Arete Healthcare LLC

Independent Practices | Shared Infrastructure | Clinician-Led Governance

Key Findings

In Pictures

Five views examine the diversity and career-stage structure of the 15 specialties: the overall snapshot; early-versus-late career change by gender and race/ethnicity; decomposition of non-White change; the chiropractic trajectory by years since graduation; and the balance of early- versus late-career clinicians overall and by sex. All career-stage comparisons are cross-sectional cohort contrasts, not same-person longitudinal change.

Figure 1. Gender and BIFSG-estimated race/ethnicity share across fifteen specialties. Dashed lines show U.S. Census Bureau Vintage 2020-2025 population estimates; these are not parity targets. The healthcare workforce composition is variable with most currently active providers being white females or males. Visualization was human generated.

Figure 2. Comparison of early and late career gender, race and ethnicity composition of 15 specialties. White male representation declines in 14 of 15 specialties with nurse practitioners being the exception. Much of the compositional movement is toward women, both White and non-White, with non-White women showing gains in all 15 specialties. Notable exceptions are social worker, physical therapist, and nurse practitioner, where white female representation declined in the early career cohort. Visualization was human generated.

Figure 3. Comparison of changes in non-white composition in 15 early and late career provider specialties. Estimated Asian share showed the largest percentage-point increase, growing in every specialty except psychiatry. Relatively large increases in estimated Hispanic share were observed in psychologist, social worker, nurse practitioner, and chiropractor. The largest increases in estimated Black share occurred among nurse practitioners and social workers. Visualization was human generated.

Figure 4. Gender and BIFSG-estimated race/ethnicity share for chiropractors across the number of years since graduation. Dashed lines show U.S. Census Bureau Vintage 2020-2025 population estimates; these are not parity targets. The gender and race/ethnicity mix of recent chiropractic graduates is moving closer to approximating the U.S. population profile with white males remaining over-represented. Visualization was human generated.

Figure 5. Change in specialty workforce career-stage composition by sex. Bars show the percentage-point difference in subgroup share comparing clinicians 0–10 years since graduation with those 31+ years since graduation. Positive values indicate greater representation in the newer cohort, while negative values indicate greater representation in the older cohort. Several specialties, including nurse practitioner, physician assistant, physical therapy, emergency medicine, and social work, show younger composition overall and within both female and male workforces. Across most specialties, female workforces are younger than male workforces. Chiropractic shows increasing representation among younger women, but this is not sufficient to offset an older male cohort, yielding a net older overall workforce profile. Years since graduation is a career-stage proxy, not direct age or a supply forecast. Visualization was human generated.

Arete's Take

Table 1. At a Glance

Source: ProviderGender_Updatedv4.xlsx, reaggregated from 1,954 specialty × single-year-since-graduation × gender cells. U.S. population context uses the U.S. Census Bureau’s 2020–2025 population estimates (NC-EST2025-SR11H). Comparator ranges exclude chiropractic unless otherwise specified. AI-assisted reaggregation and Figure 5 metric verification: ChatGPT (OpenAI; GPT-5.6 Sol), August 26, 2026; pending DAWG verification.

Abstract

This cross-sectional descriptive analysis examines 1,807,746 unique NPIs across fifteen CMS/PECOS-enrolled specialty cohorts using a June 2026 Doctors and Clinicians National Downloadable File linked to primary NUCC taxonomy from NPPES. Specialty classification now follows the same fifteen-category primary-taxonomy architecture used in Arete's Group Size paper rather than the earlier draft's degree-precedence hierarchy. Years since graduation and reported gender come from CMS enrollment fields. Race/ethnicity is not directly reported; estimated composition is generated using Bayesian Improved First Name, Surname, and Geocoding (BIFSG), combining 2020 Census first- and last-name distributions with exact, mutually exclusive 2019–2023 American Community Survey Table B03002 geography. [1–9]

