Behavioral Health Care Provider Model Components

Published 2020

In this module:

Behavioral health care is care that addresses any behavioral problem, including mental health and substance abuse conditions, stress-linked physical symptoms, and patient activation and health behaviors. In its 2016 and 2018 reports, HRSA reported workforce supply and demand projections for: 

  • psychiatrists 
  • psychologists 
  • psychiatric/mental health nurse practitioners (NPs) 
  • psychiatric/mental health physician assistants (PAs) 
  • substance abuse and behavioral disorder counselors (addiction counselors) 
  • mental health counselors 
  • school counselors 
  • social workers 
  • marriage and family therapists (MFTs) 
  • psychiatric aides & technicians1 2

These occupations are included in HRSA’s updated 2020 report, although the definition for psychologists and social workers has evolved.  

The 2016 report2 included all psychologists trained at the master’s degree or higher, whereas HRSA’s 2018 report and 2020 update includes doctorate-level psychologists only. This change was based on feedback from the American Psychological Association that master’s-level psychologists were too different from doctorate-level psychologists to model them together. Master’s-level psychologists generally work as counselors rather than as clinical psychologists.

Due to data challenges distinguishing between clinical social workers and social workers in non-clinical roles, the criteria used to identify social workers for modeling have evolved across studies. In the 2018 report and 2020 update, the model includes social workers trained at the master’s degree level or higher. The main data constraint is that national data sources such as ACS do not identify which social workers are clinical social workers.

For the 2018 report, the research team reached out to professional associations representing behavioral health professions. They helped to identify the best available data sources, discuss trends affecting workforce supply and demand, and provide feedback on preliminary findings. The information provided in this technical documentation and in HRSA reports does not necessarily reflect the views of the associations that responded to our invitation to participate in the workforce study. There may not be clear consensus on all assumptions. Individuals from the following associations participated:

  • Association of Social Work Boards (ASWB) 
  • American Academy of Physician Assistants (AAPA) 
  • American Association for Marriage and Family Therapy (AAMFT) 
  • American Psychiatric Association (APA) 
  • American Psychiatric Nurses Association (APNA) 
  • American Psychological Association (APA)3
  • American School Counselor Association (ASCA) 
  • Association for Addiction Professionals (NAADAC) 
  • College of Psychiatric and Neurologic Pharmacists (CPNP) 
  • Council on Social Work Education (CSWE) 
  • National Board for Certified Counselors (NBCC) 

Many of these associations provided data on current supply and the number and characteristics of new graduates. The remainder of this module describes how HWSM models supply and demand for behavioral health providers. 

The 2020 update differs from the previous reports in the following ways: 

  1. Supply and demand for psychiatrists, and demand for the other behavioral health occupations, is modeled by metropolitan/nonmetropolitan designation4 to better understand geographic imbalances in adequacy of supply for nonmetropolitan areas. 
    1. Supply of adult and pediatric psychiatrists is modeled by metropolitan status using data from the AMA Masterfile to estimate the number and characteristics of psychiatrists in each county and aggregated by to the metropolitan/nonmetropolitan level. 
    2. There is insufficient information on practice location to model the other behavioral health providers by metropolitan status. 
  2. Earlier reports modeled an “Unmet Needs” scenario, which was later replaced with a “Reduced Barriers” scenario. The main reasons for this change were: 
    1. Unmet needs should be defined clinically. However, the available data to define unmet needs relied on self-reported data from survey respondents. These data were if respondents perceived they did not receive all the services they needed. The reduced barriers scenario could be modeled using a more rigorous approach. 
    2. The unmet needs scenario assumed the same gap in unmet need (estimated at 20% based on self-report survey data) across all behavioral health occupations and for both pediatric and adult services. The improved access scenario estimates the gap between current levels of care and levels required to reduce inequities in care use. These levels can differ by behavioral health type and for children versus adults. The reduced barriers scenario shows a much larger gap for pediatric psychiatrists than for adult psychiatrists. These are larger gaps than estimated for other behavioral health occupations. 
  3. The 2020 update estimates the level of behavioral health services provided by primary care physicians. 
  4. More recent data sources were used for supply and demand modeling, where available. 
  5. The 2020 update more clearly articulates interpretation of the supply and demand comparisons for select occupations where supply is growing rapidly compared to demand.
  • 1Health Resources and Services Administration. Behavioral Health Workforce Projections. U.S. Department of Health and Human Services; 2018.
  • 2 a b Health Resources and Services Administration. National Projections of Supply and Demand for Selected Behavioral Health Practitioners: 2013-2025.; 2016.
  • 3The American Psychological Association (APA) recently completed a workforce study using HWSM. The data, methods, assumptions and findings in the HRSA study are consistent with those in the APA. The projections differ slightly, though, because HRSA uses a FTE definition of 40 hours worked per week, whereas the APA projections are based on the average hours worked by psychologists (39.0 hours/week).
  • 4Centers for Disease Control and Prevention. 2013 NCHS Urban-Rural Classification Scheme for Counties.

Modeling Supply

Supply modeling consists of:

  • estimating the number and characteristics of current supply 
  • modeling the number and characteristics of new entrants to supply 
  • modeling retirement 
  • modeling workforce behavior such as hours worked patterns and geographic mobility 

In addition to the Status Quo scenario that models the continuation of current workforce participation patterns and new entrants, we also model a “Graduates Trend” scenario for the rapidly growing occupations of NPs and PAs. This scenario projects the number of future new entrants instead of using the most recent historical numbers for all future years. 

