Breadcrumb
  1. Home
  2. Data & Research
  3. Projecting Health Workforce Supply and Demand
  4. Technical Documentation for HRSA’s Health Workforce Simulation Model
  5. Specialist Physician, Advanced Practice Nurse, and Physician Assistant Model Components

Specialist Physician, Advanced Practice Nurse, and Physician Assistant Model Components

Published 2014

In this module:

This module summarizes the methodology for projecting the national supply and demand for select physician specialties, advanced practice nurses (APNs) and physician assistants (PAs). Projections also were made at the U.S. census division and region levels for detailed specialties. They were made at the state level for specialty categories. Supply and demand were projected for 36 physician specialties, three APN professions (nurse practitioners [NPs], certified nurse midwives [CNMs], and certified registered nurse anesthetists [CRNAs]), and PAs.

Internal Medicine Subspecialty Model

This section describes the supply and demand models of physicians and PAs in 11 internal medicine subspecialties (Exhibit 56). It also includes the supply and demand for physicians and nurse practitioners in critical care medicine.

Estimating the Current Active Workforce Supply

The source for estimating the current active supply of physicians at the U.S., state, and region levels is the 2013 American Medical Association (AMA) Master File Extract. The file is then adjusted for misclassification of older (aged 75 or over) retired physicians as “active”. Base year counts and age sex characteristics for PAs come from the 2013 National Commission on Certification of Physician Assistants (NCCPA) Professional Profile Survey. The counts for NPs in critical care come from NPPES. The age and sex distribution from the ACS is used to assign the age-sex characteristics for NPs.

Modeling New Entrants

Adding new entrants to this workforce is done via the creation of a “synthetic” population. This population is based on the number and characteristics of recent graduates in each internal medicine specialty. As described in Section II.B, each new clinician is assigned an age and sex that reflect the distribution seen in recent years. The primary sources of data on new graduates are:

  • AMA Masterfile for physicians
  • Association of American Medical Colleges (AAMC) 2012-2013 Graduate Medical Education Census completed by residency program directors and administrations
  • American Board of Medical Specialties (ABMS) for physician specialties (Exhibit 57)

Numbers and characteristics of new PAs come from the Physician Assistant Education Association (PAEA) survey and the NCCPA for physician assistants. The number of new NPs in critical care comes from the 2012 American Association of Colleges of Nursing (AACN) survey.

After simulating the age and sex of the new entrants, the region where new providers would practice was simulated. Region is based on a model that regressed the probability of practicing in a region on the relative difference between the projected supply and demand for services in that region.

Modeling Workforce Attrition

As in the case of primary care, the main source of retirement information is the 2012 and 2013 Florida Bi-annual Physician Licensure Survey. Retirement rates differ by medical specialty. Specialties such as allergy & immunology, cardiology, and gastroenterology tend to have later retirements compared to other specialties. Age-sex specific rates calculated from the Florida survey were combining with the age-sex specific mortality rates to derive the overall attrition rate. Exhibit 58 shows that male and female physicians have similar attrition patterns after adjusting for the slightly higher mortality rates among men. Attrition patterns for APNs and PAs were unavailable. As a result, attrition patterns of family physicians were used as proxies.

Modeling Hours Worked

Average hours worked differs by clinician age, sex, and specialty. This impacts the future FTE supply of providers because of the changing demographics of the health workforce. Data for modeling hours worked by physician specialty come from the Florida 2012-2013 Bi-annual Physician Licensure Survey. This is a survey of physicians in Florida who renewed their license. Analysis of Maryland’s physician licensure files found similar work patterns by physician age, sex, and specialty.

We then generated prediction equations for hours worked patterns by physicians in a specialty. Ordinary Least Squares regression was conducted using physicians’ reported average patient care hours per week as the dependent variable. Explanatory variables included indicators (1=yes, 0=no) for specialty, age group, female, and age-by-female interaction terms.

Physicians exhibited hours worked patterns by physician age and sex as illustrated for primary care physicians (Exhibit 15). Young, male physicians tended to work more hours per week than their female counterparts. The sex gap in hours worked largely disappeared after age 55. Hours worked patterns differed by specialty. For example, physicians in nephrology worked 13 hours more per week than dermatologists. Cardiologists work 11 hours more and gastroenterologists 10 hours more per week than dermatologists. We defined 1 FTE physician for each specialty as the average hours worked per week in that specialty.

