About the Project

The FutureDocs Forecasting Tool estimates the supply of physicians, use of healthcare services, and capacity of physician supply to meet the health care services use for the United States population. The tool is designed to engage a wide range of stakeholders including physicians, physician organizations, policymakers, health system executives and other interested parties in developing workable and practical solutions to address imbalances in the supply and distribution of physicians.

The tool is an important and innovative step forward for health care workforce modeling because it is interactive, web-based, and user-friendly. FutureDocs Forecasting Tool gives users the ability to display different estimates of supply for various specialties, healthcare services use, and shortages or surpluses for many types of services at different geographic levels between 2013 and 2030. The tool also provides the option to implement different scenarios, such as the Affordable Care Act's insurance exchanges and Medicaid expansion provisions; adjust retirement rates and work effort by physicians; and increase the use of Nurse Practitioners and Physician Assistants to meet the demand for health care services.

The statistical model that underlies the tool is innovative because it does not produce "silo-based" projections for each physician specialty. Prior healthcare workforce models generally have assumed that all physician specialties provide a uniform set of services completely different from the set of services of another specialty. This model, however, reflects the real-world nature of clinical practice, in which physicians in different specialties have overlapping scopes of services. The FutureDocs Forecasting Tool represents the first implementation of the "plasticity" methodology that recognizes this overlap. Plasticity is the idea that there are multiple configurations of physicians able to meet a community's use of health care services. For a detailed explanation of plasticity, please see the question "How do physician supply and healthcare services use meet through plasticity?" or the Academic Medicine paper in Additional Resources.

The developers of the model and forecasting tool at the Program on Health Workforce Research and Policy, of the Cecil G. Sheps Center for Health Services Research at The University of North Carolina-Chapel Hill, intend to update the model to incorporate new data sources and to reflect changing population demographics, physician practice patterns, and health care policy. We are indebted to The Physicians Foundation, which funded the development of the FutureDocs Forecasting Tool.

A downloadable version of this page is available here.

How can I use this model?

The FutureDocs Forecasting Tool is intended for use by a wide variety of stakeholders interested in health care workforce policy and planning. Examples of the types of individuals and groups for whom the tool will be useful and examples of the ways estimates from the models can be used include:    

  • Physician-practice or human-resource managers: Physician practice managers and hospital and health-system executives can use the model for workforce planning for their practices, hospitals, or health systems.
  • Professional groups and private foundations: The model provides information that state medical societies, state and national specialty societies and associations, and other groups representing physicians and other clinicians can use to educate and engage leaders and the general public on healthcare workforce issues and policy interventions needed to address those issues.
  • Local and state policymakers: Policymakers can use the model to inform local- and state-level workforce policies including decisions about whether to develop new residency positions or invest in loan repayment programs to attract more physicians to underserved areas.
  • Federal health planners or policy makers: The model will be useful to national stakeholders who need better and more timely information about physician workforce supply, healthcare services use, and the capacity of physician supply to meet demand for healthcare services. Estimates from this model can help guide federal investments in Graduate Medical Education as well as healthcare payment policy, health care workforce shortage area designation criteria, and other national policy issues.

As the figure below illustrates, "the model" is actually made up of multiple models. One model estimates the future supply of physicians and a second model projects the future use of healthcare services. The estimates produced by each of these two models are combined to produce a third model which we term: "shortage/surplus" or "relative capacity". This third model reflects the capacity of the physician workforce to meet the demand for health care services for a particular type of health care service in different health care settings at different levels of geography.

The healthcare services use "sub-model" produces estimates of population-level healthcare services based on population characteristics, such as age and gender, and prevalence in the population of health risk characteristics, such as obesity or smoking. The model measures healthcare services use in terms of visits. It estimates physician visits at the state and sub-state levels across 19 different types of conditions and within three healthcare settings: the inpatient setting, the outpatient setting, and the emergency department. For more information on how we model healthcare services use, see the question "How do we model use of healthcare services?".

The physician supply sub-model produces estimates of physician supply up to 2030 using information on:    

  • physician demographics, specialty, geographic distribution and full-time equivalents (FTEs) using 2013 as the baseline;
  • resident-in-training demographics, specialty choice, historical patterns of branching and switching between specialties, and geographic location;
  • historical patterns of physician diffusion; and
  • mortality and retirement rates by age and sex.

The supply model produces estimates of physician supply in both headcounts and patient care FTEs by age, gender, specialty, and geography. For more information on how we model physician supply, see the question "How do we model physician supply?".

Estimates from the supply and healthcare services use models are compared through a mechanism known as "plasticity". Plasticity allows the model to convert physician FTEs by specialty into visits. This allows the model to use "visits supplied" from the supply sub-model to "visits demanded" from the healthcare services use sub-model to develop a measure of "relative capacity" for visits by clinical condition, health care setting, and geographic region for all years between 2013 and 2030. "Relative capacity" is the model’s way of describing whether or not a geographic area faces a shortage or surplus of physician visits by clinical condition and healthcare setting.

For more information on plasticity, see the question: "How do physician supply and healthcare services use meet through plasticity?" or to the Academic Medicine paper by Holmes and colleagues in "Additional Resources". For more information on relative capacity, see the question "What is relative capacity, and how is it included in the model?"


The figure below describes how the FutureDocs Forecasting Tool functions. The website was built using Bootstrap, an HTML platform. The FutureDocs Forecasting Tool visualizations are built using "D3 external website", an open source JavaScript library that displays the data, maps, and charts in the model.

The algorithms for estimating future physician supply are run on Heroku external website, a cloud-based server platform. The projections produced by the physician supply model are then transferred to an aggregated data library. The aggregated data library also stores the healthcare services use estimates.

The aggregated data library interacts with the website through an application programming interface (API). Thus, when a user makes a selection on the website, the API communicates with the aggregated data library to obtain the appropriate model estimates.

The different parts of the model’s web-based interface

The FutureDocs Forecasting Tool features conceptual, methodological, process, and usability innovations that that make it distinct from other healthcare workforce models.

Conceptual: The model includes the conceptual innovation of "plasticity" — that physicians provide overlapping scopes of services, and therefore multiple configurations of physicians can address a community’s healthcare services use. This represents a departure from past healthcare workforce modeling efforts which generally assumed that each physician specialty provides a uniform set of services completely distinct from the services provided by other specialties. For a more detailed explanation of plasticity, please see the question "How do physician supply and healthcare services use meet through plasticity?" or the Academic Medicine paper in Additional Resources.

Methodological: To better reflect decisions that physicians make about where and how much to practice, the supply side model utilizes an agent-based modeling approach rather than the stock-and-flow approach used in many previous physician supply forecasts. In a stock-and-flow model, physicians are segmented into different groups, and each physician within that group behaves identically. For instance, a stock-and-flow model might assume that all primary care physicians older than 65 years of age will retire. In contrast, an agent-based modeling approach estimates the decisions of individual physicians as a factor of multiple characteristics. For example, there is a non-zero probability of retirement at any point in a physician’s career as a result of attributes such as the physician’s age, gender and specialty.

Process: A key guiding principle has been that the model should reflect "real-world" clinical practice. To accomplish this objective, the project team sought input from actively practicing clinicians to check the "face validity" of the model’s assumptions and estimates.

Usability: The FutureDocs Forecasting Tool is intended to be user-friendly and interactive. Users can customize healthcare workforce projections by choosing different model scenarios. These scenarios anticipate possible changes in the factors that affect the supply of physicians and utilization of health care services. Users can also change the geographic level at which model projections are displayed. The ability to customize displays allows the user to generate multiple estimates under different assumptions and geographies, rather than producing a single "right answer."

The Web-based interface also provides users with multiple options for visualizing the model projections: line charts, population pyramids, or maps. Users can download model data and images. Furthermore, detailed information about the model’s methodology has been posted online, so users can gain a detailed understanding of how the model functions.

The project team has tried to be as transparent as possible by describing the model’s methods and assumptions and by providing detailed resources in the form of frequently asked questions below, and a set of additional resources on the model website. Although the code for generating supply, healthcare services use and relative capacity estimates is not available without permission, model users can download supply, use, and capacity estimates. Furthermore, users can download the code used to generate the model visualizations.

What if I want to know more?

Unlike previous workforce projection models, we wanted users to be able to examine workforce trends at a smaller unit of geography than a state. To achieve this, we created "tertiary service areas" which are based on the Dartmouth Institute for Health Care Policy and Clinical Practice’s hospital referral regions external website. Hospital referral regions (HRRs) set geographic boundaries for markets for specialized care. To develop the hospital referral regions, Dartmouth examined patterns in referrals to major medical centers for complex, invasive procedures.

The building blocks of Dartmouth’s HRRs are hospital service areas (HSAs), which are built using ZIP code areas. HSAs often cross state and county boundaries. Various components of our model use county- and state-level data, so it was not feasible to produce forecasts at the hospital referral region-level.

To address this issue, we created a modified version of Dartmouth’s HRRs that is built on groups of counties rather than ZIP codes, and named them tertiary service areas (TSAs). There are 293 TSAs in the U.S. Approximately 50% of TSAs are made up of eight or fewer counties, and 32 TSAs include only a single county. On the other hand, the Wichita, KS, TSA includes 68 counties.