Across the fifteen specialties, female share ranges from 8.7% in orthopedic surgery to 87.0% among nurse practitioners, while BIFSG-estimated White share ranges from 49.9% in internal medicine to 82.1% in chiropractic. Thirteen specialties are more female in the ≤10-year cohort than in the 31+ cohort, and fourteen are less White. Chiropractic remains demographically distinct: its female share is 38.5% at ≤10 years versus 17.0% at 31+, while estimated White share is 76.5% versus 86.3%. The pattern is therefore broader than chiropractic, but the magnitude and direction of change differ materially by specialty. Career-stage renewal also differs sharply by sex: the ≤10-year cohort exceeds the 31+ cohort among women in 14 of 15 specialties but among men in only five. Chiropractic shows a +15.6-point female balance and -11.7-point male balance, producing a -4.2-point overall early-versus-late-career difference.

The career-stage analysis is operationalized as a cross-sectional comparison of years-since-graduation cohorts; the same clinicians are not observed over time. BIFSG produces probabilistic estimated counts rather than self-reported identities, and the CMS source is limited to Medicare-enrolled clinicians represented in the DAC NDF. Population comparisons use the U.S. Census Bureau's 2020–2025 population estimates (NC-EST2025-SR11H) as a contextual reference—not a parity target—and align the five race/ethnicity categories to the BIFSG model. Findings remain provisional pending second-AI citation/data validation and human source confirmation; DAWG review is complete.

Methods

2.1 Study Design, Data Sources, and Population

This is a cross-sectional, descriptive workforce study based on an extract produced in Arete's mimilabs environment. The source SQL combines (1) the latest available NPPES snapshot for primary NUCC taxonomy classification and (2) the CMS Doctors and Clinicians National Downloadable File (DAC NDF), source date June 1, 2026, for PECOS-enrollment-linked gender, provider name, practice ZIP code, and graduation year. Providers must have an individual NPI, a primary taxonomy matching one of fifteen target specialties, and non-null last name, ZIP code, and graduation year between 1960 and 2026. No Medicare FFS claims-activity filter is applied. [1,9,10]

The fifteen specialty cohorts are: chiropractic, physical therapy, podiatry, pain medicine, PM&R, family medicine, psychiatry, nurse practitioner, internal medicine, orthopedic surgery, physician assistant, emergency medicine, optometry, psychology, and social work. This matches the comparator architecture in The Organizational Outlier. The source extract labels the broad 103T*/103G* cohort “Clinical Psychologist” and the 1041* cohort “Clinical Social Worker”; following the Group Size paper, this manuscript uses the more accurate reader-facing labels “Psychologist” and “Social Worker.” [9,10]

2.2 Specialty Classification and NPI Uniqueness

The analytic unit is a unique individual NPI. Specialty assignment is based on the provider's primary NUCC taxonomy in NPPES, identified by the Y primary flag in healthcare_provider_taxonomies; each provider is therefore eligible for one study specialty based on primary taxonomy. This replaces the earlier draft's degree-precedence hierarchy and aligns classification with the Group Size paper's primary-taxonomy method. The exact inclusion logic is: 111N* chiropractic; 103T*/103G* psychology; 1041* social work; 207P* emergency medicine; 207Q* family medicine; 207R* internal medicine; 363L* nurse practitioner; 152W* optometry; 207X* orthopedic surgery; 208VP* pain medicine; 2251* physical therapy excluding 225200000X (PT assistant); 363A* physician assistant; 2081* PM&R; 213E* podiatry; and four specified psychiatry codes (2084P0800X, 2084P0802X, 2084P0804X, 2084P0805X). Internal medicine therefore includes its 207R* subspecialties; pain medicine is intentionally limited to primary 208VP* codes. [9,10]

2.3 Years Since Graduation

Years since graduation is calculated as 2026 minus DAC NDF graduation year (grd_yr), with graduation years restricted to 1960–2026. The source methods report 99.9% completeness and a resulting range of 0–66 years. For presentation, single-year values are grouped into 0–2, 3–5, 6–10, 11–20, 21–30, 31–50, and 51+ years. “Years since graduation” is used rather than “years in practice” because the field does not establish continuous active practice, licensure, or patient care. A DC (or other provider) with a long time since graduation may still hold an active NPI while working substantially or entirely in education, research, administration, or other non-clinical roles; the extract cannot distinguish this from active patient care. This paper describes NPI-registration patterns, not verified clinical activity (see Key Limitations).