Using data on PAs working at least 20 hours per week, similar regression analyses were conducted. Data from the 2013 NCCPA license files was used to model hours worked patterns of PAs. Data from the 2012 NSSRN modeled hours worked patterns for critical care NPs. An FTE is the average hours worked per clinician in that occupation and specialty. This is for clinicians working at least 20 hours per week.

Developing Internal Medicine Subspecialties’ Demand Projections

The projected demands for internal medicine physicians and PAs were derived from the common model outlined in Modeling Demand for Health Care Services and Providers. Prediction equations for use of office and outpatient services in medical subspecialties were estimated using Poisson regression with 2008-2012 MEPS data. Separate regressions were estimated for children and adults. The dependent variables were annual office visits and annual outpatient visits for each specialty. Explanatory variables consisted of the patient characteristics, socioeconomic and insurance variables, and health status variables described previously. The number of visits by individuals was aggregated using the sample weights in the population file to project future demand in each state.

Prediction equations for hospitalizations and ED visits used a logistic regression on 2008-2012 MEPS data. Separate regressions were estimated for children and adults and for each of the medical conditions categorized in Exhibit 59. These categories are defined by primary ICD-9 diagnosis or procedure codes. The equations predicted probabilities that each individual would have a hospitalization or ED visit for each of the condition categories. All ED visits were assumed to involve a consultation with an emergency physician.

A single logistic regression estimated using the 2010 NHAMCS modeled the probability that an ED visit required a consulting physician. The dependent variable is whether a second physician is seen during the visit. Explanatory variables consisted of patient demographics and insurance type, and indicators (1=yes, 0=no) for each condition category. We assume that if a visit required a consult, the consulting physician is in the medical specialty associated with the primary diagnosis code as indicated in Exhibit 59.

Predicted probabilities were applied on the simulated microdata set for future years through 2025. This generated projected service use specific to the settings where these providers work.1 Demand for cardiologists, for example, is tied to:

  • projected demand for ambulatory visits to a cardiologist
  • inpatient days where the patient’s primary diagnosis is cardiology related
    • a portion of days will involve hospital rounds
  • emergency department (ED) visits where the patient’s primary diagnosis is cardiology related
    • a portion will involve a cardiologist consult

HWSM uses provider staffing patterns to project demand for physician specialties based on demand for health care services. Staffing patterns start with calculating the portion of national FTE providers delivering care in each setting. This is then divided by current national estimates of the workload driver in that work setting (Exhibit 60). These ratios were then applied to projections of future demand for services for the Baseline demand scenario in HWSM. The Baseline demand scenario assumes the Status Quo in terms of care use and delivery patterns. Estimated FTE requirements to care for each person were then aggregated to obtain the total demand for physicians.

  • 1Due to small sample sizes, HWSM does not model profession-setting combinations where service volume is small (e.g., physicians providing care in home health and residential facilities).

Data from the 2013 NCCPA PA Professional Profile Survey were analyzed to estimates PAs providing care in each major care delivery setting and specialty. The national percentage of FTE PAs in each setting and specialty is divided by national volume of care in that setting. This estimates the portion of an FTE PA per unit of health care service delivered (Exhibit 61). For critical care NPs, a general estimate of staffing for all NPs across all medical specialties is applied. This estimate assumes that NP distribution across settings would reflect the distribution of physicians in all medical specialties by setting.

Regional provider supplies were projected by simulating the locational choice of providers. Any existing shortage/surplus were incorporated in this projection. Hours worked based on provider demographics were also included. Demand estimates were derived by pro-rating the national demand for health care services based on the population characteristics of the regions. These characteristic include age, sex, household income, insurance status, and health status.

Surgical Specialty Model

Practitioners considered in this model include physicians and physician assistants (PAs) that cover 10 surgical specialties

  • general surgery
  • cardiothoracic surgery
  • colon/rectal surgery
  • neurological surgery
  • ophthalmology
  • orthopedic surgery
  • otolaryngology
  • plastic surgery
  • urology
  • vascular surgery

Estimating the Current Active Workforce Supply

The source for estimating the current active supply of physicians is the 2013 American Medical Association (AMA) Master File Extract. The analysis is limited to active physicians. The AMA file is known to misclassify older physicians who have retired as ‘active’. As such, we deleted those over age 75 from the analysis file. Retired physicians between 50 to 75 years of age were deleted based on predicted probabilities derived from a logistic regression on age and specialty. The AMA Masterfile is also adjusted for undercounting hospitalists. A large proportion of hospitalists are listed under the specialty in which they received their training. The base year counts for PAs come from the 2013 National Commission on Certification of Physician Assistants (NCCPA) Professional Profile Survey.