Estimating the Base Year Workforce Supply 

The data sources for estimating starting supply in 2018, by state, are the following: 

  • Psychiatrists: 2018 American Medical Association (AMA) Master File  
  • Psychiatric/mental health physician assistants: National Commission on Certification of Physician Assistants (NCCPA) Statistical Profile of Certified Physician Assistants5 for counts and American Academy of Physician Assistants (AAPA) Masterfile for demographics 
  • Psychiatric/mental health NPs: HRSA’s 2018 National Sample Survey of Registered Nurses (NSSRN) to estimate the proportion of new NPs entering behavioral health, combined with data from the American Association of Colleges of Nursing (AACN)6 on nurses completing NP programs 
  • Psychologists: American Psychological Association for the de-duplicated state totals of licensed psychologists7  
  • Addiction counselors, social workers, and psychiatric technicians and aides: Bureau of Labor Statistics (BLS) Occupational Employment Statistics (OES)8 9 10   
  • Mental health counselors: National Board for Certified Counselors (NBCC) for the de-duplicated state totals of licensed mental health counselors11
  • School counselors: Department of Education National Center for Education Statistics (NCES)12
  • Marriage and family therapists: Substance Abuse and Mental Health Services Administration’s (SAMHSA) Behavioral Health, United States (2012) report, with total de-duplicated marriage and family therapist licenses by state—using 2011 data and HWSM to simulate supply to a starting level in 201813

The AMA data contains information on psychiatrists’ age and sex. The NSSRN contains responses to similar questions regarding psychiatric NPs. The AAPA Masterfile contains demographic information about psychiatric PAs. The above other data sources indicating total number of licensed or active providers by state do not contain provider characteristics. To fill in this gap, 2014-2018 ACS data, which contain demographic information for each occupation, were used.  

Multiple years of ACS data were combined to ensure a sufficient sample size in each state to draw a representative sample (with replacement) of the workforce population by demographic. The sample size is equal to the state-level estimates of licensed or active providers from the above data sources. For occupations with licensed provider counts, which include psychologists, marriage and family therapists, and mental health counselors, a random sample is drawn from all such providers (active and non-active) in the ACS. For occupations with active provider counts, including social workers, addiction counselors, and school counselors, a random sample of active providers is drawn from ACS. For purposes of the ACS draw, active status is based on an employment status variable. Responses of “not in the labor force” are considered inactive. 

For social workers, the sample is drawn from the population of social workers in ACS with education level of master’s degree or higher. For addiction counselors, school counselors, marriage and family therapists, and mental health counselors, the sample is drawn from the ACS responses of each specific occupation. (Previously, the ACS grouped all these occupations together. Starting with the 2018 ACS, that is no longer the case). The mental health counselor sample is drawn from the population of counselors in ACS with education level of master’s degree or higher. The sample of addiction counselors is drawn from the population of counselors with education level of associate degree or higher.  

HWSM uses a microsimulation modeling approach that simulates labor force participation decisions at the individual provider level. For modeling, using the above data sources, a database was created containing a simulated population of the behavioral health workforce in each state. This database has an individual record containing occupation, state, age, sex, and race/ethnicity. HWSM uses this database as the starting point to project future supply through 2030.  

The rurality for psychiatrists and psychiatric PAs is based on the county of his/her office location listed in the AMA data or AAPA Masterfile. Counties are assigned one of six designations of rurality14 , with NCHS classifications 1-4 designated as metropolitan areas and classifications 5-6 designated as nonmetropolitan areas. Psychiatric NPs in the NSSRN are available only at the MSA/non-MSA rurality level due to survey confidentiality restrictions. Because the ACS does not have a geographic indicator smaller than PUMAs (Public Use Microdata Areas), supply modeling by metropolitan designation is not available for the behavioral health professions other than psychiatrists, NPs, and PAs.  

Modeling New Entrants 

To model additions to the workforce each year, a synthetic population for each occupation was created. In it, the number of newly created individuals reflects annual new graduates and the demographics of these new graduates (Exhibit 18). 

A total of 1,241 psychiatrists completed their residency training in 2017-2018. This total consists of 1,224 psychiatrists completing residency from an Accreditation Council for Graduate Medical Education program and 17 from American Osteopathic Association accredited programs. Of the 1,224, 390 physicians completed training in child and adolescent psychiatry.15 16 The AMA Masterfile contains the year each psychiatrist completed his or her graduate medical education. The age and sex distribution of new graduates from 2010-2018 were used to calculate the age and sex distribution of new psychiatrists.  

For psychiatric NPs and PAs, the annual number of new graduates is estimated using data from two source. For NPs, the NSSRN has the proportion of NPs who have graduated in 2010 or later that entered behavioral health. For PAs, a survey of PAs certified in the last six months from NCCPA notes the proportion of new PAs entering behavioral health.17 Demographics of new psychiatric NPs and PAs came from analyzing the NSSRN and the American Academy of Physician Assistants (AAPA) Masterfile, respectively. New NPs and PAs are defined as those who have graduated in 2010 or later unless more data are needed. In those cases, data are based on graduates in 2000 or later. 