2.4 Reported Gender

Gender is the CMS/PECOS gndr field with values F, M, and U. U/unknown is retained rather than excluded or imputed. In the updated extract, 1,594 of 1,807,746 NPIs (0.09%) are coded U. Female and male shares use all NPIs as the denominator unless noted. The field does not represent non-binary gender identities. When compared with Census population data, an additional definitional limitation applies: the Census benchmark is a sex measure, while the CMS source field is labeled gender.

2.5 Estimated Race/Ethnicity Using BIFSG

Race/ethnicity is imputed rather than self-reported. The source implementation applies Bayesian Improved First Name, Surname, and Geocoding (BIFSG), an extension of RAND's two-component Bayesian Improved Surname Geocoding (BISG) that adds first name as a third predictor. RAND reports its indirect estimation approach as 41% more efficient than using surname alone and 108% more efficient than using geocoding alone [12]; this paper's three-component extension has not been separately validated against that specific benchmark. For each NPI, surname, first name, and practice ZCTA contribute race/ethnicity probabilities. Surname and first-name priors come from the U.S. Census Bureau’s 2020 frequently occurring names by race/Hispanic origin; geographic priors now come directly from 2019–2023 ACS Table B03002 (Hispanic Origin by Race) at the ZCTA level, using exact non-Hispanic race rates rather than an approximated geography. All three BIFSG input components (surname, first name, geography) now use identically-defined, mutually exclusive race/ethnicity categories that separate Hispanic origin from race at every stage, eliminating the residual-allocation approximation used in earlier drafts. Unmatched components fall back to national marginal rates, and a 0.001 floor is applied to geographic priors to prevent zero-probability collapse. [2,3,5,6,7,8,10]

Five mutually exclusive modeled categories are reported: White (non-Hispanic White alone), Black (non-Hispanic Black alone), Asian (non-Hispanic Asian plus Native Hawaiian/Other Pacific Islander alone), Hispanic (Hispanic or Latino, any race), and Other (non-Hispanic American Indian/Alaska Native plus non-Hispanic multiracial). The source *_expected fields are sums of provider-level posterior probabilities, not hard class assignments. Percentages are recomputed from summed expected counts divided by summed npi_count. Reader-facing text therefore uses “estimated [group] share,” never individual race assignments. With exact, mutually exclusive geographic priors now in place at every BIFSG input stage, the estimated Other category averages 4–6% across specialties in this extract, consistent with the general population benchmark; this is a material reduction from the 11–15% range produced by the earlier, approximated geographic prior. What has been resolved is the known race/Hispanic-overlap and residual-allocation artifact in the geographic prior specifically; BIFSG's underlying model and matching uncertainty—not estimated by any confidence interval or posterior calibration here—remains and is unaffected by this change.

2.6 Reaggregation and Career-Stage Comparisons

All npi_count and *_expected fields in ProviderGender_Updatedv4.xlsx are additive and were reaggregated by summation. The updated extract contains 1,954 specialty × single-year-since-graduation × gender cells and sums to 1,807,746 NPIs. Race/ethnicity percentages were recomputed as summed estimated counts divided by summed npi_count; row-level percentages were not averaged. Average years since graduation is npi_count-weighted. Gender shares are calculated from summed npi_count by gender with U/unknown retained. [10]

The analysis is descriptive. It compares overall composition, seven presentation bands, and two summary cohorts: ≤10 years and 31+ years since graduation. The phrase “workforce evolution” refers to differences across career-stage cohorts in one 2026 cross-section. It is not a longitudinal panel of the same clinicians, and no causal interpretation is made. No hypothesis tests, confidence intervals, or causal models are applied. For Figure 5, career-stage balance is defined as the share of a specialty—or of a sex-specific specialty subgroup—within 10 years of graduation minus the corresponding share 31+ years since graduation. Positive values indicate greater early-career representation; negative values indicate greater late-career representation.

