Angus DC, Kelley MA, Schmitz RJ, White A, Popovich, Jr J, for the Committee on Manpower for Pulmonary and Critical Care Societies
(COMPACCS) . Current and Projected Workforce Requirements for Care of the Critically Ill and Patients With Pulmonary DiseaseCan We Meet the Requirements of an Aging Population?. JAMA. 2000;284(21):2762–2770. doi:10.1001/jama.284.21.2762
Author Affiliations: Department of Anesthesiology and Critical Care Medicine and Center for Research on Health Care, University of Pittsburgh, Pittsburgh, Pa (Dr Angus); Department of Medicine, Henry Ford Hospital, Detroit, Mich (Drs Kelley and Popovich); and Mathematica Policy Research (Dr Schmitz), Abt Associates Inc (Dr White), Cambridge, Mass.
Caring for the Critically Ill Patient Section Editor: Deborah J. Cook, MD, Consulting Editor, JAMA.
Context Two important areas of medicine, care of the critically ill and management
of pulmonary disease, are likely to be influenced by the aging of the US population.
Objective To estimate current and future requirements for adult critical care
and pulmonary medicine physicians in the United States.
Design, Setting, and Participants Analysis of existing population, patient, and hospital data sets and
prospective, nationally representative surveys of intensive care unit (ICU)
directors (n = 393) and critical care specialists (intensivists) and pulmonary
specialists (pulmonologists) (n = 421), conducted from 1996 to 1999.
Main Outcome Measures Influence of patient, physician, regional, hospital, and payer characteristics
on current practice patterns; forecasted future supply of and demand for specialist
care through 2030. Separate models for critical care and pulmonary disease.
Base-case projections with sensitivity analyses to estimate the impact of
future changes in training and retirement, disease prevalence and management,
and health care reform initiatives.
Results In 1997, intensivists provided care to 36.8% of all ICU patients. Care
in the ICU was provided more commonly by intensivists in regions with high
managed care penetration. The current ratio of supply to demand is forecast
to remain in rough equilibrium until 2007. Subsequently, demand will grow
rapidly while supply will remain near constant, yielding a shortfall of specialist
hours equal to 22% of demand by 2020 and 35% by 2030, primarily because of
the aging of the US population. Sensitivity analyses suggest that the spread
of current health care reform initiatives will either have no effect or worsen
this shortfall. A shortfall of pulmonologist time will also occur before 2007
and increase to 35% by 2020 and 46% by 2030.
Conclusions We forecast that the proportion of care provided by intensivists and
pulmonologists in the United States will decrease below current standards
in less than 10 years. While current health care reform initiatives and modification
of existing practice patterns may temporarily forestall this problem, most
anticipated effects are minor in comparison with the growing disease burden
created by the aging US population.
Meeting the health care needs of the American people in the coming decades
requires accurate forecasts of likely requirements, not only in terms of dollars,
but also in terms of staffing and facilities. Unfortunately, such forecasts
are difficult because many factors influence health care delivery. One factor
likely to increase demand for services is the aging of the US population.
During the next 30 years, Medicare enrollment will grow by more than 50%.1 Because the elderly consume considerable health care
resources, demand for medical services is expected to outpace other indices
of the economy.2
This increased demand might be offset by other factors, such as the
growth of managed care. It is argued that physicians create part of the demand
for their services.3,4 One influential
study found managed care plans, by carefully screening physician services,
lowered demand for specialist care.5 Because
of anticipated growth in managed care, the Council on Graduate Medical Education
predicted specialists would now be oversupplied.5,6
Concern over such potential changes in physician requirements generated other
forecasts of the physician labor market in the 1990s.3,5- 9
Not all studies reached the same conclusion and some found little evidence
to suggest a significant surplus of specialists in the near future.10,11 This controversy stems from different
assumptions about future health care12 and
has stimulated several physician groups to conduct further studies.13- 16
Two of the largest areas in medicine are the care of the critically
ill and the management of pulmonary disease. These areas are likely to be
profoundly influenced by an increase in the elderly population. Therefore,
the American College of Chest Physicians, the American Thoracic Society, and
the Society of Critical Care Medicine formed the Committee on Manpower for
Pulmonary and Critical Care Societies (COMPACCS). The goal of this committee
was to determine current patterns of care for the critically ill and patients
with pulmonary disease, anticipate how demand for care might change in the
future, and project the supply based on the current workforce and training.