For occupations other than the psychiatrists, psychiatric nurse practitioners and psychiatric physician assistants, the 2018 Integrated Postsecondary Education Data System (IPEDS) data are used to determine the number of annual new graduates. It also provides data on the sex and race distribution of the new graduates. IPEDS collects the number of new graduates by sex and race for each Classification of Instructional Programs (CIP) code and degree level. Aside from addiction counselors, only master’s degrees for the most appropriate CIP code were counted in this analysis. For addiction counselors, it included associates, bachelor’s and master’s degrees. 

To calculate the age distribution of new entrants to the workforce for the non-physician occupations, the number of providers of a particular age (e.g., age 29) in 2014-2018 ACS data were compared to the number who were one year older (e.g., age 30) in the subsequent year’s file. The amount by which the number of providers age 30 exceeds the number of providers age 29 reflects the net number of new providers entering the workforce at age 29-30 in subsequent years of the survey. With this information, the approximate age distribution of new entrants to the workforce is estimated. Multiple years of data were used to increase sample size by individual age and occupation. The state distributions of providers aged 30-39 years were used to assign future new entrants in the model to a state.

Modeling Workforce Participation

For all occupations except psychiatry (physicians, NPs, and PAs), labor force participation rates were calculated as 1 minus the proportion of such individual providers in the 2014-2018 ACS data who were classified as “not in the labor force.” The rates were calculated for provider age 50 or less by age, sex, race/ethnicity, and occupation. Data for modeling starting supply of psychiatrists come from the AMA Masterfile. While these data contain little information with which to model labor force participation patterns for physicians, published studies suggest physicians commonly retiring between age 60 and 69.18 We model all physicians under age 50 as active. Attrition/retirement patterns differ depending on the occupation being modeled. 

Physicians: Attrition probabilities for psychiatrists are based on self-reported expected retirement age for 1,294 physicians who participated in AAMC’s 2019 National Sample Survey of Physicians (NSSP). These 1,294 were categorized by IHS Markit as “other specialties” physicians, which includes psychiatrists and other specialties not categorized as primary care, medical subspecialties, or surgical specialties.19 The sample size of the NSSP is too small to create a retirement pattern based solely on psychiatrists. To conduct this survey, AAMC contracted with Toluna, an external firm that recruited active physicians from proprietary panels of healthcare professionals. The survey started February 25, 2019 and concluded March 25, 2019 upon reaching the desired quota of 6,000 participants. Survey responses were weighted to be representative of practicing physicians by specialty, age group, sex, and International Medical Graduate status consistent with the 2018 AMA Master File. NSSP findings indicate that physicians intend to retire earlier than in previous HRSA reports. We assume that all physicians have retired by age 90—though very few remain active past age 75. Many older physicians have greatly reduced work hours. 

Nurse practitioners: Attrition probabilities for psychiatric NPs are based responses to the following question in HRSA’s 2018 NSSRN:  

  • Approximately when do you plan to retire from nursing?  
    • Within a Year 
    • In 1-2 years 
    • In 3-5 years 
    • More than 5 years from now 
    • Undecided 
    • Already retired 

We used the responses “already retired”, “within a year”, and “in 1-2 years” as indications of imminent retirement and created a distribution of NP retirement age for NPs age 50 to 74. We assume that all NPs have retired by age 75. Because NPs are approximately 90% female, there is insufficient data to create separate retirement patterns by gender. As such, NPs supply is modeled using a single retirement pattern. 

Physician assistants: Attrition probabilities for psychiatric PAs are based on responses to the question in AAPA’s 2015 National Survey of PAs asking whether the PA plans to retire in the next three years. We created a retirement age distribution of PAs age 50 to 74 based on yes and no responses. We assume that all PAs have retired by age 75. The sample of older PAs in this survey is insufficient to obtain reliable estimates of retirement intention by individual age and specialty. The retirement patterns used for modeling are based on all PAs regardless of specialty—with separate patterns for male and female PAs. 

Other occupations: Attrition patterns for psychologists, social workers, and the counselor and therapist occupations were based on 2014-2018 ACS data for the same occupation categories as used to estimate labor force participation rates. Labor force attrition probabilities were constructed based on comparing whether an individual is employed in a given year and the subsequent year. Cohort differences across subsequent years of data were used to estimate the net number of people leaving the workforce each year. For example, net attrition between age 65 and 66 is estimated by comparing active supply of providers age 65 in a given year to active supply of such providers age 66 in the subsequent year. Multiple years of data were used to increase sample size. The attrition probabilities vary by occupation and age. The sample size is insufficient to model attrition by sex or race/ethnicity. 

Modeling Hours Worked 

Ordinary Least Squares regression is used to model hours worked patterns using a separate regression for each health occupation. The dependent variable is total hours worked per week. Explanatory variables included provider age group (age <35, 35-44, 45-54, 55-64, 65-74, and 75+ years), sex, and age group by sex interaction term. For regressions based on ACS data, additional explanatory variables were race/ethnicity and ACS survey year. The data source differs by occupation.  