The average tertiary service area had a population of 1 million in 2010. Tertiary service areas for Detroit, MI; Portland, ME; San Francisco, CA; and El Paso, TX contained approximately 1 million people each in 2010. In 2010, the Ormond Beach, FL, tertiary service area had the smallest TSA by population, containing 90,000 people. The Los Angeles, CA, tertiary service area, with 9 million people, was the most populated tertiary service area.

The model estimates physician supply by physician specialty, by state or tertiary service area (TSA), and by year. Users can view physician supply estimates in terms of numbers of physicians ("headcount") or patient care full-time equivalents (FTEs).

The model uses an agent-based approach to estimating future physician supply. This means that all physicians involved in patient care —and their career decisions from training through retirement— are estimated individually in the model.

A physician "agent" can enter the model in one of two ways. First, all physicians active in patient care in the U.S. in 2013 are included in the model. Data on the demographic characteristics, specialty, and geographic distribution of these physicians were developed based on algorithms that combined data from the American Medical Association’s Masterfile and American Board of Medical Specialties’ certifications. These algorithms produced a baseline estimate of physician supply in 2013 by age, sex, specialty and geographic location.

Second, a physician can enter the supply model through the model’s Graduate Medical Education pipeline. We used data from the AAMC’s GME Track to estimate the demographic characteristics, location, specialty, length of training and probability of branching/switching specialties while in training of each new medical resident in the U.S.

In each year after a physician "enters" the model, the model updates 1) the number of patient care FTEs that physician will provide in the year, and 2) that physician’s geographic location. (For more information on physician migration in the model, see "How does the model diffuse physicians to different geographies?".)

In the supply model, a physician can work 0 patient care FTEs, 0.5 patient care FTEs, or 1.0 patient care FTE. The probability that a physician works 0 patient care FTEs, 0.5 patient care FTEs or 1 patient care FTE is based on a physician’s age, gender, and specialty. These probabilities were calculated using on physician work hours from the North Carolina Health Professions Data System. (For more information on patient care FTEs, see "Why measure headcount and patient care FTE?")

The model assumes certain limits to the supply based on agent characteristics. For example, when a physician in the model is assigned 0 patient care FTEs in a given year, that physician is considered to have left the patient care workforce. The model also assumes that all physicians retire before reaching 80 years of age. The model also incorporates attrition from the workforce due to physician mortality. To estimate the mortality rates of clinically active physicians, the model uses population mortality rates by age and gender from the Centers for Disease Control and Prevention.

The model accounts not only for how much a physician works but also where a physician works. In each year in which a physician is considered clinically active in the model, the model calculates the probability that that individual physician will move. In this way, the model estimates patterns of physician migration between 2013 and 2030. (For more information on how the model incorporates physician migration, see "How does the model diffuse physicians to different geographies?" below.)

The model estimates individual physicians’ career decisions for the entire physician workforce between 2013 and 2030. The model then aggregates the effects of the decisions to develop overall estimates of physician supply in terms of both headcount and patient care FTE and by specialty, age, and gender.

To establish the baseline number of physicians in 2013, data from the American Medical Association’s Physician Masterfile (MF) were used in conjunction with certification data from the American Board of Medical Specialties (ABMS). The model includes all physicians listed as actively working according to the AMA Masterfile, whether it is work in direct patient care, administration, medical research, or teaching. Physicians classified as retired, semi-retired or not active were excluded. While previous studies have shown that most physicians begin to reduce their hours or retire after 65, we chose to use a higher retirement age of 80 to capture physicians who are actively teaching or in an administrative position because they have an important role in the education of future physicians. Physicians whose major professional activity was unclassified in the Masterfile were included in the model. Physicians in training—specifically residents and fellows—were excluded from the baseline workforce data, but are included in the model in the graduate medical education (GME) pipeline. Federal physicians were included since they often provide health care services to civilians during and after retiring from federal service.

To simplify the model and its predictions, we collapsed the 315 specialties identified by the American Medical Association (AMA) into 35 categories:      

  1. We grouped specialties with common training pathways into a single category. For example, all surgical specialties for which individuals first train in general surgery residencies are grouped with the "surgery" category. The exception to this rule is internal medicine sub-specialties, which are each placed within their own categories for modeling purposes (e.g., endocrinology, cardiology, nephrology, etc.).

  2. American Board of Medical Specialties (ABMS)-recognized "certificates of added qualifications" such as sleep medicine, hospice and palliative medicine, and sports medicine that have multiple entry routes are grouped back into their stem training pathway and primary board. For example, the AMA code "sleep medicine-IM" is grouped in the "internal medicine" specialty category and "sleep medicine-pediatrics" is grouped in the "pediatrics" category.

  3. While we have sought to apply consistent decision rules when assigning specialties into categories, the many unique situations for some specialties required some flexibility, including:
    • A few specialties important for policy reasons received their own category, such as geriatrics.
    • Category assignments for hybrid and joint specialties signified by a dash or slash within the AMA’s listing were made on a case-by-case basis to best reflect the nature of the work their practitioners typically perform. For example, "Emergency Medicine/Family Medicine" is mapped to the Emergency Medicine category but "Family Medicine/Preventive Medicine" is mapped to "Preventive Medicine."

  4. Pediatric specialties like pediatric cardiology and pediatric endocrinology in which individuals must first complete a general pediatric residency were grouped within a "pediatric non-surgical specialties" category. Individuals in other pediatric specialties can follow several training pathways and often first complete an adult training program in their area, such as pediatric psychiatry and pediatric emergency medicine. These specialties are mapped to their adult specialty counterpart. Exceptions include pediatric surgery specialties that train from surgery, which are mapped to pediatric surgical specialties.

  5. If a specialty includes a second (root) specialty in parentheses, then the specialty in parentheses becomes the assigned specialty category. For example, "Endovascular Surgical Neuroradiology (Neurological Surgery)" is mapped to Neurological Surgery.

  6. Hospitalists are grouped with internal medicine even though other specialties also serve as hospitalists, particularly family medicine. This means that hospitalists are grouped under the "Primary Care" category although they provide only inpatient services.

Using a combination of specialty data from the AMA Masterfile and certification data from the American Board of Medical Specialties (ABMS), we assigned physicians to 35 specialty categories using 5 sequential decision rules:

      Rule 1: If no ABMS certifications identified, use AMA primary specialty.
      Rule 2: If primary specialty in MF matches first ABMS certificate, this is physician specialty.
      Rule 3: If primary specialty in MF does not match the first ABMS certificate and no other ABMS certificate is identified, use first ABMS certificate.
      Rule 4: If primary specialty in MF does not match the first ABMS certificate, but matches the second ABMS certificate, use the second ABMS certificate
      Rule 5: If rules 1 to 4 have not been applied, use first ABMS certificate.
For a more detailed discussion of the methodology and the list of specialties that were captured in each of the 35 categories, please see Additional Resources.

One of the most significant drivers of "effective workforce supply" is the full-time equivalent (FTE) number of hours physicians spend seeing patients. The amount of time that physicians spend on non-clinical responsibilities including paper work, implementing electronic health records and attending to activities other than seeing patients can be significant. This means that the supply of physicians needed to meet the population’s use of health care services in a given geographic area often requires more physicians than suggested by a simple headcount: each physician spends part of his or her time doing non-clinical work.

In describing FTEs, the Association of American Medical Colleges (AAMC) uses this definition: "FTEs, or Full Time Equivalents, represent the number of physicians if every physician worked as many hours as the average physician worked in the baseline year. external website" The approach taken in our model departs from the AAMC’s approach in that we do not use an "average" number of hours. Instead, we allow FTEs to vary by specialty, age and gender in every year, including the baseline or first year of the model. Also different from past models is that our FTE calculations are based on patient care hours, not the total hours worked.

Hours worked in patient care are not available from the American Medical Association’s Masterfile (AMA MF), and robust data on patient care hours by sex, age and specialty are not readily available from other national data sets. For this reason, we used data on hours spent seeing patients from the North Carolina Health Professions Data System (HPDS). The HPDS is a longstanding and well-respected source of workforce data collected from initial licensure forms and annual licensure renewal data. The structure of the North Carolina physician workforce roughly matches national distribution in specialty, age and gender distributions with 31.0% of North Carolina physicians in primary care versus 34% nationally, 30.0% female in both North Carolina and nationally and 44.0% older than 50 years in North Carolina and 49% nationally.

Based on data from the HPDS, we determined that 1 patient care FTE (PCFTE) should equal the number of hours spent seeing patients by the 95th percentile of the physicians in each specialty. The next step was to assign a PCFTE to each individual physician in the model. The simple solution would have been to assign each physician’s patient care FTE based according to the average for their specialty, adjusted by age and gender. Using this approach, a 50-year old male working in a 60 hour/week specialty (e.g. Rheumatology) would be assigned (according to data from the NC HPDS) a PCFTE of 0.72, and a 65 year old female a PCFTE of .56. However, because our model first calculates the number of visits required by each clinical service area and then searches for the appropriate number of PCFTEs by specialty to provide this care, this approach requires extensive calculation and iteration to determine the optimal and precise match of PCFTE to meet the utilization required. The model simplifies the approach by assigning the PCFTE of physicians as either a 1.0, 0.5 or 0 FTE. Using just 3 categories of PCFTE reduces the calculations required yet produces an aggregate number of modeled hours very close to the actual number of hours worked in patient care. For more information about how physicians were assigned a PCFTE of 0, 0.5 or 1.0 PCFTE please see information in Additional Resources.

Modeling physician supply requires understanding the number of medical residents entering training by age, gender, and location. Modeling physician training pathways also requires determining the rate of attrition from training programs, the number of empty residency positions "backfilled" with preliminary-year residents, the number of years required to complete a residency in each specialty, and the number of residents who pursue switch training pathways or go on to subspecialty training in fellowships. We obtained this information from the AAMC’s GME Track, a census of medical residents and residency programs.

Information on residency length by specialty, age, gender, and location of residents—as well as the number of residents who leave residency programs in a given year—allowed the model to estimate the number, distribution, and demographic composition of residents entering the physician workforce.

The model recognizes that many residency graduates, particularly those who have completed internal medicine or general surgery, go onto to further training in order to subspecialize. For example, rather than becoming general internists, many internal medicine residency graduates move into fellowships and eventually practice in specialties like cardiology, rheumatology, and oncology. Therefore, the model uses data from GME Track to estimate the number of internal medicine, pediatrics, and general surgery residency graduates that branch into other specialties. It also allows a non-significant number of residents to switch training pathways based on estimates of these switching behaviors from the GME Track data.

"Physician diffusion" describes the migration of physicians over time within the U.S. Understanding physician diffusion is crucial to estimating future physician supply at the state and tertiary service area levels (TSA).

Whether or not a physician in the model moves to a new location in a given year is based first on historical patterns of physician diffusion, and second on relative capacity in each TSA.

If there is no change in the capacity of physicians in a TSA to meet the demand for health care services in that area, an individual physician’s probability of moving to a particular state is equal to the historical probability of moving to that state. Within a given state, a physician’s probability of moving to a particular TSA within that state is proportional to that TSA’s population relative to the population of the entire state. We calculated historical state-to-state patterns of physician diffusion using data from the 2006 and 2011 American Medical Association Masterfiles. (For additional information on state-to-state patterns of physician diffusion, see the link to Tom Ricketts’ Academic Medicine article in Additional Resources.)

This model is unique because it diffuses physicians not only on historical "push" factors but also on "pull" factors that arise when the demand for physician services exceeds the capacity of physicians in that TSA to provide those services . For example, if there is limited physician capacity in a TSA—that is, visit use exceeds visit supply—then the probability that physicians migrate to that TSA increases above historical rates. On the other hand, if there is excess relative capacity in a TSA—visit supply exceeds visit use—physicians migrate out of that TSA at levels above historical rates. The size of the change in relative capacity in a TSA affects how much physician migration in or out of a TSA differs from historical rates.

We forecast the number of physician visits between 2013 and 2030 in 19 clinical service areas. In our model, a visit is the operative unit of healthcare services use. We model utilization of healthcare services at both the state level and the sub-state, or tertiary service area, level. (For a description of tertiary service areas, see the "What is a Tertiary Service Area?" question.) We model healthcare services use for three health care settings: the outpatient setting, which includes physician offices and hospital outpatient departments; the hospital inpatient setting; and the emergency room.

We used a national sample of patients contained in the 2008 and 2009 Medical Expenditure Panel Survey (MEPS) external website to examine use of healthcare services by clinical service area and setting by age, sex, race, smoking, obesity, diabetes, poverty, health insurance coverage status (uninsured or not), rurality, and region of the country. We employed these estimates to calculate each county’s expected use of healthcare services based on numbers of people in a county and their demographic and health risk characteristics. For example, if the MEPS data indicate that white women over 85 years old with a specific set of health risk characteristics use three outpatient circulatory visits and one inpatient respiratory visit on average per year, a county that contains 100 women that fit this risk profile will "contribute" 300 outpatient circulatory visits and 100 inpatient respiratory visits to that county’s total utilization of visits.

It is important to note that these baseline projections are estimated using local population characteristics and do not represent actual area-level observed use of healthcare services. This means that a population’s estimated use of healthcare services in a county will deviate from the national average due to 1) prevalence of risk factors in an area and 2) the size of the population with those risk factors in that area. The relative importance of factors vary by type of health care service used; for example, the smoking rate has double the effect on the visit rate for cancer than on the visit rate for circulatory visits; the smoking rate’s effect on injuries is about half that of its effect on circulatory.

We use population estimates to calculate the population’s future use of healthcare services based on the age, sex and race of the population. The estimates come from ProximityOne external website, a private vendor that produces the estimates based on data from the Census, American Community Survey, and other data sources. These estimates do not include the effect of change in age-race-gender specific health risks like obesity, smoking, poverty or uninsurance. The model assumes that the age-race-gender prevalence of these factors do not change in the future and therefore any changes in the incidence of diabetes, for example, is driven by the aging of the population. We calculate future utilization estimates for every five year interval, rather than every year. In the intervening years, we assume a "straight line" trend in healthcare services use rates.

Although we calculate physician visit use at the county level, we sum county-level visits by CSA up to the tertiary service area level. The website displays per-population physician visit use at the tertiary service area level, calculated as the number of TSA visits divided by the TSA population. This avoids "scaling" issues due to the wide variation in population at the TSA level. We do not present county-specific estimates because some counties have very small populations and could produce relative high error rates.

The model’s 19 clinical service areas (CSAs) describe the primary medical condition associated with a patient’s visit to a physician (i.e. their primary complaint). For example, a patient visit for hypertension would be classified under the circulatory CSA, as would a patient visit for acute myocardial infarction. The table below lists the 19 clinical service areas included in the model as well as examples of conditions for each clinical service area. The CSAs are based on the Agency for Healthcare Research and Quality’s 18 conditions in its Clinical Classification Software external website but we added an additional category of preventive care. For a full list of the conditions under each of the CSAs please go to Additional Resources.     

List of Clinical Service Areas (CSAs)
Clinical service area label Full name of clinical service area Examples of diagnoses in each clinical service area
Blood Diseases of the blood and blood-forming organs Sickle cell anemia
Circulatory Diseases of the circulatory system Essential hypertension
Acute myocardial infarction
Congenital Anomalies Congenital anomalies Pulmonary artery anomalies
Polycystic kidney
Digestive Diseases of the digestive system Appendicitis and other appendiceal conditions
Liver disease; alcohol-related
Endocrine/Immunity Endocrine; nutritional; and metabolic diseases and immunity disorders Diabetes mellitus without complication
Gout and other crystal arthropathies
Genitourinary Diseases of the genitourinary system Acute and unspecified renal failure
Chronic kidney disease
Infectious Infectious and parasitic diseases Tuberculosis
E. Coli septicemia
Injury Injury and poisoning Fracture of upper limb
Complication of device; implant or graft
Mental Mental illness Attention deficit disorder and Attention deficit hyperactivity disorder
Depressive disorders
Musculoskeletal Diseases of the musculoskeletal system and connective tissue Osteoarthritis
Systemic lupus erythematosus
and connective tissue disorders
Neoplasms Neoplasms Breast cancer
Nervous System Diseases of the nervous system and sense organs Parkinson’s disease
Perinatal Certain conditions originating in the perinatal period Respiratory distress syndrome
Hemolytic jaundice and perinatal jaundice
Pregnancy/Childbirth Complications of pregnancy; childbirth; and the puerperium Ectopic pregnancy
Early or threatened labor
Preventive Care Preventive care Contraceptive management
Screening for viral diseases
Residual Codes Residual codes; unclassified; all E codes E codes include external causes of injury, such as traffic accidents or falls
Other codes in this category: findings without conclusive diagnosis
Respiratory Diseases of the respiratory system Influenza
Acute bronchitis
Skin Diseases of the skin and subcutaneous tissue Chronic ulcer of leg or foot
Skin and subcutaneous tissue infections
Symptoms & Signs Symptoms; signs; and ill-defined conditions and factors influencing health status Syncope
Allergic reactions

In this model, "plasticity" serves as the mechanism by which patient visits are mapped to physician supply to determine whether supply will be sufficient to meet local healthcare services use. The term plasticity describes the idea that there are multiple configurations of healthcare providers able to meet a community’s use of healthcare services because physicians in different specialties have overlapping scopes of service provision. For example, multiple physician specialties—such as cardiologists, family physicians, and internists—manage circulatory conditions. This model reflects that in real world of clinical practice, different physician specialties provide overlapping sets of clinical services. (To read more about plasticity, see the Academic Medicine article on plasticity in Additional Resources).

We implement plasticity—and thus compare healthcare services use and physician supply estimates—in the model using a plasticity matrix. The matrix represents a national average allocation of patient visits to each physician specialty by clinical service area and by health care setting. A simplified and example only matrix is shown below.


The matrix can be used to examine each specialty’s scope of service provision. For example, the last row of matrix shows that 4% of internal medicine visits are for neoplasms (cancer), 54% are for circulatory conditions, 40% are for respiratory conditions, and 3% are for pregnancy/childbirth. Similarly, the matrix can also be used to understand how visits for clinical service areas—such as circulatory conditions or injury—are divided across specialties. For instance, the second column of the matrix shows that 34% of all circulatory visits to cardiologists, none are to dermatologists, 38% are to family physicians, and 1% are to obstetrician/gynecologists.

