.
Results Electrocardiograms were ordered in 4.4% of visits to patients without
reported cardiac disease. Among the 10 group practices, ECG ordering varied
from 0.5% to 9.6% of visits (adjusted rates, 0.8%-8.6%). Variations between
individual providers were even more dramatic: adjusted rates ranged from 0.0%
to 24% of visits, with an interquartile range of 1.4% to 4.7% and a coefficient
of variation of 88%. Significant predictors of ECG use were older patient
age, male sex, and the presence of clinical comorbidities. Additional nonclinical
predictors included Medicare as a payment source, older male providers, and
providers who billed for ECG interpretation.
Conclusions Variations in ECG ordering are not explained by patient characteristics.
The tremendous nonclinical variations in ECG test ordering suggest a need
for greater consensus about use of screening ECGs in primary care.
INTRODUCTION
NONINVASIVE diagnostic testing for cardiac disease has great potential
to evaluate conditions that are prevalent and for which specific therapy alters
outcomes. Electrocardiograms (ECGs) are the oldest, most widely available,
and most frequently used cardiac test.1 Annually,
ECGs are ordered in 20 million US physician office visits (2.6% of all visits)
and, as a diagnostic test, are exceeded in frequency only by urinalysis, complete
blood cell counts, cholesterol tests, and Papanicolaou tests.2
Despite the usefulness of ECGs in evaluating sentinel symptoms of cardiac
disease, such as chest pain and palpitations, their role in routine screening
for asymptomatic cardiac disease remains controversial. There is no practical
consensus on the use of ECGs in primary care, particularly for patients without
known or suspected cardiac disease. Current guidelines, however, suggest a
limited role for routine ECG testing. Recommendations against screening ECGs
have been issued by the US Preventive Services Task Force3
and the Canadian Task Force on the Periodic Health Examination.4
Guidelines of the American College of Cardiology/American Heart Association5 and the American College of Physicians 6
suggest a potential but limited role for baseline testing and for testing
in the elderly. When specific documented cardiac risk factors or cardiac disease
are present, current guidelines offer greater support for the use of routine
ECGs although, even in these situations, the benefit of screening has not
been rigorously established.3 Often cited in
recommendations against routine use of ECGs are their low sensitivity, only
modest specificity, and contribution to health care costs. The cost of ECG
testing includes not only its direct costs but also the costs attributable
to subsequent diagnostic and therapeutic activities resulting from false-positive
or equivocal findings.
Despite the cautions raised by guidelines, physicians may nonetheless
perceive that routine screening ECGs are valuable clinically across a broad
range of patients, including those at low risk of cardiac disease. Arguments
given in favor of screening ECGs include the value of a baseline ECG before
the potential occurrence of cardiac symptoms,7
the ability of ECG testing to fulfill patients' expectations regarding evaluation,8-9 and a belief that more complete diagnostic
testing necessarily improves clinical decision making.10
It also may be that expectations of peers or patients may lead physicians
to order some ECGs. Although each of these arguments has substantial limitations,7, 11-12 they nonetheless appear
to influence physician test-ordering behavior.9
Past analyses of ECG use in primary care are limited. National assessments
of office-based physician practices indicate that ECG use has not changed
between 1979 (2.7% of all US office visits)13
and 1993 (2.6%).2 As expected, ECG use increases
with increasing patient age.14 Greater ECG
ordering also has been correlated with physician characteristics, including
large group practice (compared with small group or solo practice),15 fee-for-service practice (compared with prepaid practice),16 more recent physician graduation, and physician specialty
training.17 Several studies suggest that a
broad range of physician activities, including the use of diagnostic tests,
vary considerably among physicians.9, 18-19
The magnitude of variation tends to be greatest for those activities where
there is the most uncertainty.9
To evaluate and describe current practices regarding routine ECG use
in primary care, we examined the practices of 125 primary care providers (PCPs)
in 10 internal medicine group practices at an urban academic medical center.
Given the lack of definitive guidelines regarding ECG use and the competing
pressures faced by PCPs, we hypothesized that sizable variations in ECG ordering
rates would exist and that nonclinical factors would strongly predict ECG
use.