Physicians: AAMC’s 2019 NSSP asked 1,259 physicians categorized as “other specialties,” which includes psychiatrists, how many hours they worked during their last typical week of work excluding any week with leave. The survey also asked physicians the percent of time spent on types of activities (patient care, teaching, research, administration, other), as well as weeks worked per year. We used total hours from any of these activities to be consistent with modeling methodology for other occupations where type of activity is not available. We found little difference in hours worked patterns for male and female physicians or by physician age in weeks worked per year. As such, we used weekly hours worked for modeling. 

Nurse practitioners: HRSA’s 2018 NSSRN asked the 1,299 NPs that we categorized in behavioral health care the number of hours worked in a typical week for their primary nursing position. The weighted mean of weekly hours worked is 38.1 hours. Statistically significant differences in weekly hours worked existed across age groups and by sex. The interaction terms by age group and sex were not statistically significant—likely due to the small sample of male NPs.  

Physician assistants: AAPA’s 2019 Salary Survey collected data on hours worked per week for the primary employer for 129 PAs who reported working in behavioral health care. The average hours worked per week was 40.3.  

Other occupations: A similar OLS regression approach is employed with ACS data to separately model hours worked patterns for the other health occupations, but with a different model specification. The dependent variable is total hours worked in the previous week. Explanatory variables consisted of age group, sex, race, and a year indicator (as the ACS pooled data from 2014-2018). 

Supply projections account for changing demographics of the workforce and that average hours worked per week differs by age, sex, race, and occupation. The expected number of hours worked by each individual is converted to FTE supply by dividing the total person- hours worked by 40. This creates a uniform standard of 1 FTE as working 40 hours per week, regardless of the occupation. It also means that the initial FTE of an occupation will differ from the actual number of active providers. 

Modeling State-Level Supply and Migration 

Behavioral health occupations often have different mix of providers and levels of supply adequacy geographically. To estimate this, HWSM includes state-level supply estimates where sufficient data are available. In the ACS and OES, some occupations do not have totals reported in every U.S. state. State-level supply estimates are unavailable for these occupations. Furthermore, state-level estimates are unavailable for some behavioral health specialties or occupations (including marriage and family therapists and social workers). 

For occupations with sufficient state-level data, HWSM models future movement of behavioral health professionals across states. This is accomplished in two steps. First, a logistic regression on ACS data estimates the probability of migrating to any other state for workers under age 50 as a function of age group, sex, race, the population of the state from which the person is moving, and a year indicator. Then, the simulation compares the predicted move probability to a random number from a uniform distribution to simulate whether each provider moves to a new state. The assigned new state is based on the state distribution of destination states for individuals who were observed to have moved in the ACS data.  

For psychiatrists, instead of using ACS data for all physicians and their move probability, we used National Plan and Provider Enumeration System (NPPES) data for 2019 and 2020. Using multiple years of NPPES data allowed us to identify physicians who changed states by merging the two years using the physician’s national provider identification (NPI). The NPPES data does not contain the provider’s age. We merged NPPES data with Medicare data that had NPI and graduation year and assumed that physicians are age 28 at graduation from medical school. This was the average age at graduation for physicians in the AMA Masterfile. 

We then compare each person’s move probability to a uniform random number between 0 and 1 to simulate whether a physician moves to a new state each year over the projection horizon. In addition to provider age, sex, and occupation as explanatory variables for moving, we included OES data on physician salary by state adjusted for state cost of living using data from the Bureau of Economic Analysis. The rationale is that states with higher pay relative to cost of living will attract more physicians than will states with lower relative pay.  

Also included as an explanatory variable is state population. The assumption is that states with large populations have more opportunities for intra-state movements and thus fewer inter-state movements. Using NPPES data to analyze physician migration is a new component of the HWSM in 2020. Future areas of improvement include examining the assumptions used to impute physician age and evaluation of the optimal explanatory variables in the logistic regression. 

Modeling Demand

Modeling Demand for Health Care Services and Providers provides an overview of the data sources and approach to model demand for health care services and providers. This section provides additional details on modeling demand for behavioral health providers. 

Demand Scenarios

Demand for behavioral health providers traditionally has been modeled under two scenarios. The Status Quo scenario models future demand under current care use and delivery patterns. It accounts for changing demographics and variation across individuals in patterns of seeking behavioral health services. This scenario sets national demand equal to supply in 2018 to enable extrapolating a “2018 level of care” to future populations. The exception is for psychiatrists. That demand starts 6,894 active psychiatrists higher than supply to reflect the number of providers required to remove all mental health professional shortage area (HPSA) designations.20 This scenario is used to assess whether the nation’s future behavioral health workforce will be sufficient to provide at least the current level of care, even though many may consider current levels inadequate given reports of unmet need.21

Past behavioral health reports have modeled an unmet needs scenario. This scenario reflects the additional providers required to address both demand and unmet need.22 23 The rationale for modeling this unmet need scenario is that the health system continues to improve access to and comprehensiveness of behavioral health services. These trends include:  

  • better affordability through increased coverage by insurance 
  • more intensive use of screening 
  • efforts to decrease stigma 
  • better understanding of how to address behavioral health issues 
  • increased integration of primary care and behavioral health 
  • greater use of team-based care with a broader range of behavioral health providers that provides opportunities for task shifting and providing a more comprehensive range of behavioral health services 
  • greater use of technology such as telemedicine to reduce barriers to accessing care in rural areas and improve care to patients with mental health needs in emergency departments, nursing facilities, and other care settings 

