Thorndike AN, Rigotti NA, Stafford RS, Singer DE. National Patterns in the Treatment of Smokers by Physicians. JAMA. 1998;279(8):604-608. doi:10.1001/jama.279.8.604
From the Medical Services, Massachusetts General Hospital, General Medicine Division, and the Department of Medicine, Harvard Medical School (Drs Thorndike, Rigotti, Stafford, and Singer), Tobacco Research and Treatment Center (Dr Rigotti), Health Policy Research and Development Unit (Dr Stafford), and Clinical Epidemiology Unit (Dr Singer), Boston, Mass.
Context.— Routine treatment of smokers by physicians is a national health objective
for the year 2000, a quality measure for health care plans, and the subject
of evidence-based clinical guidelines. There are few national data on how
physicians' practices compare with these standards.
Objective.— To assess recent trends in the treatment of smokers by US physicians
in ambulatory care and to determine whether physicians' practices meet current
Design.— Analysis of 1991-1995 data from the National Ambulatory Medical Care
Survey, an annual survey of a random sample of US office-based physicians.
Setting.— Physicians' offices.
Patients.— A total of 3254 physicians recorded data on 145716 adult patient visits.
Main Outcome Measures.— The proportion of visits at which physicians (1) identified a patient's
smoking status, (2) counseled a smoker to quit, and (3) used nicotine replacement
Results.— Smoking counseling by physicians increased from 16% of smokers' visits
in 1991 to 29% in 1993 (P<.001) and then decreased
to 21% of smokers' visits in 1995 (P<.001). Nicotine
replacement therapy use followed a similar pattern, increasing from 0.4% of
smokers' visits in 1991 to 2.2% in 1993 (P<.001)
and decreasing to 1.3% of smokers' visits in 1995 (P=.007).
Physicians identified patients' smoking status at 67% of all visits in 1991;
this proportion did not increase over time. Primary care physicians were more
likely to provide treatment to smokers than were specialists. All physicians
were more likely to treat patients with smoking-related diagnoses.
Conclusions.— US physicians' treatment of smokers improved little in the first half
of the 1990s, although a transient peak in counseling and nicotine replacement
use occurred in 1993 after the introduction of the nicotine patch. Physicians'
practices fell far short of national health objectives and practice guidelines.
In particular, patient visits for diagnoses not related to smoking represent
important missed opportunities for intervention.
CIGARETTE SMOKING is the single most important cause of death and disability
in the United States.1 Because an estimated
70% of smokers visit a physician each year, physicians have the opportunity
to promote smoking cessation.2 Brief physician
counseling and the use of nicotine replacement therapy (NRT) have been shown
to increase the smoking cessation rates of patients.3- 5
The likelihood that a physician will counsel a smoker is increased by routine
identification of a patient's smoking status in the medical record.6,7
This evidence has led several professional and government organizations
to make recommendations to physicians about the treatment of smokers.8- 12
In 1996, the Agency for Health Care Policy and Research released an evidence-based
clinical guideline that directed primary care physicians to identify a patient's
smoking status at every visit, counsel smokers at every visit, and offer NRT
to patients planning to quit.10 National health
promotion objectives for the year 2000 call for increasing to 75% the proportion
of primary care providers who routinely advise cessation and provide assistance
to their patients who smoke.11 The rate of
physicians' advice to smokers is now a quality measure for US health plans.12
Despite this consensus, it is unclear how frequently physicians identify
and treat smokers. Surveys have produced conflicting results depending on
whether information is obtained from patients or physicians. Seventy percent
to 98% of surveyed physicians report that they routinely ask their patients
about their smoking status or record the patient's smoking status on the chart,13- 16 and
46% to 77% of physicians report that they routinely counsel their patients
to quit.13,15,17- 20
In contrast, only half the smokers report having ever
been advised by their physician to quit.2,21- 24
This discrepancy is more than likely attributable to recall bias. Physicians
may overestimate how often they address smoking while smokers may underestimate
how often they are counseled. The National Ambulatory Medical Care Survey
(NAMCS) is an annual office-based survey that US physicians complete during
each patient's visit. It provides a more accurate assessment of physicians'
actual practice than previous surveys.25 We
analyzed NAMCS data collected from 1991 to 1995 to assess recent national
patterns in the routine ambulatory care of smokers.