Specifically, we wanted to test 2 competing hypotheses: first, that changes
in US health care organization will lead to an oversupply of specialists and,
second, that changes in the demography of the US population will lead to an
We used population, patient, hospital, and clinician data to determine
national estimates of the use of adult critical care and pulmonary services
in the United States, the proportion of these services provided by specialists,
and the number and practice characteristics of the current specialist workforce.
From these data, we constructed models to describe the current and future
demand for and supply of critical care specialitsts (intensivists) and pulmonary
specialists (pulmonologists). We modeled critical care and pulmonary medicine
separately. Our models were based on relationships between the size and growth
of the US population; the age- and disease-specific use of intensive care
units (ICUs); the age-specific prevalence of pulmonary disease; and the training,
retirement, work hours, and practice patterns of intensivists and pulmonologists.
We tested the robustness of our model design and results in sensitivity analyses,
which explored the effect of alternative scenarios of disease prevalence,
physician work effort, number of new specialists trained, and health care
organization. Our methods followed the general approaches for modeling current
and future workforce supply and requirements described elsewhere.17,18 The full COMPACCS report is available
and a copy of the physician supply and demand forecast model can be downloaded
We derived current and projected age-specific population estimates from
the US Census (available at: http://www.census.gov). We determined
the number of ICU days by disease and by age from the 1995 Healthcare Financing
Administration Medical Provider Analysis and Review file for those aged 65
years and older and the 1995 New Jersey Hospital Reporting System hospital
reporting file for those aged 18 to 64 years. We used age-specific population-based
ICU use rates to generate national estimates. We drew similar data on inpatient
days for patients hospitalized with pulmonary conditions from the fifth release
of the nationwide inpatient sample of the Healthcare Cost and Utilization
Project-319 and generated national totals using
weights provided on the file. We determined the number of ambulatory visits
for pulmonary conditions from the 1997 enrollee and encounter data from United
HealthCare Corp (Minneapolis, Minn), one of the nation's largest health care
insurance companies. United HealthCare represents approximately 2.5 million
person-years of care from more than 30 geographically dispersed health plans.
We calculated disease- and age-specific use rates to generate national estimates
of all pulmonary ambulatory visits and the proportion provided by pulmonologists.
We gathered similar data from the 1994 Washington state Medicare Part B claims
data, which had been linked previously to enrollee, hospital, and clinician
data to generate population-based use rates.20
To document specialist practice patterns, we surveyed 10% of a stratified
random sample of all physicians (n = 10,244) in the 1996 American Medical
Association Physician Masterfile who were currently practicing in the United
States with a primary or secondary specialty of critical care or pulmonary
medicine.21 The sampling frame was stratified
by age, sex, specialty, and region—classifying region both by population
size—using the Beale coding system,22,23
and by managed care penetration, using the 1997 University HealthSystem Consortium
classification (University HealthSystem Consortium Member Market Classification,
Oak Brook, Ill). We altered sampling probabilities to ensure that particular
smaller subgroups (eg, women, surgeons, and rural physicians) were adequately
To understand physician practice and organization within ICUs, we surveyed
a stratified random sample of hospital-appointed ICU directors at all adult
noncardiac ICUs in the United States. We selected all acute care hospitals
(n = 4440) in the United States with 1 or more ICUs from the 1996 SMG Hospital
Market database (SMG Marketing Group, Chicago, Ill). From this list we produced
a sampling frame that was stratified by hospital size and region and classified
as above. We oversampled larger hospitals because they have a greater proportion
of ICUs, including specialty ICUs, such as trauma or burn units. We telephoned
each hospital in the sample to determine the number of ICUs and the number
of hospital-appointed directors of each unit. We then selected units within
each hospital (n = 1173), using a random stratified sampling technique to
ensure a representative sample of general (medical/surgical), medical, surgical,
and specialty ICUs.