Modeling New Entrants

We add new entrants to this workforce by creating a “synthetic” population based on the number and characteristics of recent graduates in each occupation. As described in section II.B, each new clinician is assigned an age and sex that reflect the distribution seen in recent years.

Estimates of total annual new physicians and PAs and the specialty distribution came from multiple sources. The primary sources of data on characteristics of new graduates are:

  • Association of American Medical Colleges (AAMC) 2012-2013 Graduate Medical Education Census completed by residency program directors and administrators
  • 2013 AMA Masterfile
  • American Board of Medical Specialties (ABMS) for physician specialties.

Numbers and characteristics of new PAs come from the Physician Assistant Education Association (PAEA) and the NCCPA for physician assistants (Exhibit 63).

Modeling Workforce Attrition

Data sources for modeling attrition patterns of physicians by individual specialty are limited. The primary source of retirement information is the 2012 and 2013 Florida Bi-annual Physician Licensure Survey. The survey asks active physicians about their intention to retire in the upcoming five years (Exhibit 58). Age-sex specific rates calculated from the Florida survey were combining with the age-sex specific mortality rates to derive the overall attrition rate. Exhibit 58 shows that male and female physicians have similar attrition patterns after adjusting for the slightly higher mortality rates among men. Retirement rates, however, differ by medical specialty. The attrition pattern for PAs was unavailable. As a result, the attrition pattern of family physicians is used as proxy.

Modeling Hours Worked

Average hours worked differs by clinician age, sex, specialty. This impacts the future FTE supply of providers because of the changing demographics of the health workforce. Data for modeling hours worked by physician specialty comes from the Florida 2012-2013 Bi-annual Physician Licensure Survey. This is a survey of physicians in Florida who renewed their license. Ordinary Least Squares regression was conducted. Physicians’ reported average patient care hours per week was the dependent variable. This OLS regression generated prediction equations for hours worked patterns by physicians. Explanatory variables included specialty indicators (1=yes, 0=no) for age group, female, and age-by-female interaction terms. Hours worked patterns differed by specialty. Relative to family medicine, for example, physicians in neurological surgery and general surgery work 8 and 7 additional patient care hours more per week. Similar regression analysis was conducted using 2013 NCCPA license files to model hours worked patterns of PAs.

Developing Surgical Subspecialties’ Demand Projections

Projected demand for physicians and PAs is derived from the common model outlined in Modeling Demand for Health Care Services and Providers. HWSM uses provider staffing patterns to project demand for physician specialties based on demand for health care services. The consulting physician is in the surgical specialty associated with the primary diagnosis code as indicated in Exhibit 64. Staffing patterns use the portion of national FTE providers delivering care in each setting. This then divided by current national estimates of the workload driver in that work setting. These ratios were then applied to projections of future demand for services. These projections assume the Status Quo in terms of care use and delivery patterns.

For PAs, a process similar to estimating the physician staffing ratio is used to estimate current and project future FTE demand for PAs (Exhibit 66). Data from the 2013 NCCPA PA Professional Profile Survey were analyzed to provide estimates of PAs providing care in each major care delivery setting and specialty.

The regional provider supplies were projected by simulating the locational choice of providers in light of the existing shortage/surplus. Supply estimates also incorporated hours worked based on provider demographics. The demand estimates were derived by pro-rating the national demand for health care services based on the population characteristics of the regions. Characteristics include age, sex, household income, insurance status, and health status.

Other Medical Specialties

This section summarizes the methodology for projecting the national supply and demand for physicians and non-physician providers. Non-physician providers include physician assistants (PAs) and certified registered nurse anesthetists (CRNAs) in Anesthesiology, Emergency Medicine, Neurology and Physical Medicine and Rehabilitation.

Estimating the Current Active Workforce Supply

The primary source for estimates of physicians currently active in the aforementioned specialties is the 2013 American Medical Association (AMA) Master File Extract. The analysis is limited to active physicians under age 75. Physician specialty is identified using the 2013 AMA Masterfile along with the American Board of Medical Specialties (ABMS) file on physician specialties.