METHODS
DATA SOURCES
We obtained data on ECG ordering from October 1, 1996, through September
30, 1997, for 10 primary care internal medicine group practices associated
with the Massachusetts General Hospital, Boston. These practices included
4 private practices associated with the Massachusetts General Physicians Organization,
3 hospital-based practices, and 3 affiliated health center practices. These
practices varied in size from 6 to 20 PCPs, nearly all of whom were internists.
Two electronic billing systems, IDX Systems Corp (Burlington, Vt) (for the
4 private practices) and Transition Systems, Inc (Boston) (for the other 6
practices), were used to gather information on patient visits and ECG ordering.
In both billing systems, ECG ordering during PCP visits was identified by
the presence of at least 1 of the following Current Procedural
Terminology20 codes: 93000, 93005, 93010,
93012, 93014, 93224-93237, 93268, 93270, 93271, or 93272.
In addition, these 2 billing systems provided information on patient
age, sex, payment source, site of care, and principal and secondary diagnoses.
We also linked information about individual patient visits to provider characteristics
that were available from the hospital's physicians organization. Provider
characteristics included provider type, specialty, years since medical school
graduation, and sex. From our billing information, we also determined each
provider's annual visit volume and whether providers interpreted their own
ECGs or billed for this service. Institutional review board approval was obtained
for our data collection and analysis protocol.
Our database initially included information on 220 305 visits by
74 738 patients to more than 300 PCPs. To ensure sufficient sample sizes
at the provider level, we excluded providers with fewer than 200 annual patient
visits. The excluded providers were mostly internal medicine residents or
part-time faculty who had limited clinical practices (total visits, 9882;
4% of total) and were not representative. Most of the 125 included providers
were faculty-level internists (86%), with the remainder being nurse practitioners
(8%), nurses (3%), and others (3%). Providers were nearly equally distributed
between private practices (n = 40), hospital-based practices (n = 42), and
affiliated community-based health centers (n = 43).
We focused our analysis on ECGs ordered for screening or routine purposes
rather than for diagnosis or clinical monitoring. We excluded an additional
22 205 patient visits (11% of visits to the included PCPs) with a principal
or secondary International Classification of Diseases, Ninth
Revision (ICD-9)21
diagnosis of cardiac disease (codes 391-429, except for 401, 403, and 405
[hypertension and hypertensive renal disease]) or with reported symptoms of
chest pain (codes 786.50-786.59) or palpitations (codes 785.0-785.3). Our
final database was composed of 125 providers, 69 921 patients, and 190 238
patient visits.
STATISTICAL METHODS
Our analysis had 2 goals: (1) to quantify variation in ECG testing among
primary care group practices and providers and (2) to identify patient, provider,
and group practice characteristics associated with ECG ordering. Our principal
outcome measure was the likelihood of an ECG being ordered at a given patient
visit.
We calculated unadjusted rates of ECG ordering for group practices or
individual providers as the ratio of ECGs ordered to the total number of visits.
We also calculated adjusted rates of ECG ordering using indirect standardization22 to account for the differing patient characteristics
between providers. These adjusted rates of ECG use accounted for systematic
variation in ECG use that we observed by age, sex, and clinical diagnosis.
Following exploratory analysis of ECG rates by patient age, we defined age
groups of 30 years or younger, 31 to 50, 51 to 65, and older than 65 years.
Exploratory analysis of ECG rates by principal diagnosis identified 5 clinical
conditions statistically associated with increased ECG use: hyperlipidemia
(ICD-9 codes 272.0-272.7); hypertension (ICD-9 codes 401.0-401.9); dizziness, syncope, and giddiness (ICD-9 codes 780.2, 780.4, 781.2, and 781.3); malaise and
fatigue (ICD-9 code 780.7); and general medical examination
(ICD-9 codes V70.0-V76.9).21
A sixth diagnostic category included all other visits.
Using data from all providers, we developed standard rates of ECG ordering
for all 48 combinations of age (4 categories), sex (2 categories), and diagnosis
(6 categories). These standard rates were applied to the distribution of these
characteristics for each provider's visits to derive an expected number of
ECG. The calculation of expected ECGs assumed that for each patient category
the provider's practices would follow the age-, sex-, and diagnosis-specific
standard rates derived for all providers. Adjusted rates of ECG ordering were
then calculated as the ratio of actual to expected ECGs multiplied by the
mean rate of ECG ordering for all providers. This adjustment method allowed
us to compare the practices of different providers while accounting for the
demographic and clinical characteristics of their patients.
We described variations in ECG ordering behavior between providers using
the coefficient of variation and interquartile range. Using provider-specific
ECG ordering rates, we calculated the coefficient of variation as the SD divided
by the overall mean ordering rate, both weighted by patient visits. The interquartile
range represented the ordering rate of the 25th percentile provider compared
with the 75th percentile provider.
To further explore potential explanations for variations in ECG ordering,
a multiple logistic regression model was developed to predict the likelihood
of ECG ordering at specific visits. In addition to the variables of patient
age, sex, and diagnosis used to calculate adjusted rates of ECG ordering,
this model included provider sex, provider years postgraduation, annual provider
visit volume, expected source of payment for the visit, type of primary care
practice, and whether the PCP billed separately for ECG interpretation. After
20 visits with incomplete information were excluded, the sample for this analysis
included 190 218 office visits. Initial testing of this model indicated
significant statistical interaction between 2 pairs of predictor variables.
For this reason, we defined combinations of patient sex and age, as well as
provider sex and years postgraduation, to capture these interactive effects.
We interpreted the effect of predictor variables on ECG ordering by calculating
adjusted odds ratios and their 95% confidence intervals.
Our methods using patient visit as the unit of analysis do not explicitly
account for the clustered nature of our data, where multiple visits to hospital-affiliated
physicians were made by individual patients. However, we also analyzed patterns
of ECG use with patients as the unit of analysis, with the probability of
1 or more ECGs in a year as our outcome variable. This analysis indicated
similar patterns and variations in ECG use compared with the findings presented
here. We opted to use visits as the unit of analysis because diagnostic data
and provider assignment were specific to visits.
RESULTS
For the 190 238 visits by patients without reported cardiac disease
included in our study, ECGs were ordered in 8357 or 4.4% of visits. There
was tremendous variation in ECG ordering at the level of both primary care
group practices and PCPs that was not altered substantially by adjustment
for patient characteristics.
The ECG ordering rates for the 10 group practices varied from 0.5% to
9.6% of visits by patients without cardiac disease. Only modest differences
were observed in the expected rates of ECG ordering based on applying global
age-, sex-, and diagnosis-specific rates to each practice's visit characteristics
(Figure 1). After adjustment for
expected ECG ordering, the adjusted rates of the group practices varied from
0.8% to 8.6% of visits. Group practices at hospital-affiliated health centers
had an adjusted ECG use rate of 2.0% of visits (range, 1.5%-2.6%), considerably
lower than all but 1 of the other group practices. Hospital-based group practices
(mean, 6.5%; range, 4.2%-9.6%) and 3 of 4 private practices affiliated with
the hospital's physician organization (mean, 6.7%; range, 4.2%-9.6%) had higher
rates. The 1 exception to this pattern was a private women's health practice
(mean, 0.5%).
|
|
|
|
Screening electrocardiogram (ECG) use by primary care practice, Massachusetts
General Hospital, October 1, 1996, to September 30, 1997. The actual ECG rate
is the proportion of applicable visits where a screening ECG was ordered.
The expected ECG rate is the rate expected if overall age-, sex-, and diagnosis-specific
rates were applied to the characteristics of each group's patient visits.
|
|
|
Differences between the 125 individual providers were of even greater
magnitude, with ECG ordering rates varying from 0.0% to 32% of visits. The
coefficient of variation at the provider level was 111% (SD of physician-specific
rates [4.9%] divided by overall mean ECG rate [4.4%]). Adjusting for sex,
age, and diagnosis narrowed this gap somewhat, although provider-specific
adjusted rates still varied between 0.0% and 24% with a coefficient of variation
of 88%. The median rate of screening ECG test ordering for all PCPs was 2.5%
with an interquartile range from 1.3% (25th percentile) to 5.4% (75th percentile).
After adjustment, the interquartile range for the likelihood of ECG testing
during a visit was 1.4% to 4.7%. The 63 providers whose testing rates were
above this median accounted for 7301 (87.4%) of 8357 ECGs ordered.
We evaluated whether other factors beyond patient demographic and clinical
factors explained the observed variations in ECG use. We used a logistic regression
model that included additional predictors of ECG use, including provider characteristics. Table 1 presents the odds ratio from a
logistic regression model developed to predict the likelihood of ECG use at
specific visits. This model confirmed that older age, male sex, and specific
clinical conditions were independently associated with an increased likelihood
of ECG ordering. As expected, older patients were more likely to receive ECGs.
Male patients had greater ECG use, but only at younger ages. For patients
older than 65 years, adjusted male vs female differences were minimal. All
5 of the selected comorbidities independently increased the odds of ECG ordering.
|
|
|
|
Independent Predictors of Electrocardiogram (ECG) Use at Primary Care
Visits Based on Multiple Logistic Regression, Massachusetts General Hospital,
October 1996 Through September 1997
|
|
|
In the logistic regression model, a range of nonclinical factors had
an impact on ECG ordering. Payment source affected ECG ordering so that patients
with Medicare as their primary source of payment were the most likely to receive
ECGs, whereas patients with Medicaid coverage were the least likely. Providers
with either high or low numbers of patient visits ordered ECGs less often
than providers in the 2 middle quartiles of annual visits. Male providers
were more likely to order ECGs than female providers. This difference was
more extreme for providers who were more than 15 years postgraduation (male
vs female odds ratio, 2.27) compared with younger providers (odds ratio, 1.39).
The strongest predictor of ECG ordering was type of practice, with providers
in the hospital-affiliated health centers ordering ECGs only a third as often
as providers in hospital-based or private practices.
While not apparent in unadjusted comparisons, screening ECG ordering
at primary care visits was 16% more likely for providers who billed for their
ECG interpretations. Health center providers (with lower use rates) tend to
interpret their own ECGs. Therefore, in the absence of an effect of billing
status, on an unadjusted basis, one would have expected ECG use to be much
lower in providers who billed for ECG interpretation.
Despite this broad range of clinical and nonclinical predictors, this
logistic regression model explained the equivalent of only 5% of the total
variation in ECG use at the level of individual visits.
COMMENT
For 10 primary care internal medicine practices affiliated with a large
urban academic medical center, we found tremendous variations in the ordering
of screening ECGs. These variations were evident both at the level of group
practices and individual providers and were not explained by patient sex,
age, and principal diagnosis.
A logistic regression model developed to evaluate other potential predictors
revealed additional factors associated with variations in ECG use. A range
of clinical and nonclinical factors were found to be associated with ECG use.
However, most of the statistical variation in ECG use was not explained by
these predictors. Contrary to past studies,17
younger providers were not consistently more likely to order ECGs. Female
providers were far less likely to order ECGs, particularly if they had been
in practice more than 15 years.
For several of these nonclinical predictors, the results suggest that
financial incentives may directly or indirectly influence ECG ordering, a
finding consistent with past work.16 Patients
with private insurance or Medicare were more likely to have ECGs ordered compared
with patients without health insurance or with Medicaid coverage. The hospital's
health centers, where ordering rates were lowest, have traditionally had clinical
missions emphasizing services to their communities rather than financial performance.
Finally, providers who bill for interpretation of the ECGs had a slightly
greater likelihood of ECG ordering than those who did not.
The large magnitude of unexplained variations between providers may
result from the lack of a clear consensus on the use of screening ECGs and
the potentially conflicting demands faced by providers in applying cardiac
diagnostic technology. While evidence-based clinical guidelines generally
discourage the use of screening ECGs, this diagnostic test is a traditional
feature of office-based internal medicine practice. To the extent that the
general public perceives ECGs to be integral to comprehensive ambulatory care,
physicians may be compelled to negotiate specific demands from patients regarding
ECG ordering.
Although the cost associated with a single ECG is relatively modest,
the aggregate cost of ECG testing is substantial given the frequency of ECG
ordering among primary care visits. Using an average Medicare-allowed charge
of $30 for ECG performance and interpretation (Current Procedural
Terminology code 93000),23 we conservatively
estimate that $250 000 is expended in our setting annually for screening
ECGs by PCPs. By extension, the annual national cost of all ECG ordering likely
exceeds $600 million. In addition, the cost implications of ECG testing extend
beyond the test itself. These costs include those associated with follow-up
testing of suspicious or false-positive findings. Finally, false-positive
findings may serve as an entry point into a diagnostic and therapeutic pathway
eventually leading to costly revascularization.
Several limitations of our analysis should be kept in mind. We have
used billing information on ECG ordering and patient diagnoses that has not
been validated. Because of our data source, we had a limited range of variables
available for use in our multivariate statistical model. Other information
on providers and patients might have increased our ability to explain the
variations we noted. Our results apply to academic internal medicine group
practices and may not be generalizable to other settings. However, the observed
4.4% ECG use rate is not substantially different from the 2.6% rate reported
for all visits to all US office-based physicians.2
Finally, while existing guidelines suggest that ECGs may be overused, this
study has not attempted to evaluate whether the ordering of specific ECGs
was appropriate clinically.
The tremendous variation in ECG ordering that we observed suggests that
decision making about this diagnostic test is subject to considerable discretion
and uncertainty. Although variation in clinical practice is not necessarily
inappropriate, extreme variations suggest that providers are making decisions
without a consistent relationship to patient outcomes. The existence of variations
despite widely disseminated clinical guidelines suggests that these guidelines
have not been particularly influential in guiding clinical practice. More
aggressive efforts to fully implement these guidelines into clinical care
may be appropriate. Beyond clinical uncertainty about ECG use, an additional
contributor to variations may be the lack of feedback available to providers.
The ability of providers to compare their practices with their peers also
could help PCPs make appropriate decisions about screening ECG use.
AUTHOR INFORMATION
Accepted for publication February 22, 2001.
This project was supported by the Primary Care Operations Improvement
Initiative at Massachusetts General Hospital, Boston.
Presented in part at the 22nd Annual Meeting of the Society for General
Internal Medicine, San Francisco, Calif, April 29, 1999.
Special thanks to Michael J. Barry, MD, John H. Farhat, MPH, Sue Clifford,
MBA, and Robert Murray, MPH, for their editorial and data processing assistance
on this article.
Corresponding author: Randall S. Stafford, MD, PhD, Stanford Center
for Research in Disease Prevention, 1000 Welch Rd, Palo Alto, CA 94304 (e-mail: rstafford{at}stanford.edu).
From the Primary Care Operations Improvement Team and Institute for
Health Policy, Massachusetts General Hospital, Boston (Dr Stafford and Ms
Misra); and Howard University College of Medicine, Washington, DC (Ms Misra).
Dr Stafford is now with the Stanford Center for Research in Disease Prevention,
Palo Alto, Calif.
REFERENCES
| |
1. Fye W. A history of the origin, evolution, and impact of electrocardiography. Am J Cardiol. 1994;73:937-949.
FULL TEXT
|
ISI
| PUBMED
2. Schappert S. National Ambulatory Medical Care Survey: 1992 Summary. Hyattsville, Md: National Center for Health Statistics; 1994. Advance
Data From Vital and Health Statistics, No. 253.
3. US Preventive Services Task Force. Guide to Clinical Preventive Services. 2nd ed. Baltimore, Md: Williams & Wilkins; 1996.
4. Canadian Task Force on the Periodic Health Examination. The periodic health examination: 1984 update. CMAJ. 1984;130:2-15.
5. American College of Cardiology/American Heart Association. Guidelines for electrocardiography: a report of the American College
of Cardiology/American Heart Association Task Force of the Assessment of Diagnostic
and Therapeutic Cardiovascular Procedures (Committee on Electrocardiography). J Am Coll Cardiol. 1992;19:473-481.
ISI
| PUBMED
6. Eddy D, ed. Common Screening Tests. Philadelphia, Pa: American College of Physicians; 1991.
7. Lee TH, Cook EF, Weisberg MC, Rouan GW, Brand DA, Goldman L. Impact of the availability of a prior electrocardiogram on the triage
of the patient with acute chest pain. J Gen Intern Med. 1990;5:381-388.
ISI
| PUBMED
8. Eisenberg J. Physician utilization: the state of research about physicians' practice
patterns. Med Care. 1985;23:461-483.
FULL TEXT
|
ISI
| PUBMED
9. Eisenberg J. Doctors' Decisions and the Cost of Medical Care. Ann Arbor, Mich: Health Administration Press; 1986.
10. Axt-Adam P, van der Wouden J, van der Does E. Influencing behavior of physicians ordering laboratory tests: a literature
study. Med Care. 1993;31:784-794.
ISI
| PUBMED
11. Norell M, Lythall D, Coghlan G, et al. Limited value of the resting electrocardiogram in assessing patients
with recent onset chest pain: lessons from a chest pain clinic. Br Heart J. 1992;67:53-56.
FREE FULL TEXT
12. Edlavitch SA, Crow R, Burke GL, Huber J, Prineas R, Blackburn H. The effect of the number of electrocardiograms analyzed on cardiovascular
disease surveillance: the Minnesota Heart Survey (MHS). J Clin Epidemiol. 1990;43:93-99.
FULL TEXT
|
ISI
| PUBMED
13. National Center for Health Statistics. Public Use Data Tape Documentation: National Ambulatory
Medical Care Survey, 1979. Hyattsville, Md: National Center for Health Statistics; 1979.
14. Health Services Utilization and Research Commission. Anatomy of a practice guideline: tradition, science, and consensus
on using electrocardiograms in Saskatchewan. Can Fam Physician. 1995;41:37-40, 43-49.
ISI
| PUBMED
15. Epstein AM, Begg CB, McNeil BJ. The effects of group size on test ordering for hypertensive patients. N Engl J Med. 1983;309:464-468.
ABSTRACT
16. Epstein AM, Begg CB, McNeil BJ. The use of ambulatory testing in prepaid and fee-for-service group
practices: relation to perceived profitability. N Engl J Med. 1986;314:1089-1094.
ABSTRACT
17. Epstein AM, McNeil BJ. Physician characteristics and organizational factors influencing use
of ambulatory tests. Med Decis Making. 1985;5:401-415.
18. Ashton CM, Petersen NJ, Souchek J, et al. Geographic variations in utilization rates in Veterans Affairs hospitals
and clinics. N Engl J Med. 1999;340:32-39.
FREE FULL TEXT
19. Komaromy M, Lurie N, Osmond D, Vranizan K, Keane D, Bindman A. Physician practice style and rates of hospitalization for chronic medical
conditions. Med Care. 1996;34:594-609.
FULL TEXT
|
ISI
| PUBMED
20. Physicians' Current Procedural Terminology (CPT 98). 4th ed. Chicago, Ill: American Medical Association; 1997.
21. International Classification of Disease, Ninth Revision,
Clinical Modification. Washington, DC: Public Health Service, US Dept of Health and Human
Services; 1988.
22. Fleiss JL. Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Wiley & Sons Inc; 1981.
23. Health Care Financing Administration. Physician Fee Schedule Payment Amount File, National. Baltimore, Md: Health Care Financing Administration; 2000. Available
at: http://www.hcfa.gov/stats/pfalldwn.htm. Accessed July 27, 2001.
 CiteULike  Connotea  Del.icio.us  Digg  Reddit  Technorati
What's this?
RELATED ARTICLE
Archives of Internal Medicine Reader's Choice: Continuing Medical Education
Arch Intern Med. 2001;161(19):2388-2389.
FULL TEXT
THIS ARTICLE HAS BEEN CITED BY OTHER ARTICLES
The Impact of Peer Management on Test-Ordering Behavior
Neilson et al.
ANN INTERN MED 2004;141:196-204.
ABSTRACT
| FULL TEXT
|