Historically, the unmet needs gap was modeled at 20%, reflecting a likely lower bound on the true level of unmet need.24 However, there are several issues with the 20% unmet needs estimate. One, this estimate was calculated based on adult patient self-reporting of having received any services. It did not clinically evaluate whether the services are sufficient. For example, adults with any mental illness (AMI) who receive no services likely have lower mental health needs, on average, as compared to adults who receive services and adults with AMI who received services may still have unmet need.25 26 Two, children and adolescents possibly have higher rates of unmet need than do adults. MHSS, however does not collect sufficient information to quantify unmet need for mental health services for children and adolescents. Three, even within the behavioral health workforce, the level of unmet need likely varies by occupation. For example, the unmet needs estimate for addiction counselors likely differs from the unmet needs estimate for mental health.  

For this updated study we analyzed MEPS data to develop a Reduced Barriers scenario. This scenario estimates the number of behavioral health FTEs required to close the gap between care use for populations facing access barriers compared to populations perceived to have fewer barriers to behavioral health care. Those facing access barriers include nonmetropolitan, minority, or without health insurance populations. Those perceived to have fewer barriers include metropolitan, White, non-Hispanic, or with health insurance populations. 

The Reduced Barriers scenario uses a more rigorous approach to quantify the gap. It also allows for the size of the gap to differ for adults and for children and adolescents. This scenario also is consistent with HRSA’s strategic plan to “…address health disparities through access to quality services.....”27 This scenario helps quantify the impact of each factor when a certain level of change of each factor is expected. For example, when a proportion of uninsured population is expected to be covered by insurance, a rough estimate on the impact to the demand of health workforce can be proportionally developed. 

The rationale for modeling this goal is to illustrate the wide disparities in use of behavioral health services. Modeling this scenario also is consistent with SAMHSA’s aim to improve behavioral health equity, which is defined as “the right to access quality health care for all populations regardless of the individual’s race, ethnicity, gender, socioeconomic status, sexual orientation, geographical location and social conditions through prevention and treatment of mental health and substance use conditions and disorders.28

If underserved populations had equal access to care as populations with fewer access barriers, we estimate that demand for psychiatrists would increase by 43%. This includes a 37% increase for adult psychiatry and a 69% increase for child psychiatry (Exhibit 19). This Reduced Barriers scenario is a hypothetical scenario describing the workforce demand implications if policies and programs could close the gap in access to psychiatry services. The gap is smaller for some occupations reflecting, in part, that change in demand associated with the Reduced Barriers scenario applies only to care delivered in healthcare provider offices, emergency departments, and hospitals. It does not apply to care delivered in nursing homes, residential care facilities, school-based settings, and public health settings.

  • 20Bureau of Health Workforce, Health Resources and Services Administration. Designated Health Professional Shortage Areas Statistics: Third Quarter of Fiscal Year 2019 Designated HPSA Quarterly Summary. U.S. Department of Health and Human Services; 2019.
  • 21Christidis P, Lin L, Stamm K. An unmet need for mental health services. Accessed July 23, 2019.
  • 22Health Resources and Services Administration. Behavioral Health Workforce Projections. U.S. Department of Health and Human Services; 2018.
  • 23Health Resources and Services Administration. National Projections of Supply and Demand for Selected Behavioral Health Practitioners: 2013-2025.; 2016.
  • 242017 National Survey on Drug Use and Health: Detailed Tables. Substance Abuse and Mental Health Services Administration; Center for Behavioral Health Statistics and Quality; 2018.
  • 25Han B, Compton WM, Blanco C, Colpe LJ. Prevalence, Treatment, And Unmet Treatment Needs Of US Adults With Mental Health And Substance Use Disorders. Health Affairs. Published online October 12, 2017. Accessed May 2, 2018.
  • 26Walker ER, Cummings JR, Hockenberry JM, Druss BG. Insurance Status, Use of Mental Health Services, and Unmet Need for Mental Health Care in the United States. Psychiatric Services. Published online March 1, 2015. doi:10.1176/
  • 27Health Resources & Services Administration. Strategic Plan FY 2019-2022. Department of Health and Human Services; 2019. Accessed June 8, 2020.
  • 28Substance Abuse and Mental Health Services Administration. Behavioral Health Equity. Department of Health and Human Services; 2019.

Addressing Data Limitations 

Adapting HWSM to model demand for behavioral health services and providers required addressing critiques of HWSM and data limitations. 

One critique of the prediction equations used in HWSM has been the lack of variables related to mental health or substance abuse. Many of the patient characteristics included in HWSM are correlated with receipt of mental health services. These include: 

  • demographics 
  • family income level 
  • presence of chronic disease 
  • insurance type (especially insured by Medicaid) 
  • lifestyle choices 
  • metro/nonmetropolitan location  

However, HWSM contains no information on use of illicit drugs, alcohol consumption, and mental health variables such as presence or severity of depression. This omission reflects data challenges and the need for variables with the same definitions in two places. First, they are needed in MEPS and data sources used to estimate the relationship between patient characteristics and use of health care services. In addition, they should be found in the population file to which the prediction equations are applied.  