We asked ICU directors to specify how care was provided for each patient
in their ICU on the day the survey was completed. We used 4 care models: a
full-time intensivist model, in which all or most of a patient's care is directed
by an intensivist (an intensivist was defined as an attending physician who,
by training or experience, provides care for the critically ill in a role
broader than that provided by a consultant specialist); a consultant intensivist
model, in which an intensivist consults for another physician to coordinate
or assist in critical care but does not have primary responsibility; a multiple
consultant model, in which multiple specialists are involved (in such instances,
a pulmonologist or intensivist may be consulted for ventilator management
but no one is designated as the consultant intensivist); and a single physician
model, in which the primary physician provides all ICU care.
Response rates were 42.1% (n = 421) and 33.5% (n = 393) for the physician
and ICU director surveys. Respondents and nonrespondents to the physician
survey were generally similar in age and base specialty training and near-identical
regional distribution and professional activities. However, respondents were
more likely to be men (86% vs 79%; P<.01), more
likely to be surgeons (14% vs 10%; P<.01), and
less likely to be internists (14% vs 19%; P<.01).
There were no significant differences in response rates by region or subgroup
in the ICU director survey. We generated nationally representative estimates
from the surveys, adjusting for both sampling probabilities and response rates.
To determine demand for intensivists, we calculated ratios of total
annual hours provided by intensivists in adult, noncardiac ICUs (estimated
from the Physician Masterfile and survey) to total numbers of adult, noncardiac
ICU patient days per annum (estimated from Medical Provider Analysis and Review
file and state patient data). We estimated separate ratios for each combination
of managed care penetration level, disease category, and patient age group.
We determined total demand by summing the products of each combination's ratio
and number of ICU patient days.
We determined demand for pulmonologists similarly, constructing separate
models for inpatient and ambulatory pulmonary services. For inpatient care,
we calculated the ratio of total hours (estimated from the Physician Masterfile
and survey) reported by pulmonologists in direct, non-ICU, inpatient care
of pulmonary disease to the number of non-ICU inpatient days during hospital
admissions for pulmonary conditions (estimated from Healthcare Cost and Utilization
Project-3 nationwide inpatient sample). We assumed the proportion of cases
treated by pulmonologists and the average time per case were equal within
disease and age groups. For ambulatory care, we calculated the ratio of hours
spent by pulmonologists providing pulmonary (as opposed to general medical
or other) ambulatory care (estimated from the Physician Masterfile and survey)
to the total number of ambulatory visits for pulmonary conditions (estimated
from the United HealthCare data). We estimated the number of ambulatory visits
and the proportion involving pulmonologists both by disease and age group,
assuming the time per visit was equal across disease and age group.
Under the base case, we modeled future demand as a function of the US
census estimates of the changing age structure multiplied by current age-
and disease-specific use rates.
We defined the supply model as the product of the number of specialists
multiplied by the average annual clinical hours spent by specialists in intensive
care, inpatient pulmonary care, and ambulatory pulmonary care. We used weighted
least-squares regression to estimate the conditional distribution of total
hours worked, total clinical hours, and clinical hours by setting.24,25 Model covariates were physician age,
sex, specialty, faculty status, and degree of managed care penetration in
the physician's local market area, with interaction terms for age and sex
and for age and specialty. We tested alternative specifications of the regression
models using different measures of managed care penetration and sets of interaction
terms. In each case, the regression equation that we used for model construction
was the one attaining the highest value of the Akaike information criterion.26,27
Under the base case, we modeled future supply assuming changes in annual
hours would be due to the changing age structure of the specialist population,
the number of physicians graduating from fellowship programs, and the number
of physicians retiring from practice. We assumed the number of physicians
entering specialist practice matched the number of graduates from all critical
care and pulmonary training programs. We obtained the current number of graduates
from published data.28 Based on recent reductions
in the number of graduates and external funding restrictions on training programs,
we assumed the number of graduates from pulmonary, pulmonary/critical care,
and internal medicine/critical care programs would decrease by another 10%
in the next 10 years. We assumed the proportion of female graduates would
increase to 40% in the next 10 years. We modeled retirement based on the physician
survey responses for each clinical domain and assuming a Weibull density.26,29,30
We tested the sensitivity of our model design and base-case estimate
to (1) the effect of age-independent changes in the incidence, prevalence,
and management of pulmonary diseases and those diseases commonly managed in
the ICU; (2) the spread and penetration of managed care throughout the country;
(3) retirement patterns; (4) the distribution of clinical hours between different
clinical and nonclinical activities; and (5) changes in the number of specialty
trainees. We determined the boundaries for the sensitivity analyses (including
the size and timing of any change in the base model variables) prior to seeing
the forecast model. Boundaries were based on the stochastic error distributions
of variables, literature review, additional survey data, and the expert opinion
of the COMPACCS Committee. When considering ranges, we assumed no major changes
in the reimbursement for specialist services.