Base year counts for CRNAs come from the 2013 National Plan and Provider Enumeration System (NPPES). The age-sex distribution for CRNAs came from the 2013 ACS. The 2013 National Commission on Certification of Physician Assistants (NCCPA) Professional Profile Survey was utilized to develop the base year counts and age-sex characteristics for PAs practicing certain specialties. These specialties are Anesthesiology, Emergency Medicine, Neurology and Physical Medicine and Rehabilitation.

Modeling New Entrants

The primary sources of data on characteristics of physician graduates is the Association of American Medical Colleges (AAMC) 2012-2013 Graduate Medical Education Census. This is completed by residency program directors and administrators. Graduates were assigned to Anesthesiology according to the base year proportions reported in the 2013 AMA Master File from the American Board of Medical Specialties (ABMS).

Numbers and characteristics of new CRNAs came from the 2012 American Association of Colleges of Nursing (AACN) survey. The Physician Assistants Education Association data were used to determine the number of new PAs trained. The 2013 NCCPA Professional Profile was used to determine the characteristics of the new PAs. We assumed that the distribution of PAs by different characteristics would remain the same as in the current workforce. Regional provider supplies were projected by simulating the locational choice of providers in light of any existing shortage/surplus.

Modeling Workforce Attrition

Physician retirement rates were calculated from the 2012 and 2013 Florida Bi-annual Physician Licensure Survey. This survey asks active physicians about their intention to retire. These data were compared to the AAMC’s 2006 Survey of Physicians over Age 50. This AAMC survey collected information on age at retirement or age expecting to retire.

Both sources showed similar retirement rates. The Florida survey had a larger sample size and more detailed individual specialties. Retirement rates were combined with the age-sex specific mortality rates adjusted downward to reflect the lower mortality of healthcare workers.2 Emergency medicine, anesthesiology, and radiology showed earlier retirement rates compared to physicians in other specialties. Attrition pattern for family physicians is used as proxy for attrition rates of PAs and CRNAs.

Modeling Hours Worked

Ordinary Least Squares regressions were conducted for each occupation. Reported average hours worked per week was the dependent variable. Age group, sex, and age-sex interaction were explanatory variables. For physicians, data from the Florida 2012-2013 bi-annual Physician Licensure Survey (n=18,016) file of physicians were used. Hours worked patterns differed by specialty. An FTE is defined for each specialty as the average number of patient care hours worked in that specialty.

A regression analysis was conducted using 2013 NCCPA Professional Profile Survey to model hours worked patterns of PAs. Another regression used the 2006-2012 ACS to model hours worked patterns of CRNAs. An FTE is defined for each occupation as the average hours worked per clinician in that occupation and specialty. This uses data for clinicians working at least 20 hours per week.

Modeling Demand Projections

Projected demand for physicians, CRNAs and PAs begins with the predicted probabilities for each demographic group estimated from MEPS data. These probabilities are applied to the simulated microdata set for future years derived from the Census Bureau. The result is projected service use specific to the settings where these providers work.

We developed prediction equations for office visits, inpatient days, and emergency room visits for each type of provider using 2008-2012 MEPS data. These were developed using logistic regression and the appropriate ICD9 codes. Those codes for neurology were 320-359, 742, 781, 784, and 800-804. The codes for physical medicine and rehabilitation services were 0.4-00.5, 17.5, 35-39; and 93. Separate regressions were estimated for children and adults.

Prediction equations for ED visits used a similar approach but did not use ICD9 codes. Instead, we assume all ED visits involved a consultation with an emergency physician. Because MEPS lists only the highest level of provider seen, the 2010 NHAMCS is used to identify the probability that a PA was also seen. Provider demand in anesthesiology was determined by the demand for all surgical procedures across all settings. The predicted probabilities of service use by demographic groups when applied to the future population predicted the workload of the different occupations.

Exhibit 68 provides the staffing ratio for each type of service. This is derived by dividing the current volume of services by the number of provider FTE who currently provide these services. This is then applied to the projected service demand to obtain the predicted demand for provider FTE.

The regional provider demand estimates were derived by pro-rating the national demand for health care services. This is based on the population characteristics of the regions (e.g., age, sex, household income, insurance status, health status, etc.).

  • 2Johnson NJ, Sorlie PD, Backlund E. The Impact of Specific Occupation on Mortality in the U.S. National Longitudinal Mortality Study. Demography. 1999;36(3):355-367. doi:10.2307/2648058.
Date Last Reviewed: