The figure shows unweighted frequencies and weighted percentages (for
physicians, Community Tracking Study [CTS] survey weights were applied; for
patients, the CTS survey weight for their usual-source-of-care physician and
a factor of 20 were applied. If a patient saw more than 1 CTS physician, the
weighted population is based on a randomly selected CTS physician weight).
The usual-source-of-care physician was defined as the physician who filed
claims for the plurality of evaluation and management visits for each patient.
Eligibility criteria for each preventive service are detailed in Table 1. HbA1c indicates glycosylated
hemoglobin; UPIN, Unique Physician Identification Number.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Pham HH, Schrag D, Hargraves JL, Bach PB. Delivery of Preventive Services to Older Adults by Primary Care Physicians. JAMA. 2005;294(4):473–481. doi:10.1001/jama.294.4.473
Context Rates of preventive services remain below national goals.
Objective To identify characteristics of physicians and their practices that are
associated with the quality of preventive care their patients receive.
Design Cross-sectional analysis of data on US physician respondents to the
2000-2001 Community Tracking Study Physician Survey linked to claims data
on Medicare beneficiaries they treated in 2001. Physician variables included
training and qualifications and sex. Practice setting variables included practice
type, size, sources of revenue, and access to information technology. Analyses
were adjusted for patient demographics and comorbidity, as well as community
Setting and Participants Primary care delivered by 3660 physicians providing usual care to 24 581
Medicare beneficiaries aged 65 years and older.
Main Outcome Measures Proportion of eligible beneficiaries receiving each of 6 preventive
services: diabetic monitoring with hemoglobin A1c measurement or
eye examinations, screening for colon or breast cancer, and vaccination for
influenza or pneumococcus in 2001.
Results Overall, the proportion of beneficiaries receiving services was below
national goals. Physician and, more consistently, practice-level characteristics
were both associated with differences in the delivery of services. The strongest
associations were with practice type and the percentage of practice revenue
derived from Medicaid. For instance, beneficiaries receiving usual care in
practices with less than 6% of revenue from Medicaid were more likely than
those with more than 15% of revenue derived from Medicaid to receive diabetic
eye examinations (48.9% vs 43%; P = .02),
hemoglobin A1c monitoring (61.2% vs 48.4%; P<.001), mammograms (52.1% vs 38.9%; P<.001),
colon cancer screening (10.0% vs 8.5%; P = .60),
and influenza (50.2% vs 39.2%; P<.001) and pneumococcal
(8.2% vs 6.4%; P<.001) vaccinations. Other variables
associated with delivery of preventive services after adjustment for patient
and geographic factors included obtaining usual health care from a physician
who worked in group practices of 3 or more, who was a graduate of a US or
Canadian medical school, or who reported availability of information technology
to generate preventive care reminders or access treatment guidelines.
Conclusions Delivery of routine preventive services is suboptimal for Medicare beneficiaries.
However, patients treated within particular practice settings and by particular
subgroups of physicians are at particular risk of low-quality care. Profiling
these practices may help develop tailored interventions that can be directed
to sites where the opportunities for quality improvement are greatest.
It is well established that many US patients receive suboptimal care
and that subgroups disadvantaged on the basis of demographic or socioeconomic
characteristics are at special risk.1,2 Even
among advantaged patients, the quality of US health care has been shown to
lag well below national goals.3,4 In
the landmark Community Quality Index study, McGlynn et al3 and
Kerr et al5 documented not only that quality
of care is suboptimal, but that quality problems are not limited to a specific
set of conditions or communities.
An emerging body of literature now suggests that quality of care may
vary in association with the characteristics of individual physicians and
their practices.4,6,7 Lurie
et al8 reported differences in cervical and
breast cancer screening by physician sex, while O’Malley and Mandelblatt9 found that patients at community health centers were
as likely as those in private physicians’ offices to receive preventive
services, but these associations have not been examined for a nationally representative
group of physicians.8-12
We studied the relationship between attributes of physicians and their
practices, such as experience, training, sex, and practice setting, and the
extent to which their Medicare patients received preventive services. We hypothesized
that patients treated by less well-trained physicians, or in less well-equipped
health care settings, would be less likely to receive preventive care services.
Physician Data. The Community Tracking Study
(CTS) Physician Survey is a biannual, nationally representative telephone
survey of nonfederal US physicians conducted in 60 randomly selected metropolitan
statistical areas and supplemented by a national sample. Primary care physicians
are oversampled. In Round 3 (2000-2001), the response rate was 59%. (Details
of the survey have previously been published and are available at http://www.hschange.org/index.cgi?data=04.) The survey included physicians who reported at least 20 hours per
week of direct patient care in an office- or hospital-based practice, including
Bureau of Primary Health Care sites. Residents and fellows and certain specialties
such as pathology or anesthesiology were excluded. Physicians received a letter
beforehand describing the purpose of the survey, and interviewers accepted
physicians’ willingness to complete interviews as implicit consent.
Patient Visit Data. The Medicare program provides
insurance for 97% of individuals aged 65 years and older in the United States.
In 2001, the program covered 40 million persons, 86% of whom were enrolled
under Part A and B indemnity insurance, for which physicians submit detailed
claims for rendered services to the Centers for Medicare & Medicaid Services
for reimbursement.13 Our data were obtained
from the 2001 5% Carrier File, which contains complete claims histories for
physicians’ professional services on a 5% representative sample of Medicare
beneficiaries who had both Part A and Part B coverage. We limited our analysis
to beneficiaries aged 65 years and older as of January 2001.
Data on physicians and patient visits were linked through the “performing
physician” Unique Physician Identification Number (UPIN), which is recorded
on all claims submitted to the Medicare program.14 The
usual-source-of-care physician was defined as the physician who provided the
greatest number of evaluation and management services to a particular beneficiary
in 2001 (based on the Berenson-Eggers type of service codes as previously
reported).7 In the case of a tie between physicians,
we defined the usual-source-of-care physician as the physician with the highest
total amount of paid claims for that beneficiary. Our analysis then focused
on those beneficiaries for whom the identified physician was categorized as
a traditional primary care physician—a general internist, general practitioner,
or family practitioner—and who responded to Round 3 of the CTS. This
approach was described and validated by Weiner et al.12 To
ensure that our findings were not a byproduct of the particular selection
procedure we used, we tested the effect on our findings of (1) including the
percentage of a beneficiary’s evaluation and management visits that
were with the usual-source-of-care physician as a covariate; (2) limiting
the population to beneficiaries who had at least 5 evaluation and management
visits with their usual-source-of-care physician; (3) defining usual-source-of-care
physicians as physicians providing the majority (≥50%) of a patient’s
visits for evaluation and management; and (4) assigning a CTS usual source
of care to any patient who saw at least 1 CTS primary care physician in 2001,
whether or not they provided the plurality of the patient’s care (the
fewer than 10% of beneficiaries who had visits with >1 CTS primary care physician
were randomly assigned to 1 physician). Analyses using these alternative methods
of assignment yielded substantively similar results.
We evaluated beneficiaries’ Medicare claims to determine delivery
of diabetic monitoring (hemoglobin A1c testing, eye examinations),
cancer screening (colonoscopy/sigmoidoscopy, mammography), and vaccinations
(influenza and pneumococcal) during 2001. The methods used to identify these
events have been previously described and evaluated by other investigators,
and the codes and eligibility criteria are listed in Table 1.15-17 Based
on the recommended frequencies for each of the 6 services,18-21 we
report the expected annual rates that should be observed given 100% compliance
in Table 1. We excluded fecal occult
blood testing (FOBT) from our primary analysis because its documentation in
claims is less reliable than for sigmoidoscopy or colonoscopy,15 and
because its effectiveness outside research settings has been questioned22,23; however, in subsequent analyses,
we determined that our findings would not have been altered had we decided
to include FOBT as an outcome.
We characterized physicians by their (1) training, qualifications, and
number of years in practice; (2) sex; and (3) practice setting. Training and
qualifications included physician self-reported primary specialty (general
internal medicine or family practice/general practice); board certification
in their primary specialty; and whether their medical school education was
completed in the United States (including Puerto Rico) or Canada, rather than
another country. Practice setting variables included practice type and size
(solo/2-person, small group of 3-10 physicians, medium/large groups of ≥11
physicians, and all other practice types) and payer mix, including the percentage
of practice revenue derived from Medicare, Medicaid, and managed care. Practices
were also characterized based on physicians’ survey responses to 2 questions
about whether computers or other information technology was available to (1)
generate physician reminders about preventive services, or (2) to obtain information
about treatment alternatives or recommended guidelines. We used a composite
variable that dichotomized these responses as “neither” compared
with “1 or both,” but our results did not change in analyses using
a separate variable for each information technology tool. Practice location
was characterized as urban (00-03) or rural (04-09) based on metropolitan
statistical area codes in the 2001 Area Resource File.
Patient characteristics were ascertained from Medicare files and included
age as a continuous variable, sex, race/ethnicity (white, black, other), and
comorbidity based on the index described by Klabunde et al.24 Community
variables were derived from 2000 US Bureau of the Census data in the Area
Resources File and included median household income of residents aged 65 years
and older in the beneficiary’s ZIP code; percentage of adults aged 25
years and older in the county who completed 12 or more years of schooling;
and in the model for delivery of mammography, the number of radiologists per
1000 capita in the county.
Individual beneficiaries were the unit of analysis. Each patient was
assigned to a single usual-source-of-care physician, but each physician could
serve as the usual source of care for multiple beneficiaries. Reported percentages
are therefore weighted to represent estimates for the national population
of Medicare beneficiaries aged 65 years and older, using CTS survey weights
to take into account the complex physician sampling strategy.
We used logistic regression to analyze the association between physician
and practice characteristics and beneficiary delivery of each of the 6 preventive
services. We used SUDAAN software to adjust estimates and variances given
the complex survey sampling strategy and the clustering of beneficiaries within
physicians.25 This study was approved by the
institutional review contractor for Mathematica Inc. P = .05
was set as significant.
Of 12 406 physicians who responded to the CTS survey, 8517 (83%)
had claims represented in the 5% Carrier File. Of these, 3660 (22%) were both
traditional primary care physicians (general internal medicine or family/general
practice) and served as the usual source of care for at least 1 of 24 581
beneficiaries (Figure). The mean number
of evaluation and management visits with the usual-source-of-care physician
for these beneficiaries was 4.5 (range: 1-47), representing a median of 70%
(interquartile range, 50%-100%) of each beneficiary’s total evaluation
and management visits.
The 24 581 beneficiaries derived from a population of 1 332 985
Medicare beneficiaries aged 65 years and older who had at least 1 claim in
the 2001 5% Carrier File. The beneficiaries included in our study were largely
similar to the 1 308 404 beneficiaries not included along the dimensions
of mean age (75.4 vs 75.6 years), race (88.2% vs 87.4% white and 7.6% vs 7.4%
black), sex (63.5% vs 60.4% female), median income in the ZIP code ($49 944
vs $47 903), and comorbidity score (0.55 vs 0.54). Our study population
was also similar to the 15 435 beneficiaries aged 65 years and older
who saw a CTS primary care physician for evaluation and management at least
once in 2001 but for whom that physician was not their usual source of care
(data not shown).
In 2001, the proportions of eligible beneficiaries receiving the 6 recommended
preventive services were 47.9% for diabetic eye examinations, 55.9% for hemoglobin
A1c monitoring, 46.7% for mammography, 9.0% for colon cancer screening,
46.5% for influenza vaccination, and 8.0% for pneumococcal vaccination.
Although preventive care increased in conjunction with median income
in beneficiaries’ ZIP codes, it was suboptimal even among beneficiaries
living in areas with the highest incomes. A total of 53.2% of beneficiaries
residing in ZIP codes in the highest-income tercile received diabetic eye
examinations vs 44.9% for those in the lowest-income tercile. The analogous
comparisons were 59.5% vs 50.9% for hemoglobin A1c monitoring,
50.8% vs 39.8% for mammography, 10.3% vs 8.0% for colon cancer screening,
50.8% vs 41.5% for influenza vaccination, and 8.7% vs 7.3% for pneumococcal
Physician Training, Experience, and Delivery of Preventive
Services. Physician training was consistently associated with better
delivery of preventive services (Table 2).
Beneficiaries with board-certified physicians as their usual source of care
were more likely to receive each of the preventive services evaluated except
for diabetic eye examinations. Similarly, beneficiaries cared for by physicians
who graduated from US or Canadian medical schools were more likely to receive
each preventive service (P<.05 for all comparisons
except hemoglobin A1c monitoring). The comparative advantage for
beneficiaries with general internists as their usual source of care rather
than family/general practitioners was seen for diabetic eye examinations,
mammograms, colon cancer screening, and pneumococcal vaccination. We found
no association between delivery of services and the number of years that the
usual-source-of-care physician had been in practice.
Physician Sex and Delivery of Preventive Services. Sex of the physician was not associated with delivery of services,
except that beneficiaries with female physicians as their usual source of
care were more likely to receive mammograms and less likely to receive influenza
vaccination than those whose physician was male.
Physician’s Practice Setting and Delivery of
Preventive Services. For each of the 6 preventive services, beneficiaries
treated by physicians in group practices (≥3 physicians) were more likely
to receive the service than those in other types of practices; there was no
clear relationship between increased group size and delivery of services.
Similarly, beneficiaries cared for in practices with lower relative Medicaid
revenues were significantly more likely to receive each of the services, with
a consistent graded relationship between the percentage of practice revenues
derived from Medicaid and lower likelihood of service delivery (P<.05 for all comparisons between lowest and highest revenue terciles
except for colon cancer screening). We found that availability of information
technology to access clinical guidelines or to generate reminders for preventive
care was associated with delivery of diabetic eye examinations and pneumococcal
vaccination, but not other services (Table 2).
Neither urban vs rural practice location nor the percentage of practice revenue
derived from managed care or Medicare seemed to influence delivery of services
(data not shown).
The bivariate findings in Table 2 largely
persisted in multivariable analyses adjusting for patient and community characteristics,
although some estimates were amplified and others attenuated (Table 3). Associations between having a board-certified usual-source-of-care
physician and delivery of each service except for the 2 cancer screening services,
and those between being cared for in a group practice and delivery of influenza
and pneumococcal vaccinations and colon cancer screening were no longer statistically
Our results did not change appreciably when we expanded the population
to all beneficiaries who had at least 1 visit with any CTS physician, or when
we limited the population to only beneficiaries who had at least 5 visits
with their CTS usual-source-of-care physician, although statistical significance
levels changed for some associations (sample sizes of physicians were reduced
by >50% in the latter case). Neither did our results change when we redefined
the usual-source-of-care physician as the physician with whom a beneficiary
had the majority (≥50%) of outpatient visits. Results were also similar
when we used FOBT, sigmoidoscopy, and colonsocopy as separate outcome variables
for colon cancer screening were no longer statistically significant.
We evaluated the delivery of preventive services to Medicare beneficiaries
who had a usual source of care to determine whether differences between primary
care physicians or their practice settings influence quality of care. We focused
on preventive care because of its importance in health promotion, and because
the best approach for improving delivery of preventive services has not been
clarified. We found that the delivery of preventive services falls well short
of the ideal in important and routinely encountered clinical scenarios. Consistent
with prior reports, all of the proportions of beneficiaries receiving each
of the services are lower than desired based on clinical guidelines (Table 1).18-21,26-32 Moreover,
we found that there were large variations in the quality of care for beneficiaries
contingent on the characteristics of the treating physicians and their practices.
The most substantial differences were those seen at the practice level.
Beneficiaries who had usual-source-of-care physicians in group practices
were more likely to receive preventive services than those treated in solo/2-person
practices or institution-based practices. Our results support hypotheses that
group practices in general may deliver higher quality of care,33-35 although
contrary to commentators who conjecture that particularly large group practices
provide higher quality care,34 we did not find
such a graded relationship. Explanations for the shortfalls in small practices
are not readily apparent. Large practices may have an easier time obtaining
access to resources and management systems, such as financial incentives,
data collection systems to support physician profiling, or the availability
of ancillary staff.36,37 Some
studies suggest that large practices may also place more emphasis on quality
monitoring, reporting, and improvement.38
We found a consistent inverse association between lower percentage of
practice revenue derived from Medicaid and delivery of preventive services.
It may be that this association reflects socioeconomic differences between
patients or practice settings not captured in our analyses. However, another
possibility worth considering is that there are spillover effects of Medicaid
participation on the quality of care delivered to other patients, due to both
the detrimental effects of lower levels of reimbursement on the practice system
(such as the need for a higher volume of visits to achieve revenue goals)
and the greater challenges inherent in caring for disadvantaged patients.
If these spillover effects are present, then concerns raised about the potential
negative impact of pay-for-performance on practices treating a larger share
of disadvantaged patients may be realized.39
We also found that information technology facilitating access to clinical
guidelines or generating physician reminders for preventive services conferred
only a limited advantage in delivery of those services. This finding contrasts
with other studies demonstrating higher rates of preventive care when physicians
use computer-based reminder systems.40,41 Our
results, like those of Gann et al,33 may more
accurately reflect the effectiveness of such information technology tools
in typical care situations. We note that on the survey, physicians reported
only whether information technology tools were available in their practice,
not whether they actually used them.
Physicians who were board certified in their primary specialty were
more likely to deliver preventive services. These findings strengthen those
of the systematic review by Sharp et al,42 demonstrating
the direct association between quality and board certification. Although viewed
as important by the public and often required for hospital privileges, certification
is currently not included as a condition for physician participation in many
large insurers, including the traditional fee-for-service Medicare program.43 The practical implications of such a requirement
could be substantial in terms of the effect on access. Physicians who are
not board certified made up 15% of the physicians in our study, and these
physicians disproportionately provide care to black patients.7 Our
results suggest, however, that certification may be an important marker of
quality and should be considered as a quality assurance measure.
General internists delivered some preventive services more often than
did family or general practitioners, a set of findings not entirely explained
by differences in practice setting or other covariates. General internists
seemed in particular to be more likely to deliver those services that require
referral to specialists—endoscopy for colon cancer screening, mammography,
and diabetic eye examinations. There was no real difference for services typically
delivered in the primary care physician office—vaccinations and hemoglobin
A1c evaluation. These results are consistent with earlier studies.11,44 Yet there has been little discussion
in the literature to date of potential reasons for this difference across
primary care specialties, for which preventive services delivery is a central
tenet of care. Our findings suggest that family and general practitioners,
when referring their patients to subspecialists for preventive care, may have
either less established subspecialist referral networks or may in general
have patients who are more reluctant to pursue subspecialty evaluation.
We also found that beneficiaries treated by physicians graduating from
a US or Canadian medical school rather than a medical school in another country
were more likely to receive each of the 6 preventive services. These results
add to emerging literature on the relationship between country of medical
school education and the quality of care delivered but do not establish a
definitive relationship. In a systematic review, Mick and Comfort45 found little evidence for a quality difference between
US and foreign graduates, noting that most studies lacked detailed data on
patient and practice characteristics. Other studies have noted that international
medical graduates have extremely heterogeneous backgrounds,46 and
our findings mask such variation.
Our study should be viewed within the context of its limitations. Our
approach to identifying beneficiaries’ usual-source-of care physicians,
although previously validated, is likely imperfect. We evaluated the robustness
of our findings by using some reasonable alternatives, and we did not see
meaningful alterations in our reported effects, but we cannot be certain that
the physicians in our analysis considered themselves to be the source of primary
care for the beneficiaries in our study. Second, we had incomplete data on
patient socioeconomic status, and we lacked data on beneficiaries’ care
preferences and their rates of refusal of preventive services. Third, our
reliance on claims data to measure delivery of services may have introduced
bias in cases where inaccuracies or incompleteness in claims are associated
with important physician or practice characteristics, such as practice type
(eg, if large group practices are more or less efficient at filing Medicare
claims) or percentage of revenue derived from Medicaid (eg, if practices with
high Medicaid volume practices have less administrative infrastructure for
filing claims). One potential critique of our findings is that claims do not
adequately capture services rendered. Busy physician practices may be lax
about coding for services such as influenza vaccination that are associated
with modest levels of reimbursement. However, we would expect such a phenomenon
to vary by both the reimbursement amount (endoscopies are more expensive than
hemoglobin A1c testing) and whether the primary care physician
usually performed the service (vaccinations vs mammography). That our findings
were consistent across both categories of services suggests that we are capturing
true associations, not ones reflecting differences in billing. Fourth, although
nationally representative, the survey sample size limits our power to detect
urban vs rural differences, as rural physicians were not oversampled. Regarding
colon cancer screening, as it is difficult to measure ideal adherence to guidelines
when physicians can choose from many combinations of different testing modalities,
the endoscopy rates we report should be interpreted only as measures of relative
performance. Finally, claims will not capture delivery of some preventive
services, such as some mammograms delivered through community outreach programs.
However, a large number of these programs do bill Medicare for these services
when the patient is a beneficiary, and so many of these events have been captured
in our study.
Prior studies have reported that the delivery of preventive services
remains below national goals; our results confirm this conclusion.3,47 We found that this shortfall is neither
uniform for all beneficiaries nor explained entirely by characteristics of
the beneficiaries such as their race or income level. Rather, it appears that
some beneficiaries are treated in practice settings or by physicians who deliver
preventive services at particularly low rates. Our results suggest that these
variations in quality are substantial and seem to be greatly influenced by
the structure and revenue sources of physician practices. If we can understand
the mechanisms underlying these relationships, it would be much easier to
identify the key leverage points for quality improvement.
Corresponding Author: Hoangmai H. Pham,
MD, MPH, Senior Health Researcher, Center for Studying Health System Change,
600 Maryland Ave SW, Suite 550, Washington, DC 20024 (firstname.lastname@example.org).
Author Contributions: Dr Pham 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: Pham, Hargraves,
Acquisition of data: Pham, Bach.
Analysis and interpretation of data: Pham,
Schrag, Hargraves, Bach.
Drafting of the manuscript: Pham, Schrag,
Critical revision of the manuscript for important
intellectual content: Pham, Schrag, Hargraves, Bach.
Statistical analysis: Pham, Bach.
Obtained funding: Pham, Bach.
Administrative, technical, or material support:
Pham, Schrag, Hargraves,
Study supervision: Bach.
Financial Disclosures: None reported.
Funding/Support: This study was supported by
grants from the Robert Wood Johnson Foundation (to Drs Pham and Hargraves
and for research conducted at the Center for Studying Health System Change),
the National Cancer Institute (grant RO1CA090226 to Dr Bach), and the American
Cancer Society (grant RSGT-04-012-01-CPPB to Dr Schrag).
Role of the Sponsor: The funding sources were
not involved in the design, data collection, analysis, or manuscript preparation
for this study.
Acknowledgment: We wish to thank Beny Wu, MS,
of Social and Scientific Systems for invaluable aid in data analysis.