More than half of all ICU days (55.8%) were incurred by patients older
than 65 years with the number of days per year per 1000 person-years varying
from 37.3 for adults younger than 65 years to 178.4, 244.9, and 230.9 for
those aged 65 to 74, 75 to 84, and older than 85 years, respectively. Half
of the patients (45.6%) were managed under the multiple consultant model with
23.1% managed under the full-time intensivist model, 13.7% under the consultant
intensivist model, 14.2% under the single physician model, and 3.4% under
other arrangements. Intensivists provided care to at least 1 patient in 59.1%
of all ICUs and provided that care as a full-time intensivist in 29% of ICUs.
Intensivists were significantly more likely to be involved in the care of
patients with respiratory insufficiency, multiple organ failure and sepsis,
and less likely to be involved in the care of postoperative patients and patients
In regression analysis, intensivists were more likely to provide care
in medical ICUs (odds ratio [OR], 1.4; P = .006),
ICUs in larger hospitals (ORs, 4.3 [P = .01] and
3.4 [P = .01] for hospitals with 301-500 and >500
beds), and ICUs with a high proportion of patients covered by managed care
(OR, 1.6 for ICUs with >30% managed care case-mix [P
Two thirds (66.8%) of all inpatient pulmonary days were incurred by
patients older than 65 years with the number of days per year per 1000 person-years
increasing from 38.3 for adults younger than 65 years to 261.7, 443.1, and
634.1 for those aged 65 to 74, 75 to 84, and older than 85 years, respectively.
Pneumonia, chronic obstructive pulmonary disease, and respiratory failure
accounted for 80.5% of all days. Pulmonologists were significantly more likely
to care for patients with chronic obstructive pulmonary disease, asthma, and
respiratory failure and less likely to care for patients with pneumonia.
Unlike inpatient care, most pulmonary ambulatory visits were incurred
by patients younger than 65 years (80.1% and 67.8% of all pulmonary disease-related
visits to generalists and to pulmonologists). The number of visits to pulmonologists
per 1000 person-years increased from 23.9 for patients younger than 65 years
to 84.9, 103.1, and 67.7 for those aged 65 to 74, 75 to 84, and older than
85 years, respectively. With the exception of interstitial lung disease (50.3%)
and respiratory failure (50.4%), pulmonologists generally provided less than
10% of all ambulatory pulmonary care. In the Washington state Medicare population,
the use of pulmonologists was 23.7% and 28.5% higher in those aged 65 to 74
and 75 to 84 years (compared with the United Healthcare data) but similar
(66.5 visits per 1000 person-years) in those older than 85 years.
Characteristics of current intensivists and pulmonologists are outlined
in Table 1. Most are trained in
pulmonary and critical care medicine. The remainder have base specialty training
in internal medicine, anesthesiology, or surgery and are younger, more likely
to be university-affiliated, and include a higher proportion of women. Most
are certified in their base specialty but only half are certified in critical
care, although this varies substantially across subspecialties. Only a small
proportion of the workforce is salaried, and a smaller proportion still are
employed by managed care organizations.
Specialists worked 61 hours per week for 48 weeks (2933 hours per year),
with 78% (2284 hours) of time spent in clinical activities. They spent 26.1%
of their clinical time in the ICU, although this ranged from 23.4% for those
with pulmonary training to 46.2% for nonpulmonary internal medicine-based
intensivists. Pulmonologists spent 67.1% of their non-ICU clinical time providing
pulmonary services. In regression analyses, women worked 307 fewer total annual
hours (P = .01) and 379 fewer clinical hours (P<.01) than men. Specialists in areas with greater than
25% managed care penetration worked 344 more total hours (P<.01), 314 more clinical hours (P<.01),
202 more ICU hours (P = .01), and 268 more pulmonary
hours (P<.01) than others. Older physicians and
full-time faculty also worked fewer clinical hours.
In 1996, there were 354 graduates from pulmonary and pulmonary/critical
care medicine programs, 110 graduates from internal medicine critical care
programs, 67 graduates from surgical critical care programs, and 63 graduates
from anesthesiology critical care programs.
The numbers of intensivists demanded and supplied as projected by the
base model and sensitivity analysis are shown in Figure 1 and in Table 2.
Intensivist hours supplied and demanded remain in rough equilibrium until
2007, both increasing by 10% more than their 1997 values. Thereafter, demand
grows rapidly while supply remains near constant, yielding a shortfall of
intensivist hours equal to 22% of demand by 2020 and 35% by 2030. This discrepancy
is due primarily to the increased elderly population as baby boomers age.
In sensitivity analysis, the shortfall widens markedly if there is demand
for intensivist care to a greater proportion of ICU patients. Otherwise, most
supply and demand variables have only a small effect on the ratio of supply
to demand in comparison with the effect of an aging population. Although changes
in health care delivery, such as the effect of increased managed care penetration,
will affect the balance of supply and demand during the next 10 to 15 years,
no variable led to a large reduction in the ever-widening shortfall predicted
The numbers of pulmonologists demanded and supplied as projected by
the base model and sensitivity analysis are shown in Figure 1 and in Table 3.
Again, this model shows an increasing demand with time while supply is near
constant. The shortfall develops sooner because the current workforce is anticipating
retirement at a rate faster than the graduation rate. The shortfall in pulmonologist
hours is predicted to be 35% by 2020 and 46% by 2030.
In sensitivity analyses, significant deployment of nonpulmonary clinical
time to pulmonary care delays the onset and reduces the magnitude of the shortfall
but demand still outstrips supply by 10% in 2020 and 16% in 2030. Again, the
finding of a significant discrepancy in supply and demand, driven principally
by the aging population, is generally robust to the sensitivity analysis,
including the likely effects of managed care.
Our results suggest that the proportion of care provided by specialists
in critical care and pulmonary medicine cannot be maintained in the United
States for more than 10 years. The principal reason is the increased demand
for care as the population ages. We found variation in current practice patterns,
including differences associated with health care reform initiatives. However,
our sensitivity analyses suggest these initiatives, and other potential changes
in supply and demand, will not prevent and may even promote the shortfall.
Reports on Medicare have predicted increased costs and demand for services
as the US population ages. Lacking from these reports, however, are details
on how to deliver these services.2,31
Simultaneously, a common view has emerged that graduate medical training should
remain constant or decrease.2 This has created
a public policy paradox. Despite widespread recognition that elderly health
care needs will grow, there is no planning for who will actually provide this
This situation is the legacy of a decade-long debate whose premise was
that the US had an oversupply of physicians in general and of specialists
in particular.3,32 In 1994, the
Council on Graduate Medical Education predicted an increase in demand for
generalist physicians and a decrease in demand for specialists.32
The Council on Graduate Medical Education relied heavily on assumptions of
managed care growth and associated efficiencies. Others, using managed care
use data, reached similar conclusions.5 These
predictions have proven inaccurate as patient resentment of managed care restrictions
and the threat of litigation have prompted managed care companies to liberalize
access to specialty care.
Our study differs substantially from most previous workforce models.
First, our forecast horizon extends beyond 2010, thereby capturing the major
demographic shift toward the elderly. Second, as discussed above, previous
studies relied on the flawed assumption 11
that managed care would decrease demand. Our study suggests that, at least
for intensivists, managed care may actually increase demand. Third, we treated
demand and supply independently. Some models of physician labor markets assumed
physician supply drives demand.33 However,
as pointed out by Reinhardt,4 this assumption
has proven overly simplistic, failing to account for much of the variation
in physician work patterns. In addition, demand for critical care and pulmonary
services is likely to be less discretionary than that for more procedure-driven
specialty services. Finally, intensivists and pulmonologists provide only
a portion of critical care and pulmonary services. Therefore, while there
may be some coupling of demand to supply, such an effect is likely to be weaker
than for other physician groups.
We found intensivists currently provide one third of ICU care and, without
increased supply, will provide a smaller proportion in the future. It is therefore
key to determine the optimal proportion and types of ICU patients who should
be treated by intensivists. Our study was not designed to answer this question.
However, studies suggest care by intensivists is associated with decreased
resource use and improved patient outcomes.34- 38
Therefore, it is possible that the current proportion of ICU patients managed
by intensivists is already less than ideal. For patients with pulmonary disease,
there are no studies comparing outcomes for those treated by pulmonologists
compared with other physicians.
We explored potential responses to the predicted shortfall of specialist
hours. Specialists could dedicate more hours but this would be insufficient
to reverse the shortfall and would require financial incentives, contradict
trends in specialist work habits,20 and potentially
lead to burnout.39 Health care reform initiatives
are unlikely to affect our predictions. While such initiatives have reduced
the length of hospital stays for many conditions,40
total age-specific use rates have not decreased but actually have grown.41 Furthermore, managed care organizations have shown
little ability to reduce resource use by ICU patients.42
Although technologic advances might theoretically reduce need for specialist
care, the more usual pattern is for technology to increase demand for care.41,43
We found no epidemiological data to suggest future decline in the incidence
of pulmonary diseases or those disorders requiring critical care. These conditions
are common complications of many illnesses and are driven by patient demographics.
Greater use by the elderly of advanced directives to limit intensive care
at the end of life could be encouraged. However, the consequences in our sensitivity
analysis were small and other studies suggest that the elderly recover well
from critical illness.44- 46
Thus, if tacit rationing of health care is to be minimized, we must explore
mechanisms to ensure that qualified health care clinicians are available to
care for the growing burden of critical care and pulmonary disease.
The implications of our study outside the United States are less clear.
There is wide international variation in intensivist training and accreditation47 and in the availability of ICU beds.48
In some countries, including Australia, New Zealand, and several western European
nations, virtually all ICU care is delivered by intensivists while in others
there is little formal critical care training.49
There is little information on international variation in supply and demand
for pulmonary services, and we are not aware of forecast models from other
countries. However, many nations have aging populations and increasing health
care costs and may face similar increases in demand for care. Given that the
increased demand of an aging population was the strongest driving force of
our forecast model, our findings may generalize to countries with similar
demography, such as many industrialized nations. Indeed, a recent study from
the United Kingdom demonstrated that demand for ICU beds has increased in
recent years, leading to premature ICU discharges and increased hospital mortality.50
Our study has several potential limitations. The American Medical Association
Physician Masterfile, while commonly used as a source for the US physician
workforce, may not capture all practicing physicians. In some specialties,
such as surgery, the number we identified is less than the number who have
graduated during the past 10 years. However, our forecast model suggests the
number of new graduates, based on well-documented training program records,
maintains supply at a near-constant level. If the existing workforce were
larger than our estimate, then the projected number of new graduates would
not maintain current supply, worsening the projected shortfall.
The use of administrative patient-level data also has potential inaccuracies.
We used these data principally to understand the distribution of resource
use by disease and by age. Sensitivity analyses suggested our results were
generally robust to variation in our initial estimates. In addition, because
information on ICU resource use for adults younger than 65 years came from
only 1 state, we compared overall age-specific ICU use estimates with 2 other
states (data not shown) and found differences of less than 10%. Some ranges
in our sensitivity analysis might be considered too low. For example, the
effects of managed care that we measured were average effects and particular
health care systems may use intensivists in ways more extreme than we found.
If such models were to proliferate disproportionately, the managed care effects
could be larger. Our model also relies on data from 2 surveys. Although we
found minimal differences between respondents and nonrespondents, survey results
could still be inaccurate. Indeed, there is potential imprecision and error
in all the estimates incorporated in our model. However, we believe our sensitivity
analyses suggest our results are generally robust to such errors.
In conclusion, we predict the aging of the US population will create
a demand for care that will outpace the future supply of intensivists and
pulmonologists. The resulting shortfall is troubling both because the current
provision of intensivist care is arguably already low and because the aging
population will likely create similar shortfalls in other areas of medicine.
These observations contradict the common perception that specialists will
be oversupplied in the future and raise concerns regarding the provision of
other components of care, such as the supply of other health care workers
and of adequate health care facilities. Because potential solutions, such
as changes in training and education or development of alternative delivery
models, require significant lead time to affect care, we believe prompt attention
is necessary. Otherwise, the growing disease burden in the United States will
stress an unprepared health system, and, in particular, disadvantage the elderly.