2.7 U.S. Population Context Benchmark

U.S. population gender/race/ethnicity estimates were derived from The Annual Estimates of the Resident Population by Sex, Race, and Hispanic Origin for the United States: April 1, 2020 to July 1, 2025 (NC-EST2025-SR11H) file [4]. The estimates are based on the 2020 Modified Age and Race Census (MARC) file and include changes from Count Question Resolution, the Post-Census Group Quarters Review, and geographic revisions. The estimates add births to, subtract deaths from, and add net migration to the April 1, 2020 estimates base. Race data in the Vintage 2025 estimates are based on the 1997 Office of Management and Budget Standards for Maintaining, Collecting, and Presenting Federal Data on Race and Ethnicity.

For population estimates methodology statements, see [11]. Responses of “Some Other Race” from the decennial census are modified to be consistent with the race categories that appear in our input data. This contributes to differences between the population for specific race categories shown and those published from the 2020 Census. To learn more about the “Modified Race” process, go to https://www.census.gov/programs-surveys/popest/technical-documentation/research/modified-race-data.html. All geographic boundaries for the 2025 population estimates series are as of January 1, 2025.

The U.S. population is 50.9% female, using the July 1, 2025 estimate. For race/ethnicity, the file's sex, race, and Hispanic-origin detail is mapped to the same five analytic categories as the BIFSG model: non-Hispanic White alone, non-Hispanic Black alone, non-Hispanic Asian alone plus Native Hawaiian/Other Pacific Islander alone, Hispanic (any race), and Other (non-Hispanic American Indian/Alaska Native alone plus non-Hispanic Two or More Races). Unlike the ACS-based benchmark used in earlier drafts, this file's five categories sum to the full population with no residual “Some Other Race” category requiring exclusion or renormalization. The resulting context, using the July 1, 2025 estimate, is White 56.1%, Black 12.3%, Asian/NHPI 6.9%, Hispanic 20.5%, and Other 4.3%. These are population context values, not expected workforce quotas or parity targets. [4]

Results

1. Workforce Size and Career Stage Across Fifteen Specialties

The updated extract contains 1,807,746 NPIs across fifteen primary-taxonomy specialty cohorts. Specialty size ranges from 4,938 NPIs in pain medicine to 426,739 nurse practitioners; chiropractic contributes 43,018. Average years since graduation ranges from 9.4 among nurse practitioners to 24.4 in internal medicine. Chiropractic averages 22.2 years since graduation, with 23.2% of its cohort within 10 years and 27.4% at 31+ years.

Career-stage distributions differ substantially by specialty. Nurse practitioner (66.6%), physician assistant (54.4%), and physical therapy (50.3%) have the largest shares within 10 years of graduation; internal medicine (14.3%), orthopedic surgery (14.9%), and pain medicine (15.9%) have the smallest. These differences reflect training pathways, Medicare enrollment, cohort growth, attrition, and other selection mechanisms; they should not be read as direct measures of entry, retention, or retirement.

Table 2. Specialty-level workforce profile.

Race/ethnicity values are BIFSG-estimated shares. Counts are unique NPIs classified by primary NUCC taxonomy. AI-assisted reaggregation: ChatGPT (OpenAI; GPT-5.6 Sol); pending DAWG verification.

2. Overall Demographic Composition and Population Context

Female representation varies nearly tenfold across the fifteen cohorts: orthopedic surgery is 8.7% female and nurse practitioner 87.0%. Chiropractic is 27.3% female, third-lowest after orthopedic surgery and pain medicine (14.8%). Optometry (52.3%) and family medicine (46.5%) are closest to the 50.9% Census population context on this single dimension, while physical therapy is 57.2%. These comparisons are descriptive; population representation is not a normative parity target.

BIFSG-estimated race/ethnicity also varies markedly. Chiropractic has the highest estimated White share at 82.1%; psychology is 75.3%; social work is 73.7% and orthopedic surgery 73.3%. Internal medicine is lowest at 49.9%, with psychiatry at 53.1% and pain medicine at 55.4%. Internal medicine has the highest estimated Asian share (29.0%), while nurse practitioner has the highest estimated Black share (12.1%). Estimated Hispanic share ranges from 5.3% in orthopedic surgery to 10.5% in psychiatry—below the 20.5% Census contextual benchmark in every specialty.

Figure 1 places female share and estimated White share against the Census population context. It shows why “representativeness” is multidimensional: a specialty can be near the population benchmark on one dimension and far from it on another. Because BIFSG is a probabilistic model and because workforce entry requires profession-specific education and licensure, the population lines should be treated as context rather than expected endpoints.

3. Career-Stage Cohort Change Is Broad but Not Uniform

Across the fifteen specialties, 13 have a higher female share among NPIs within 10 years of graduation than among NPIs 31+ years out. The largest gaps are optometry (+38.1 percentage points), family medicine (+25.5), psychology (+24.6), podiatry (+23.4), and chiropractic (+21.5). Physical therapy (−2.0) and nurse practitioner (−7.5) are the two exceptions; both are already majority-female overall, and the NP 31+ stratum is very small (1.4% of the specialty cohort).

Estimated White share is lower in the ≤10-year cohort than in the 31+ cohort for 14 of 15 specialties. The largest differences occur in podiatry (−21.0 percentage points), optometry (−19.4), psychology (−19.0), nurse practitioner (−14.8), and social work (−14.1). Pain medicine is essentially flat (+0.7). Figure 2 summarizes both dimensions: most specialties move toward more female and less White representation in newer cohorts, but the magnitude is highly specialty-specific.

4. Chiropractic: Demographic Transition From an Unusually Homogeneous Baseline

Among chiropractic NPIs within 10 years of graduation, female share is 38.5% versus 17.0% at 31+ years. Estimated White share is 76.5% versus 86.3%; estimated Hispanic share is 9.1% versus 4.1%; estimated Asian share is 4.1% versus 1.7%; and estimated Black share is 5.0% versus 4.2%. The seven-band profile (Figure 4) is monotonic in the two most visible measures: female share falls from 45.5% at 0–2 years to 17.4% at 31–50 years, while estimated White share rises from 74.4% to 86.2%.

Even after that cohort shift, chiropractic remains the highest-White specialty in the ≤10-year comparison at 76.5%; physician assistant (70.6%), orthopedic surgery (70.3%), and physical therapy (69.7%) are next. The evidence therefore supports a “transition without convergence” interpretation for chiropractic: newer cohorts are materially more diverse and more female than older cohorts, but the profession remains distinct from both the comparator set and the population context.

The Organizational Outlier independently finds chiropractic to be the most structurally distinct of the same comparator specialties on organizational measures: predominantly small, single-specialty, single-state practices with comparatively infrequent claims endpoints. Read alongside that paper, this demographic profile — the highest estimated White share and, historically, the most male composition among the fifteen specialties — describes a third axis, workforce composition, on which chiropractic is also an outlier, even as its newest graduating cohorts are diversifying. This is a descriptive parallel, not a causal claim linking practice structure to workforce composition; the two papers use different cohorts, time frames, and units of analysis, and no formal test of association between them has been performed. [9]

5. Different Specialties Are Evolving Along Different Trajectories

The added optometry, psychology, and social-work cohorts materially broaden the interpretation. Optometry shows the largest female career-stage difference (+38.1 percentage points) and one of the largest estimated White declines (−19.4). Psychology shows +24.6 points female and −19.0 points White. Social work is already 83.6% female overall, so its female cohort difference is modest (+5.6), while its estimated White share is 14.1 points lower in the ≤10-year cohort. These patterns show that “diversification” is not a single trajectory: professions can start from very different baselines and move at different rates across gender and estimated race/ethnicity dimensions.

The physician and advanced-practice cohorts also differ. Family medicine, internal medicine, emergency medicine, and physician assistant all show sizeable increases in female representation in newer cohorts (+19.4 to +25.5 points) along with lower estimated White shares (−6.2 to −8.6). PM&R and psychiatry show smaller White-share differences, while pain medicine is essentially unchanged on estimated White share. Across every specialty, estimated Hispanic share remains below the Census population benchmark, including in the ≤10-year cohort; the highest early-career estimate is 13.9% in psychology. This recurring gap is descriptive and warrants pipeline research, but BIFSG limitations preclude treating it as a precise measure of underrepresentation without direct demographic data.

Figure 3 decomposes the estimated non-White share into its component groups by career stage. Estimated Asian share increases in every specialty except psychiatry; estimated Hispanic share increases most among psychologists, social workers, nurse practitioners, and chiropractors; and the largest increases in estimated Black share occur among nurse practitioners and social workers.

6. Workforce Renewal Differs by Specialty and Sex

Figure 5 compares, within each specialty, the share of clinicians within 10 years of graduation with the share 31+ years since graduation, overall and separately among women and men. Seven of 15 specialties have a positive overall balance. Nurse practitioners (+65.2 pp), physician assistants (+51.4), physical therapists (+39.8), emergency medicine (+21.3), and social workers (+16.2) show positive balances overall and within both women and men. Optometry (+8.4) and PM&R (+0.8) are also positive overall, but their male workforces have negative balances.

Sex-stratified patterns diverge sharply. Female early-career share exceeds late-career share in 14 of 15 specialties; internal medicine (-2.4 pp) is the only exception. Among men, only the five specialties above are positive. Chiropractic illustrates the split: women +15.6 pp, men -11.7 pp, overall -4.2 pp. Thus, increasing representation of younger women does not fully offset the older male career-stage composition.

Discussion

These patterns suggest that workforce change involves both diversification and differential renewal across specialties. A negative career-stage balance does not establish a future shortage: years since graduation is not age or retirement timing, and this analysis does not measure entrants, exits, full-time-equivalent effort, hours worked, geographic distribution, or demand. Where late-career share materially exceeds early-career share, however, the pattern identifies a potential workforce-replacement signal worth monitoring.

In chiropractic, the finding adds nuance to the demographic transition described above. Newer cohorts are more female and more racially/ethnically diverse, yet the overall profession remains older in career-stage balance because the male workforce is disproportionately concentrated in later cohorts. If persistent, that structure could create recruitment and supply pressure even as the profession becomes more diverse. Prospective workforce data would be needed to estimate any future shortfall.

Key Limitations

• Specialty assignment uses primary NUCC taxonomy and is mutually exclusive within the study comparator set. A provider whose relevant specialty appears only as a secondary taxonomy is not captured in that specialty cohort. Primary taxonomy is self-reported in NPPES and may not reflect every credential, subspecialty, or practice role.

• Race/ethnicity is imputed, not self-reported. BIFSG outputs probabilities and estimated counts; individual identity should never be inferred from these data, and aggregate estimates may be biased in a healthcare workforce whose demographic selection differs from the general population used to build priors.

• Race/ethnicity categories are mutually exclusive by construction, but remain model-based. All three BIFSG input components — surname, first name, and geography — now use identically defined, mutually exclusive race/ethnicity categories that separate Hispanic origin from race at every stage, including the geographic prior, which is now built directly from ACS Table B03002 exact non-Hispanic race rates at the ZCTA level rather than an approximated allocation. This resolves overlap at every category-construction stage but does not eliminate model error, calibration differences, or uncertainty arising from names or practice geography.

• The “Other” category is heterogeneous. It combines non-Hispanic American Indian/Alaska Native and multiracial populations. With exact geographic priors now in place, this category averages 4–6% across specialties in this extract — down from the 11–15% range under the earlier, approximated geographic prior — but it remains a composite of distinct groups and should not be interpreted as a coherent demographic group in its own right.

• The cohort is CMS-enrolled, not the full licensed workforce. Providers who do not participate in Medicare are excluded, and enrollment propensity differs by profession, career stage, setting, and possibly demographics. This paper cannot characterize what CMS-non-enrolled individuals are doing — continuing in cash-pay or other non-Medicare clinical practice, moving into education, research, or administration, or leaving the profession entirely — and the career-stage patterns described here should not be read as describing that population.

• Physical therapy is identified from primary 2251* NUCC taxonomy codes with physical therapy assistants excluded. This aligns the profession definition with the Group Size paper and removes the earlier draft's “Physical Therapist in Private Practice” limitation, but the cohort remains limited to clinicians represented in the CMS/PECOS DAC NDF.

• Years since graduation is not age or continuous practice tenure. It does not account for training length, career breaks, delayed entry, retirement, or time outside clinical practice. For physician specialties, very early years may include residents or fellows billing under their own NPI.

• Career-stage balance is not a workforce-supply forecast. Figure 5 compares early- and late-career shares using years since graduation; it does not measure retirement timing, workforce exits, new graduate entry, full-time-equivalent effort, hours, geographic distribution, or demand. Negative balances should therefore be interpreted as potential replacement signals for monitoring, not projected shortages.

• Gender is limited to M/F/U in CMS enrollment data. The field is non-null in the supplied extract, but U means unknown and the source does not capture non-binary identities.

• Cross-sectional cohort contrasts are not longitudinal trends. The observation that newer graduation cohorts are more female or less White in many specialties does not establish that the same workforce changed at a particular rate, nor can it separate cohort replacement from enrollment, migration, attrition, delayed entry, or other selection mechanisms.

• No inferential testing was performed. Differences are descriptive; uncertainty in BIFSG estimated composition is not represented by conventional binomial confidence intervals. The Census population benchmark adds context but is not age-, education-, or occupation-matched to the provider cohorts and should not be interpreted as a parity target.

Practice location is used for geocoding. The BIFSG geography component uses the provider practice ZIP from the DAC NDF, not a residential address; practice location may be less informative than residence for some clinicians.

Name-race priors and the provider roster are from different vintages. The BIFSG name distributions come from the 2020 Census and are applied to a 2026 provider roster; cohort-specific demographic shifts may affect calibration, particularly for newer graduates.

Internal medicine is deliberately broad. The 207R* taxonomy pattern includes general internal medicine and its subspecialties under one study cohort, which can mask demographic differences among subspecialties.

Disclosures and Publication Status

This work is classified as a public-data workforce and health-services analysis under the Arete Research Library Governance Framework; it is not PBRN-sponsored human-subjects research. Arete Healthcare supports independent outpatient practices, including several professions represented in the analysis, and therefore has a professional and commercial interest in workforce composition, recruitment, practice sustainability, and infrastructure. That interest does not alter the requirement that descriptive Findings remain independent of Advocacy / Commentary. The DAWG review is intended to test methods, reproducibility, limitations, and the boundary between findings and Arete interpretation.

AI Use Disclosure: This work used AI-assisted technologies during preparation and quality assurance; AI systems are not authors and are not cited as evidentiary sources. ChatGPT (OpenAI; GPT-5.6 Sol) assisted with source-data reaggregation, methodology reconciliation following the Author's BIFSG v3 geographic-prior re-run, independent computation of the career-stage-by-sex balances presented in Figure 5, and manuscript drafting, organization, and assembly across versions. Claude (Anthropic; Claude Sonnet 4.6 and Claude Sonnet 5) performed independent reproduction of the headline statistics against the source workbook, Style Guide compliance implementation, DAWG reviewer comment incorporation, and drafting support. All AI-assisted calculations, code, factual statements, and manuscript text incorporated into this work were reviewed by human authors against the underlying data, executed queries, or primary sources, as documented in the internal AI Assistance and Reproducibility Record maintained under the Arete Research Library Governance Framework. Human authors and reviewers remain fully responsible for accuracy, methodology, interpretation, and the publication decision.

Recommended citation: Arete Healthcare. A Workforce in Transition: Gender, Career Stage, and Estimated Race/Ethnicity Across Fifteen Health Professions and Specialties. Arete Original Research ARETE-OR-2026-007 (candidate; pending Suite Index confirmation), Version 3.9. Published August 26, 2026. https://doi.org/10.13140/RG.2.2.17854.04160

References

1. Centers for Medicare & Medicaid Services. Doctors and Clinicians — National Downloadable File, Provider Data Catalog. https://data.cms.gov/provider-data/dataset/mj5m-pzi6

2. U.S. Census Bureau. Frequently Occurring First Names and Last Names in the 2020 Census. Released April 14, 2026. https://www.census.gov/topics/population/genealogy/data/2020_names.html

3. U.S. Census Bureau. 2019–2023 American Community Survey 5-Year Estimates, Table B03002: Hispanic or Latino Origin by Race. https://data.census.gov/table/ACSDT5Y2023.B03002

4. Annual Estimates of the Resident Population by Sex, Race, and Hispanic Origin for the United States: April 1, 2020 to July 1, 2025 (NC-EST2025-SR11H). https://www.census.gov/data/datasets/time-series/demo/popest/2020s-national-detail.html

5. Elliott MN, Morrison PA, Fremont A, McCaffrey DF, Pantoja P, Lurie N. Using the Census Bureau's Surname List to Improve Estimates of Race/Ethnicity and Associated Disparities. Health Services and Outcomes Research Methodology. 2009;9(2):69–83. doi:10.1007/s10742-009-0047-1.

6. Voicu I. Using First Name Information to Improve Race and Ethnicity Classification. Statistics and Public Policy. 2018;5(1):1–13. doi:10.1080/2330443X.2018.1427012.

7. Consumer Financial Protection Bureau. Using Publicly Available Information to Proxy for Unidentified Race and Ethnicity: A Methodology and Assessment. 2014. https://www.consumerfinance.gov/data-research/research-reports/using-publicly-available-information-to-proxy-for-unidentified-race-and-ethnicity/

8. Fiscella K, Fremont AM. Use of Geocoding and Surname Analysis to Estimate Race and Ethnicity. Health Services Research. 2006;41(4 Pt 1):1482–1500. doi:10.1111/j.1475-6773.2006.00551.x.

9. Arete Healthcare. The Organizational Outlier: Group Size, Specialty Composition, and Multi-Specialty Scaling of Chiropractic Compared to Fourteen Professions and Specialties. Arete Original Research ARETE-OR-2026-001, Version 3.10. Draft dated August 17, 2026.

10. Arete Healthcare Research & Analytics. ProviderGender_Updatedv4.xlsx and UpdatedMethodsandSQLFromMimilabs_v3.docx. Internal reproducibility/source files supplied August 18–26, 2026.

11. U.S. Census Bureau. Population Estimates Methodology. https://www.census.gov/programs-surveys/popest/technical-documentation/methodology.html

12. RAND Corporation. Bayesian Improved Surname Geocoding (BISG). https://www.rand.org/health-care/tools-methods/bisg.html

Appendix A. Supplemental Comparative Tables

A1. Full Fifteen-Specialty Demographic Profile

White/Black/Asian/Hispanic/Other are BIFSG-estimated shares. They are not self-reported demographic counts. “Psychologist” corresponds to source 103T*/103G*; “Social Worker” corresponds to 1041*.

A2. Chiropractic by Career Stage

Career-stage bands are constructed directly from the single-year CSV: 0–2, 3–5, 6–10, 11–20, 21–30, 31–50, and 51+ years since graduation.

A3. Early- vs Late-Career Cohort Differences Across Fifteen Specialties

Positive gender Δ means higher female share in the ≤10-year cohort. Negative White Δ means lower estimated White share in the ≤10-year cohort. Cross-sectional cohort comparison only.

A4. Female Share by Career Stage Across Fifteen Specialties

Values are percentages within each specialty × career-stage band. Cross-sectional cohorts only; the 51+ cells can be small in some specialties.

A5. Estimated White Share by Career Stage Across Fifteen Specialties

Values are BIFSG-estimated White percentages within each specialty × career-stage band. They are not self-reported race/ethnicity.