MEPS contains information on patients’ visits to psychiatrists, psychologists, and social workers.  Negative binomial regressions are used for modeling ambulatory (office and outpatient) visits  where: 

  • the dependent variable is the number of annual visits to a specific type of provider 
  • the explanatory variables consist of 
    • demographics 
    • lifestyle variables
    • family income 
    • insurance information
    • presence of select chronic conditions 
    • resident county level of rurality 

However, MEPS does not specifically identify visits to mental health counselors, marriage family therapists, and addiction counselors. These occupations are not differentiated in MEPS but rather are listed under the “all other non-physician” category. Similarly, while MEPS identifies visits to NPs and PAs, it does not specifically identify visits to psychiatric NPs and PAs. 

Associated with each visit is a reason code. This code can indicate whether the primary reason for the visit is associated with mental health. The sample size for mental health visits to NPs and PAs is too small to provide reliable regression results. Therefore, to construct prediction equations for patients’ annual use of ambulatory services for these occupations, we did the following: 

  • Mental health counselors, marriage family therapists, and addiction counselors: Total annual ambulatory behavioral health visits were used to predict the relationship between patient characteristics and use of mental health services. The analysis included visits to psychiatrists, psychologists, social workers, and the “all other non-physician” category where the visit is indicated as a visit for mental health or substance abuse. Using this broad category of behavioral health visits assumes that patient characteristics correlated with greater number of mental health visits to psychiatrists, psychologists, clinical social workers or other non-physicians. It also assumes they area correlated with greater use of services provided by mental health counselors, marriage family therapists, and addiction counselors. (Further adjustments were made to model demand for addiction counselors). Using projected total behavioral health visits as the workload driver will overstate the actual number of visits to individual provider occupations. It will, however, serve as a scalar for the projected growth rate in demand and a scalar for how demand varies across different geographic areas. 
  • Psychiatric NPs and PAs: Due to insufficient sample size in MEPS, annual mental health visits to NPs and PAs could not reliably be modeled. Also, in visits where the patient saw a psychiatrist in addition to an NP or PA, only the psychiatrist is listed as the provider seen. We estimated prediction equations for patient use of mental health services, regardless of who provided such services, as a proxy for use of psychiatric NPs and PAs. Total visits will overstate the actual number of visits where NPs and PAs provide services. This workload driver, however, will provide a scalar of how demand for services likely will grow over time and a scalar for level of demand across geographic areas. The decision to use total visits irrespective of provider type (e.g., primary care provider, psychiatrist office, psychologist office, other health provider) reflects two considerations:
    • NPs and PAs often provide mental health-related services to patients irrespective of the provider’s area of specialization (e.g., treating patients for depression during a primary care visit).
    • Certain patient characteristics might be correlated with visits to non-physicians instead of physicians (e.g., whether residing in a metropolitan area, insurance type including insured by Medicaid, and family income level).

The above prediction equations apply only to modeling services in provider office and outpatient settings. Different assumptions and prediction equations are used for modeling demand for these occupations in other care delivery or employment settings (see Exhibit 21). 

Model validation activities, discussed below, suggest that the prediction equations and resulting workforce demand projections capture much of the variation across states in demand for mental health services. However, the prediction equations do not adequately capture variation across states in demand for addiction counselors. When comparing (a) state-level estimates of 2018 demand for addiction counselors per 100,000 adults and per 100,000 youth and adolescents to (b) state-level estimates of prevalence of substance use disorder (SUD) among the populations age 18 and older and age 12-17, there appears to be low correlation (i.e., correlation coefficient <0.2).29 This low correlation suggests that patient characteristics not included in HWSM (e.g., illicit drug use and alcohol abuse) vary systematically by state. In addition, the patient characteristics in the model (i.e., demographics, lifestyle variables, disease prevalence, family income, insurance type) do not adequately explain need for addiction counselor services.30

Therefore, state-level multiplicative scalars were used to adjust demand for addiction counselors to reflect state variation in SUD prevalence. To create the scalars, SUD prevalence for each state from SAMHSA’s 2017 & 2018 National Survey on Drug Use and Health (NSDUH) is divided by the national average prevalence (Exhibit 20).31

  • 29Substance Use Disorder is defined as meeting criteria for illicit drug or alcohol dependence or abuse.
  • 30Future research might consider using National Survey on Drug Use and Health (NSDUH) data to statistically match people in the constructed population file used in HWSM with people in NSDUH to add mental health and substance use/abuse data. The HWSM prediction equations, based on MEPS, have small predictive power for identifying individuals requiring behavioral health counseling or treatment. However, when individual predictions of demand for behavioral health services are aggregated to the state level, HWSM can predict how broad demographic and socioeconomic factors will affect overall demand for behavioral health services.
  • 31Center for Behavioral Health Statistics and Quality. 2017-2018 NSDUH Estimated Totals By State: Substance Use Disorder in the Past Year, by Age Group and State (Table 23). Substance Abuse and Mental Health Services Administration; 2020.

If a state had 10% higher or 10% lower prevalence relative to the national average, the scalar would be 1.1 and 0.9, respectively. While these scalars changed the state-level projections, they had only a negligible effect on national projections when projected to 2030. 

NSDUH data from 2015 to 2018 suggests that SUD prevalence among adults has remained relatively steady at about 6.6% for adults age 26 or older and 15% for adults age 18-25. Prevalence among youth age 12-17 has declined slightly over the past years from about 5% in 2015 to 3.7% in 2018. We assume that current rates will continue through 2030. 

Converting Demand for Visits to Demand for Providers 

The approach to modeling demand for behavioral health workers is similar to the general approach described in Modeling Demand for Health Care Services and Providers for other health workers. First, national estimates of total FTE providers in each care delivery setting were estimated. The primary source is the 2018 ACS and 2019 OES. Occupation-specific data sources are used for psychiatrists, nurse practitioners, physician assistants and psychologists (Exhibit 21). Total workload measures were divided by FTE supply in 2018 to calculate staffing ratios by occupation and care delivery setting (Exhibit 22). The Status Quo demand scenario models the continuation of current provider-to-visits ratios during the projection period. It also models state-level demand for services and providers by applying national patterns to the characteristics of each state’s population.  

Primary Care Providers as a Source of Behavioral Health Services 

This study analyzed the role of primary care providers (PCPs) in the delivery of behavioral health services, where such activities include:  

  1. Screening: Primary care is often the entry point to the health care system. PCPs help screen patients and identify the need for behavioral health treatment. The US Preventive Services Task Force (USPSTF) recommends that PCPs screen children, adolescents and young adults for behavioral health outcomes.33 34 USPSTF also recommends screening and counseling in primary care settings for adults regarding excessive alcohol use and depression.35 36
  2. Treatment: PCPs sometimes are involved in the direct treatment of patients. This is especially seen in medically underserved areas without adequate behavioral health infrastructure to refer patients to other providers. Treatment includes counseling, prescribing medications for depression and anxiety, and prescribing methadone to treat opioid use disorder.37
  3. Collaboration within multidisciplinary teams: Integration of behavioral health into primary care practices is increasing the number of PCPs who are part of multidisciplinary teams that include professional behavioral health providers.38

One goal of this analysis was to estimate the proportion of primary care physicians’39 time spent providing treatment for patients’ behavioral health disorders. Another was to better understand if this proportion of time differed by patient characteristics—e.g., residing in nonmetropolitan county or patient’s gender. This investigation is based on analysis of patient visits to primary care physician offices in the 2014, 2015 and 2016 National Ambulatory Medical Care Survey (NAMCS).  

NAMCS is based on a representative sample of physician office visits. For each participating physician practice, information was collected for a random sample of patient visits obtained through record extraction. Visit information includes the type of physician seen, the length of time (in minutes) the physician spent with the patient, and up to five diagnosis codes. The NAMCS sampling frame includes three specialties categorized as primary care: family medicine, general internal medicine, and general pediatrics. In addition to these three specialties, we created a fourth proxy category for geriatric medicine visits. We created this category by using all patients aged 75 and older in the family medicine or general internal medicine specialties across all three years of data.  

Our analysis found that 13.8 percent of visits to a primary care physician included a behavioral health diagnosis code (Exhibit 23).40 For patients with multiple diagnoses for the visit, insufficient information exists to know what proportion of physician time was spent addressing behavioral health diagnoses. Consequently, the visit time was pro-rated by dividing total behavioral health diagnoses by total diagnoses. Say a 15-minute visit had three diagnoses and one was behavioral health. In this case, one third (5 minutes) of the visit was counted as time spent providing behavioral health services. Physician screening or counseling for behavioral health services was not counted as time providing behavioral health services in the absence of a behavioral health diagnosis. 

Using this approach, an estimated 5.8 percent of physicians’ direct patient care time is spent providing behavioral health services. The time ranges from a high of 6.4 percent for family medicine to a low of 4.0 percent for geriatric medicine. Using the 5.8 percent value multiplied by our estimate of 256,220 FTE primary care physicians in 2018 yields approximately 14,900 FTE primary care physicians providing behavioral health services. NAMCS data are insufficient to calculate the proportion of primary care NP and primary care PA time spent providing behavioral health services. 

  • 33U.S. Preventive Services Task Force. Final Recommendation Statement: Depression in Children and Adolescents: Screening.; 2016.
  • 34O’Connor E, Thomas R, Senger CA, Perdue L, Robalino S, Patnode C. Interventions to Prevent Illicit and Nonmedical Drug Use in Children, Adolescents, and Young Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA. 2020;323(20):2067-2079. doi:10.1001/jama.2020.1432.
  • 35US Preventive Services Task Force, Curry SJ, Krist AH, et al. Screening and Behavioral Counseling Interventions to Reduce Unhealthy Alcohol Use in Adolescents and Adults: US Preventive Services Task Force Recommendation Statement. JAMA. 2018;320(18):1899. doi:10.1001/jama.2018.16789.
  • 36Siu AL, and the US Preventive Services Task Force (USPSTF), Bibbins-Domingo K, et al. Screening for Depression in Adults: US Preventive Services Task Force Recommendation Statement. JAMA. 2016;315(4):380. doi:10.1001/jama.2015.18392.
  • 37Livingston JD, Adams E, Jordan M, MacMillan Z, Hering R. Primary Care Physicians’ Views about Prescribing Methadone to Treat Opioid Use Disorder. Substance Use & Misuse. 2017;53(2):344-353. doi:10.1080/10826084.2017.1325376.
  • 38Cantone RE, Fleishman J, Garvey B, Gideonse N. Interdisciplinary Management of Opioid Use Disorder in Primary Care. The Annals of Family Medicine. 2018;16(1):83-83. doi:10.1370/afm.2184.
  • 39Insufficient information was available to estimate the proportion of time spent providing behavioral health services by nurse practitioners and physician assistants in primary care.
  • 40ICD-10 diagnosis codes that begin with “F” (including F00 – F99) were included as a behavioral health diagnosis. Insomnia and sleep disorders (ICD-10 code G470) were also considered a behavioral health diagnosis. ICD-9 diagnosis codes that describe mental health disorders (beginning with 290 – 319) were also included as a behavioral health diagnosis code.