We assume that the distribution of conditions that a specific specialty sees is identical in both inpatient and outpatient settings. For instance, the model assumes that both 54% of outpatient visits and 54% of inpatient visits to internists are for circulatory conditions.

The model treats emergency department visits differently than it treats inpatient and outpatient visits. We do not categorize emergency department visits by clinical service area. Rather, the plasticity matrix "allows" ED visits to be seen by a select set of physician specialties that tend to work in emergency departments.

We populated the matrix using a combination of data from the 2008-2009 Medical Expenditure Panel Survey, surveys to practicing specialists asking about their scope of service provision and expert opinion from physicians.

The first step in using the plasticity matrix to compare estimated physician supply and healthcare services use is to establish the number of total visits that physicians can see in a year. Using data from the Medical Group Management Association (MGMA), we determined that inpatient visits substitute for outpatient visits at a rate of approximately 1 inpatient visit to 5 outpatient visits. In other words, the more time a physician spends in the inpatient setting, the fewer patients he or she will be able to see in a year.

The second step in the process of reconciling estimated supply and healthcare services use is to allocate estimated visits by clinical service area to physicians in each specialty within a tertiary service area. These visits are allocated to physicians according to the national average distribution of visits by specialty in the plasticity matrix. For example, if internists nationally spend 54% of their visits addressing circulatory conditions, the model assumes that 54% of internist visits in a particular tertiary service area will be for circulatory conditions.

After visits have been assigned to physicians in a given area, the model can then compare supply of visits for a particular clinical service area to the predicted use of care in that clinical service area.

The term "relative capacity" describes how the model measures shortages or surpluses in physician supply. In this model, relative capacity is the ratio of number of visits demanded in a particular geographic area divided by the number of visits supplied in that geographic area. If relative capacity is less than 0.85 —that is, the ratio of the number of visits supplied to the number of visits demanded is less than 0.85— then there is "excess demand" for visits in area that cannot be met by local physicians. If relative capacity is between 0.85 and 1.15, the model considers visit demand and visit supply to be in balance. If relative capacity is greater than 1.15, then there is an excess supply of visits—there are more physicians supplying visits than there are individuals using those visits.

Model users can assess relative capacity at both the state and tertiary service area levels and by health care setting and clinical service area.

In other healthcare workforce models, shortages or surpluses tend to be measured in terms of physician headcount or FTE by specialty. However, we chose to measure shortages and surpluses by the ability of different workforce configurations in different geographic areas to meet the demand for different types of health care services. Therefore, our shortage/surplus estimates are calculated in terms of visits, not numbers or FTEs of specific physician specialties.

The model projects physician supply, health care service use, and relative capacity under multiple scenarios. The scenarios were chosen based on potential and likely changes that might occur to factors influencing the supply of physicians and the demand for health care services. Examples of scenarios in the model include implementation of the Patient Protection and Affordable Care Act’s coverage expansion provisions, changes in the distribution of Graduate Medical Education slots, and changes to physician retirement rates. Providing users with a variety of scenarios—as well as options for adjusting the scenarios—allows users to customize the model’s output to suit their beliefs about what the future might look like and to develop interventions to address potential shortages and imbalances. For information on how to implement model scenarios in the web-based tool, please see the Help page.

  1. Baseline: The baseline scenario is the model’s default scenario. This scenario assumes that there are no changes to factors influencing healthcare workforce supply and healthcare services use between 2013 and 2030. It is important to note that the baseline scenario begins in 2013 and assumes that the Patient Protection and Affordable Care Act is not implemented.

  2. Retirement scenarios: These scenarios allow users to understand how retirement rates affect physician supply. There are two options for this scenario: a low retirement rate and a high retirement rate. The baseline retirement rate is the average of the high retirement rate and the low retirement rate.

    The low retirement rate is derived from the US Centers for Disease Control and Prevention’s mortality rates by age and gender. In other words, this scenario assumes that all physicians either work until they die or work until they are 80 years old. The high retirement rates are derived from analyses that the Lewin Group and Thomas R. Konrad of the Sheps Center conducted using Association of American Medical College survey data from 2006. While the baseline and low retirement rates vary by both age and gender, the high retirement rates vary only by age. In other words, in the high retirement scenario, the retirement rates of female and male physicians are identical.

  3. Full-time equivalent (FTE) scenarios: The model includes scenario options that allow users to understand how increases and decreases in physician work effort—measured in terms of patient care FTEs—affect physician supply and physician shortages and surpluses. One scenario option allows users to reduce patient care FTEs supplied by all clinically active physicians by 5% beginning in 2013. In this scenario, FTEs are reduced equally across all specialties and between both genders. Similarly, users are able to increase the patient care FTEs provided by clinically active physicians by 5%.

  4. Increased use of nurse practitioners (NPs) and physician assistants (PAs): This scenario estimates how nurse practitioners (NPs) and physician assistants (PAs) will contribute to meeting future healthcare services use. The scenario uses data from the Medical Expenditure Panel Survey (MEPS), the Area Health Resource File (AHRF), and the North Carolina Health Professions Data System (HPDS). To implement this scenario, we updated the model’s plasticity matrix to include two additional "specialty" rows—one for NPs and one for PAs—and we projected NP and PA supply under two growth rates. We then compared the combined physician and NP and PA supply projection with healthcare services use under the baseline scenario. This allowed us to obtain estimates of surpluses and shortages of visits that incorporate the important and growing role that NPs and PAs play in health care delivery. For a more detailed explanation of this scenario see the Additional Resources section.

  5. Graduate Medical Education (GME) scenario: The question of whether, and in which specialties and geographies, to expand Graduate Medical Education (GME) is a topic of intense policy interest. Our model includes a scenario that redistributes GME slots rather than increases the number of positions. The purpose of the scenario is to redistribute GME from the geographies and specialties in which there is the greatest capacity to states and specialties in which there will not be enough physicians to meet the population’s demand for healthcare services.

    It should be noted that redistributing GME slots away from some states and specialties is not equivalent to asserting that those states or specialties have “too many” clinically active physicians. Rather, this scenario is an attempt to narrow the wide range in relative physician capacity over specialties and states. The outcome of the scenario is to bring those states and specialties with the greatest excess demand (indicated by dark brown on the maps) up to the next level of “lesser demand” (lighter brown on the maps).

    Assumptions and rules: There are a number of assumptions and rules required to implement this scenario. First, we assume that redistribution of GME takes place at the state level since GME is modeled at the state level in the FutureDocs Forecasting Tool.

    The baseline shortage/surplus estimates in 2030 determine how GME slots are redistributed. In this scenario, GME slots are redistributed beginning in 2015. The pattern of GME slot distribution in 2015 is maintained until 2030. For example, if 5 residency slots are redistributed from State A to State B in 2015, then State B will retain those five additional slots from 2015 to 2030.

    We assume that redistributed GME slots are always filled, that states have the capacity to train in the specialties allocated to them and that the redistribution of residency slots does not affect patterns of healthcare service use.

    Finally, we assume that redistributing GME slots affects physician supply only after residents have finished training. On average, redistributing GME slots provides states that receive additional slots with about 10 years of additional physician supply. This is because the residency slots are redistributed in 2015, and the first graduates from the residency slots enter the physician workforce between 2018 and beyond depending on the length of training. Since the model projects physician supply until 2030, a state is allocated approximately 1/10th of the additional GME physician supply they are short per year to mitigate a shortage forecast in 2030.

    Method for reallocating GME: There are several steps required to carry out this scenario. First, in each state, we “assign” visits that are in shortage to the physician specialties that would, according to our plasticity matrix, have seen those visits. We use this assignment methodology to determine which states and which specialties need additional GME slots to address shortages. Second, we again use the plasticity matrix to assign visits that are in surplus to the physician specialties that see these visits. This allows us to determine from which states and specialties GME slots could be withdrawn. Finally, we redistribute GME slots across states.

  6. ACA health insurance scenario: Our model includes a scenario that estimates the effect of the Patient Protection and Affordable Care Act's (ACA) Medicaid expansion and health insurance exchanges on use of healthcare services and on shortages and surpluses. All scenarios in the model assume all states have implemented health insurance exchanges. The default baseline option the model shows reflects Medicaid expansion occurring in the 32 states that have accepted the ACA's Medicaid expansion as of March 14, 2015. There is also the option to select the "Medicaid expansion in all states" scenario to see the impact of all states expanding Medicaid.
  7. For more information about how the ACA scenarios were developed, see the Additional Resources section.

For more information about how GME positions were redistributed between states and specialties see the Additional Resources section.

Additional Resources

Healthcare Cost and Utilization Project. Multi-level CCS Category and Diagnoses. Rockville, MD: Agency for Healthcare Research and Quality. March 8, 2013.

Program on Health Workforce Research and Policy, Cecil G. Sheps Center for Health Services Research. Developing an open-source model for projecting physician shortages in the United States. Chapel Hill, NC: Physicians Foundation; 2012.

Holmes GM, Morrison M, Pathman DE, Fraher E. The contribution of "plasticity" to modeling how a community's need for health care services can be met by different configurations of physicians. [published online ahead of print October 14, 2013]. Acad Med. 2013.

Ricketts T. The migration of physicians and the local supply of practitioners: a five-year comparison. [published online ahead of print October 14, 2013]. Acad Med. 2013.