The NAMCS is an ongoing annual survey of US office-based physicians
conducted by the National Center for Health Statistics (NCHS).25
Doctors of medicine and osteopathic medicine are selected by stratified random
sampling from the American Medical Association and the American Osteopathic
Association listings of all practicing physicians in the United States. The
unit of analysis is the patient visit. Each participating physician completes
a 1-page encounter form after each ambulatory care visit during a randomly
assigned week. Outpatient care provided in hospital settings, by telephone,
or by nonphysician providers is excluded. Physicians record information about
patient demographics, smoking status, expected source of payment, reasons
for the visit, diagnoses, counseling and education provided, and current medications.
Missing data are limited to approximately 5% of patient visits.25
The cross-sectional nature of NAMCS permits patterns in physician practices
to be followed over time but does not allow individual physicians or patients
to be followed longitudinally.
The NCHS uses a complex 3-stage sampling design that has previously
been described.26 To produce unbiased national
estimates, each patient visit is assigned an inflation factor called the patient
visit weight that is based on the probability of selection, the differences
in response rates, and the specialty distributions. All statistical estimates
presented in the results of this study are weighted to reflect national estimates.
The NCHS provides relative SEs for estimates to gauge the reliability of an
estimate for an individual year. An estimate with a relative SE greater than
30% could be unreliable.25
We analyzed data collected from 1991, when smoking status was first
included on the survey, to 1995, the most recent year available. Physician
response rates varied between 70% and 73% for the 5 years.25
All visits by patients aged 18 years or older were included. We examined changes
in physician practices from 1991 to 1995 and then combined data from the 1994
and 1995 surveys to describe recent physician practices. We examined 3 outcomes:
(1) identification of a patient's smoking status, (2) provision of smoking
counseling, and (3) reporting of NRT use. Physicians identified a patient's
smoking status by answering the question, "Does patient smoke cigarettes?"
Smoking status was categorized as "known" if the answer was yes or no; otherwise,
smoking status was "unknown." Physicians recorded smoking counseling by checking
the appropriate box under "Counseling/Education." Nicotine replacement therapy
that included both nicotine gum and patches was recorded on the survey form
under "Medications." Nicotine replacement products were available only by
prescription during the survey years. All adult patient visits were included
in the analysis of smoking status. Analyses of smoking counseling and NRT
were restricted to visits by patients identified as smokers.
Independent predictors of smoking status identification, smoking counseling,
and NRT use were determined with weighted multiple logistic regression.27 Covariates included in the models for all 3 outcomes
were survey year, patient demographics (age, sex, and race), geographic region,
expected payment source for the visit, diagnoses and reasons for visit, physician
specialty, and counseling for other cardiovascular risk factors (cholesterol,
weight reduction, and exercise) provided during the visit. Because the 3-stage
sampling design could not be accounted for in the logistic regression models,
statistical significance was defined conservatively as a 2-tailed P value at a level of ≤.01. To analyze time trends, each year was
included in the multivariate models as a categorical variable using the year
1993 as the reference variable.
Four categories of diagnoses were assessed as predictor variables because
of their association with adverse outcomes from continued smoking. They were
cardiovascular disease, chronic pulmonary disease, diabetes, and pregnancy.
Each category represented a combination of reason for visit codes created
by the NCHS for the NAMCS25 and the International Classification of Diseases, 9th Revision, diagnosis codes.28 The cardiovascular disease category included hypertension,
coronary artery disease, cerebrovascular disease, and peripheral vascular
disease. Chronic pulmonary disease included chronic bronchitis, emphysema,
and asthma. A fifth category, general medical examination, was included because
we hypothesized that a physician would be more likely to identify smoking
status and to counsel smokers during this type of visit. All diagnosis categories
were created as binary variables, eg, cardiovascular disease vs no cardiovascular
disease; therefore, each visit could be included in more than 1 diagnosis
category. Physicians were categorized as primary care (general internists,
family practitioners, and generalists) or specialists (all other specialties).
Expected payment source for the visit was divided into 5 categories: health
maintenance organization, private insurance, Medicaid, Medicare, and other
To further explore the association between physician intervention and
the specific problem addressed during the visit, we calculated the rate of
smoking counseling for the most common primary diagnoses of smokers visiting
primary care physicians. A primary diagnosis is the diagnosis associated with
the patient's primary reason for making the visit and was determined by using
the first diagnosis written on the survey by the physician. We then grouped
clinically related primary diagnoses using Schneeweiss diagnosis clusters.29
Data were available on 145716 adult patient visits to 3254 physicians
from 1991 through 1995. Smoking status was identified at 95540 visits, which
represented 66% of all visits. Patients were identified as smokers at 17632
visits from 1991 through 1995, which represented 12% of all visits and 18%
of visits where smoking status was known. The proportion of visits in which
a patient's smoking status was identified did not change significantly between
1991 and 1994 but fell in 1995 from 67% of all visits in 1994 to 61% in 1995
(P<.001). The proportion of visits by patients
identified as smokers did not change significantly from 1991 to 1995 (Figure 1). (All comparisons presented in
the "Results" section are adjusted; see "Methods.")
Smoking counseling was provided at 3302 smokers' visits, and NRT was
reported at 161 smokers' visits, which represented 22% and 1%, respectively,
of visits by smokers from 1991 through 1995. Smoking counseling increased
from 16% of smokers' visits in 1991 to a peak of 29% in 1993 (P<.001) and then declined to 21% of smokers' visits in 1995 (P<.001) (Figure 2,
top). Primary care physicians counseled smokers at a significantly higher
rate than specialists in each year except 1991 (1992-1995, P<.001). Between 1991 and 1993, counseling by primary care physicians
increased from 20% to 38% of smokers' visits (P<.001),
while counseling by specialists only rose from 12% to 19% of visits (P<.001). Between 1993 and 1995, counseling rates fell
for both types of practitioners from 38% to 29% of smokers' visits to primary
care physicians (P<.001) and from 19% to 14% of
smokers' visits to specialists (P<.001).
The NRT use was reported infrequently. We caution that these estimates
may be unreliable because the relative SEs were greater than 30% (see "Methods").
The NRT use increased from 0.4% of smokers' visits in 1991 to 2.2% of smokers'
visits in 1993 (P<.001) with a subsequent decrease
in 1995 to 1.3% (P=.007) (Figure 2, bottom). Primary care physicians reported NRT use significantly
more often than specialists in 1993 through 1995 (1993, P=.008; 1994-1995, P<.001). Between 1991
and 1993 NRT use increased among both primary care physicians (0.6% to 3.0%
of smokers' visits, P<.001) and specialists (0.2%
to 1.3% of smokers' visits, P<.001). The NRT use
decreased significantly among specialists between 1993 and 1995 (1.3% to 0.4%, P<.001) but not among primary care physicians (3.0%
to 2.4%, P=.3). Nicotine gum accounted for all NRT
reported in 1991 and 1992 while nicotine patches accounted for more than 90%
of NRT reported in 1993 through 1995. Physicians provided smoking counseling
at 82% of visits at which NRT was reported.
Table 1 displays factors
independently associated with physicians' identification of a patient's smoking
status and provision of smoking counseling at visits in 1994 and 1995. Primary
care physicians were more likely than specialists to identify a patient's
smoking status and were twice as likely to counsel about smoking. All physicians
were at least 1.5 times more likely to identify a patient's smoking status
and counsel for smoking at visits by patients with cardiovascular disease,
chronic pulmonary disease, or pregnancy. Smoking counseling was more likely
to occur at a general medical examination than at other types of visits (37%
vs 22% of visits, P<.001), and at a first visit
compared with a return visit (25% vs 23%, P=.002).
Physicians were no more likely to identify a patient's smoking status at a
general medical examination or at a new patient visit. Counseling about smoking
was more likely to occur at visits that also included counseling for other
cardiovascular risk factors. Elderly patients were less likely than younger
patients to have smoking status identified and to be counseled about smoking.
Men and nonwhites were less likely than women and whites to have smoking status
identified but were no less likely to be counseled for smoking. A patient's
insurance status had little effect on the likelihood that a smoker would be
identified or counseled.
Figure 3 displays the rate
of smoking counseling for the most common primary diagnoses of smokers visiting
primary care physicians in 1994 and 1995. There is wide variability in physician
counseling depending on the patient's primary diagnosis for the visit. Physicians
were most likely (≥35% of visits) to counsel about smoking at visits for
acute and chronic respiratory disorders, cardiovascular disorders, alcohol
and drug abuse, peptic diseases, and diabetes. These diagnoses are all caused
or complicated by tobacco use. In contrast, physicians were least likely (≤20%
of visits) to counsel smokers who were seen for musculoskeletal disorders,
nonrespiratory infections, and lacerations and contusions, which are all conditions
unrelated to smoking.
This study examined national patterns in physicians' treatment of smokers
in office practice during the first half of this decade. It is distinguished
from previous work in that it analyzed data collected over 5 years from a
large nationally representative sample of US physicians who reported their
actions at the time of a patient visit rather than summarizing their practice
patterns retrospectively as most previous studies have done. This study demonstrates
that physicians' practices clearly fall short of national health goals for
the year 2000, current practice guidelines, and new performance standards
for health plans.10- 12
Primary care physicians did better than specialists but still fell far short
of these goals. However, more importantly, the study failed to find evidence
of sustained improvement in physicians' practice during the early 1990s. Physicians'
rates of providing smoking counseling and prescribing NRT increased to a peak
in 1993 but decreased thereafter, and the rate at which physicians identified
a patient's smoking status did not improve over the 5-year study period.
Physicians reported counseling about smoking at only 23% of office visits
by patients whom they identified as smokers. As expected, counseling occurred
more often at a general medical examination than at a visit for a specific
problem, but even so, physicians addressed smoking at only 37% of smokers'
general medical examinations. These smoking counseling rates indicate that
physicians' treatment of smoking is even further from recommended practice
standards than previous surveys have suggested.17,18,20
Previous physician surveys are more subject to recall bias than the NAMCS
because physicians retrospectively summarize their practice patterns, and
their reports may reflect their intentions rather than their actual practices.
While most physicians believe that smoking is an important health behavior
and rank counseling about smoking as the most important preventive service
that they can provide,14,16,20
they also report that counseling is frustrating and time-consuming.20
Our analysis documents a nearly 6-fold variation in physicians' smoking
counseling practices according to the reason for the patient's visit. Physicians
were more likely to address smoking behavior if the patient's presenting problem
was caused or exacerbated by smoking or if the patient had a chronic smoking-related
illness. This is consistent with previous surveys of patients and physicians.13,17,21,22 Recent
national guidelines recommended that smoking should be addressed even when
patients are seen for problems unrelated to smoking.10
This clearly was not included in the physician's practice in the early 1990s.
Our results represent a first nationwide look at physicians' use of
NRT at a time when it was available only by prescription, although conclusions
are limited by the small numbers of visits at which the therapy was used.
The low prevalence of NRT use is most likely due to the cross-sectional nature
of the survey, the low prevalence of smokers who are ready to quit at any
given visit, and the short-term use of the medication. The NRT use rose to
a peak in 1993, the year after the transdermal nicotine patch was introduced
to the US market, and subsequently declined. This pattern resembles the US
sales of the patch, whose annual sales rose rapidly after its introduction
in 1992 to $600 million and subsequently declined to $250 million.30
It is not surprising that the introduction of the nicotine patch led
to an increase in physicians' use of nicotine replacement products. However,
our data suggest that the introduction of the patch also contributed to a
transient increase in physician counseling about smoking even at visits where
NRT was not prescribed. Although many factors influence physician behavior,
we are unaware of any other temporal change that could explain the transient
peak in physicians' smoking counseling in 1993. The nicotine patch may have
influenced physician counseling in 2 ways. First, the patch provided physicians
with a new therapeutic option that may have encouraged them to discuss smoking
cessation with patients. Second, the patch was directly marketed to consumers
and attracted considerable media attention, creating consumer awareness and
demand for a product available only from physicians.31- 33
Only 25% of patients who were filling a nicotine patch prescription in 1994
stated that they learned about the patch from their physicians,31
and between 60% and 80% of patients who used the patch in 1992 reported that
they requested the patch from their physician.32,33
Consumer demand for the product may have increased physician smoking counseling
rates even when a prescription for nicotine replacement was not provided.
Identifying a patient's smoking status is a necessary first step because
treatment cannot be provided if the physician does not recognize that the
patient is a smoker.8- 10
While there are no previous national estimates of how frequently physicians
assess patients' smoking status, 70% to 92% of physicians surveyed in the
1980s reported that they determined the smoking status of all their patients,13- 15 and 98% of Massachusetts
physicians reported in 1994 that they "regularly gathered information about
smoking."16 Our data indicate that these surveys
overestimate actual physician practices and that physicians were unaware of
a patient's smoking status at one third of all visits. This proportion did
not change when the analysis was limited to new patient visits or visits for
a general medical examination when assessment of smoking behavior might be
more likely to occur. Physicians miss the opportunity to counsel a substantial
portion of their patients who smoke because they are unaware of their smoking
status, which may partially explain the discrepancy between patient and physician
reporting of smoking counseling in previous surveys.
The visit-based nature of the NAMCS is a strength of this study but
also presents limitations. Our estimates reflect only the probability of being
counseled at a visit not the probability of an individual patient being counseled
over a given period, such as a year. Patients who visit physicians frequently
might be less likely to be treated for smoking at an individual visit but
more likely to be treated over a year, as a recent survey of patients in 4
midwestern states observed.34 Our results may
overestimate the amount of physician intervention beyond advice to quit because
some physicians may have interpreted "counseling" to only mean giving advice
to quit. The NAMCS may have some recall bias because physicians fill out the
survey after a patient encounter, but this possibility is much less than in
previous physician surveys. The prevalence of smokers among visits where smoking
status was identified was 18%, which is somewhat lower than the US adult smoking
prevalence of 25%. 35 It is likely that some
smokers did not truthfully report their smoking status to their physician.
There may be more error in the estimates of counseling and NRT use than in
the estimates of smoking status identification, because estimates of smoking
counseling and NRT use were limited to visits by patients identified as smokers,
and these visits may not be representative of visits by all smokers. Finally,
in the logistic regression models that estimated year-to-year trends and aggregated
yearly data, we were not able to account for correlations in time between
estimates. The effect of these correlations on statistical inference is to
increase the SEs but not to affect the point estimates of the relative odds.
In conclusion, in the first half of the 1990s physicians made little
progress in the treatment of smokers. This finding highlights the importance
of efforts to institutionalize the identification of smoking status into office
practice by using system-wide interventions or assessing smoking status as
if it were a vital sign.6,7,10
Our observation of a transient increase in smoking counseling in 1993, the
year following the introduction of the nicotine patch, was unexpected and
suggests that the introduction of pharmacotherapies for smoking cessation
may influence not only physicians' prescribing practices but also their willingness
to counsel smokers. It will be important to determine the effects of more
recent events, such as the shift of nicotine gum and patches to nonprescription
status in 1996 and the introduction of new prescription drugs for smoking
cessation in 1997, on physicians' behavior.