The proportion of primary care physician time providing behavioral health services is almost twice as high in nonmetropolitan areas (9.9%) compared to metropolitan areas (5.5%). This is consistent with estimates of under-supply of behavioral health providers in nonmetropolitan areas and other possible barriers to seeing a behavioral health specialist (e.g., stigma), The time PCPs spent providing behavioral health services in metropolitan areas is equivalent to approximately 12,400 FTE primary care physicians. The time PCPs spent providing behavioral health services in nonmetropolitan areas is equivalent to approximately 2,500 FTE primary care physicians. 

Comparing the PCP time spent providing behavioral health services to the estimated FTE supply of psychiatrists in 2018 shows that PCP physicians are providing approximately 21% of all behavioral health services in metropolitan areas. Primary care physicians are providing 50% of all behavioral health services in nonmetropolitan areas. This is a significant finding that suggests the gap in behavioral health services and lack of specialists and infrastructure in nonmetropolitan areas are being filled by primary care physicians to a large extent. 

Validation Activities

International Society for Pharmacoeconomics and Outcomes Research (ISPOR) guidelines on best practices for model validation activities were followed.41 Validation of HWSM in general is discussed in HWSM Improvement, Validation, Strengths, and Limitations. For the behavioral health component of HWSM, validation activities include the following: 

  • Subject matter experts were engaged to review data inputs and preliminary findings. These included health workforce researchers from two HRSA-funded health workforce centers and representatives from nine associations that represent the behavioral health occupations modeled. 
  • The workforce demand projections were compared to external data sources not used by HWSM to develop the projections to evaluate the state-level projections. For the behavioral health workforce projections, state-level demand projections were compared to various state-level measures for mental health and substance use disorder using correlation coefficients between estimates at the state level.  
    • To compare across metrics, the demand projections were divided by size of the state’s population to calculate demand per 100,000 population.  
    • Number of days with depression/anxiety in the past 30 days data from Behavioral Risk Factor Surveillance System (BRFSS) is one metric for comparison.  
    • Other metrics come from SAMHSA’s NSDUH including: 
      • adult prevalence of any mental illness (AMI) 
      • adult prevalence of serious mental illness (SMI) 
      • youth & adolescent prevalence of major depressive episode (MDE) 
      • adult and youth & adolescent SUD prevalence. 
    • Variation across states in the severity of AMI was estimated from online depression screening data (n=27,511, data collected May 2014-Dec 2016) that reports the proportion of depression cases that are minimal, mild, moderate, moderately severe, and severe by state.42

Correlation coefficients were used to estimate the linear correlation between HWSM demand estimates and the comparison metrics for each state. Most of the estimated correlation coefficients (Rs) suggest moderate (R>0.5) to strong (R>0.7) relationships. Here, R=1.0 represents perfect correlation between the demand projections and the comparison metrics, and R=0.0 represents no correlation. Examples of these correlations include the following: 

  • Psychologist demand is highly correlated with AMI prevalence adjusted for depression severity (R=0.76). The correlation is moderate with SMI prevalence (R=0.63) and BRFSS prevalence data on self-reported days with depression/anxiety (R=0.54). 
  • Psychiatrist demand is moderately correlated with AMI prevalence adjusted for depression severity (R=0.60), with SMI prevalence (R=0.51) and with correlated with BRFFS self-reported days with depression/anxiety (R=0.64). 
  • External metrics of need for behavioral health services are not perfectly correlated with each other. NSDUH-derived prevalence for AMI, MDE, and SMI are highly correlated across states with R=0.85 for these comparisons (perhaps reflecting in part the methods used to construct these prevalence estimates). AMI prevalence from NSDUH has a weak correlation with the BRFSS-derived metric on prevalence of self-reported days with depression/anxiety (R=0.32). 

As discussed previously, demand estimates for addiction counselors had little correlation with SUD prevalence for both adults and for youth and adolescents. Hence, the state-level multiplicative scalars (Exhibit 20) were used to better capture state-level variation in factors contributing to demand for addiction counselors.

  • 41Eddy DM, Hollingworth W, Caro JJ, Tsevat J, McDonald KM, Wong JB. Model Transparency and Validation: A Report of the ISPOR-SMDM Modeling Good Research Practices Task Force-7. Value in Health. 2012;15(6):843-850. doi:10.1016/j.jval.2012.04.012.
  • 42Theresa Nguyen, Michele Hellebuyck, Madeline Halpern, and Danielle Fritze. The State of Mental Health In America 2018 (PDF - 18 MB).; 2018.
Date Last Reviewed: