Regional physician supply vs. demand — active workforce, population-driven demand, retirement attrition, and residency pipeline output modeled HRSA/AAMC-style
Every workforce forecast starts with an accurate stock count. Health workforce planners combine administrative claims, state licensure boards, and survey data (in the US, chiefly the AAMC Physician Masterfile and the HRSA Area Health Resources File) to establish how many physicians are actively practicing, in which specialties, and where. This baseline — headcount and full-time-equivalent (FTE) supply, broken out by region and specialty — is the anchor against which every future projection is measured.
Workforce supply models are built as stock-and-flow systems, borrowed from systems dynamics and demography:
• Stock: the current count of active, licensed, patient-facing physicians at time t=0 — reported as both headcount and FTE (a physician working 0.6 clinical FTE contributes 0.6, not 1.0, to capacity) • Inflows: new residency/fellowship graduates entering practice, international medical graduates (IMGs) obtaining licensure, physicians re-entering after leave • Outflows: retirement, death, disability, career change, relocation out of the modeled region • Geographic unit of analysis: HRSA typically models at the state or Health Service Area (HSA) level; sub-state models use county clusters or Health Professional Shortage Area (HPSA) boundaries
HRSA's Health Workforce Simulation Model (HWSM) and AAMC's complementary supply-demand model both start from this stock, then project forward year-by-year applying flow-rate assumptions. The choice of geographic and specialty granularity determines how actionable the resulting forecast is — a national aggregate can mask a severe rural obstetric-care shortage entirely.
Aggregate physician-to-population ratios routinely conceal severe local shortages. The US averages roughly 1 active physician per 320 residents nationally, but rural counties average closer to 1 per 660 — and nearly 20% of the population lives in areas HRSA formally designates as Health Professional Shortage Areas for primary care.
Specialty maldistribution compounds the geographic problem: psychiatry, general surgery, and obstetrics show the steepest rural drop-off, since these specialties depend on hospital infrastructure, call coverage, and patient volume that small rural facilities cannot always sustain. A regional forecasting model must therefore stratify supply along two axes simultaneously — geography and specialty — rather than reporting a single blended shortage number.
Demand forecasting is the harder half of the equation. Two broad methodologies dominate the field: utilization-based models, which extrapolate current per-capita patterns of care use forward with population change, and needs-based models, which start from epidemiological disease prevalence and clinically recommended care intensity. Most modern regional forecasts — including HRSA's — blend both, since utilization alone bakes in existing access barriers as if they were acceptable steady states.
Utilization-based models project future demand as: Demand(t) = Population(t) × current utilization rate per capita, often stratified by age band, sex, and insurance status. They are simple and defensible but silently assume today's access patterns — including today's unmet need — are the correct target.
Needs-based models instead estimate the clinically appropriate volume of care for a population's disease burden, using epidemiological prevalence data and evidence-based care guidelines, independent of what is currently being delivered. They tend to show larger shortages, because they surface currently-unmet need rather than normalizing it.
Benchmarking models, a third approach, compare a region's physician-to-population ratio against a high-performing reference region (e.g., the best-resourced HSA) and treat the gap as unmet demand. HRSA's HWSM primarily uses a hybrid utilization-need model, adjustable by specialty.
The choice of demand methodology alone can swing a projected 2034 shortage estimate by tens of thousands of physicians — AAMC's published range of 37,800 to 124,000 total physicians short by 2034 reflects exactly this modeling sensitivity, not measurement error.
Two structural forces reliably push demand curves upward faster than population counts alone would suggest:
• Age-driven utilization: patients aged 65+ use roughly 2–3× the physician visits per year of patients under 45, and this cohort is the fastest-growing age bracket in most developed-economy regions. A region with flat population growth but a rapidly aging pyramid can still see demand rise 15–25% over a decade. • Multimorbidity: the prevalence of two or more concurrent chronic conditions (diabetes, hypertension, COPD, heart failure) rises steeply with age and drives disproportionate demand for both primary care continuity and specialist referrals.
Regional forecasting models apply an age-standardized utilization multiplier — recalculating projected demand as the regional population pyramid shifts, not just as raw headcount grows. This is why a shrinking-but-aging region can face a growing physician shortage even while its total population declines.
Supply does not sit still while demand grows. Physicians retire, reduce clinical hours, change careers, or leave practice due to burnout — and the current physician workforce skews older than most professions, because training takes over a decade from matriculation to full practice. Attrition modeling applies age-specific exit probabilities to the current stock, and recent survey data shows those probabilities have been rising, not falling.
Attrition models assign each physician an annual exit probability as a function of age, typically low (<1%) below age 55, rising steeply past 60, and exceeding 8–10% per year past 70. Because roughly one in five active US physicians is already 65 or older, this age structure alone implies a sustained, predictable wave of retirements over the coming decade — a "retirement bulge" that regional models can forecast with reasonable confidence even without any external shock.
A simplified attrition-only projection compounds this exit rate against the current stock: Supply(t) = Supply(0) × (1 − attrition rate)^t. Absent any inflow, a 2.6% annual attrition rate alone would shrink a baseline workforce by over 20% across a 10-year horizon — which is why the training pipeline (Stage 4) is not optional context, but a load-bearing part of every credible forecast.
Rural and primary-care physicians are disproportionately older than the specialist workforce concentrated in urban academic centers, meaning attrition-driven shortages tend to hit already-underserved regions first and hardest — a compounding, not merely additive, maldistribution effect.
Traditional workforce models treated attrition as almost purely age-driven. Since 2020, forecasters have added a burnout-adjusted term: survey-measured burnout and intent-to-leave rates now materially raise near-term exit probabilities independent of age, particularly among mid-career physicians (ages 40–55) who would otherwise be a decade or more from retirement.
National physician surveys (AMA, Medscape, and academic burnout studies) have repeatedly found that roughly half of practicing physicians report at least one core burnout symptom, and a meaningful share report actively planning to cut clinical hours or leave the profession within one to two years. Regional models now commonly apply a burnout multiplier on top of the baseline age-attrition curve, especially for specialties and settings (emergency medicine, primary care, rural hospitals) where reported burnout runs highest.
The only durable offset to attrition is new physician supply — and that pipeline is long and structurally constrained. From medical school matriculation to unsupervised practice takes 7 to 15 years depending on specialty, and in the US the number of Medicare-funded residency training slots was effectively frozen for over two decades, creating a bottleneck between a growing pool of medical graduates and the number of accredited positions available to train them.
Unlike medical school capacity — which teaching hospitals and universities can expand relatively flexibly — residency training slots are gated by Medicare Graduate Medical Education (GME) funding, which historically pays for the bulk of US residency positions. The 1997 Balanced Budget Act capped the number of Medicare-funded residency slots per hospital at 1996 levels, and that cap held for roughly 24 years even as medical school enrollment grew by around 40% over the same period.
The result was a widening funnel mismatch: more MD and DO graduates each year, competing for a nearly fixed number of accredited residency positions. The Consolidated Appropriations Act (2021, 2023) began phasing in the first material expansion in a generation — roughly 1,200 new Medicare-supported slots added in stages — but workforce economists broadly agree this expansion is still smaller than needed to close projected shortfalls on its own.
Because residency slots — not medical school seats — are the binding constraint, simply admitting more medical students without expanding GME funding does not increase the number of new practicing physicians; it only increases competition for a fixed number of training positions.
Every policy lever pulled today — expanding residency slots, funding new medical schools, offering rural loan repayment — takes years to show up as practicing physicians. Primary care residencies run 3 years post-MD; many surgical subspecialties run 5–7 years of residency plus 1–3 years of fellowship, meaning a policy decision made this year may not add meaningfully to active supply until the mid-2030s.
Specialty choice within the pipeline also does not automatically follow population need: graduates gravitate toward specialties with better compensation, controllable lifestyle, and urban academic placement, while primary care, psychiatry, and rural general surgery — the specialties with the steepest projected shortages — consistently fill a smaller share of available slots. International medical graduates (IMGs), who fill a substantial share of primary-care and rural residency positions in the US, are a critical but visa-policy-sensitive component of this inflow.
The final step combines every prior stage into a forward projection: run the supply trajectory (baseline stock, minus attrition, plus pipeline inflow) and the demand trajectory (population growth, aging, disease burden) forward together, and read off the gap at each future year. Because every input carries real uncertainty, credible forecasts are always published as a scenario range, not a single number — and are explicitly built to be re-run against policy levers.
A regional forecast is not one line on a chart — it is a family of lines generated by varying the key assumptions: population growth rate, per-capita utilization trend, attrition rate, and residency inflow rate. Running the model across low/medium/high assumptions for each driver produces a projected shortfall range rather than a false-precision point estimate.
Sensitivity analysis identifies which assumption the projection is most exposed to. In most regional physician models, the projected gap is more sensitive to the residency-inflow assumption than to the population-growth assumption — meaning policy has more real leverage over the outcome than demographics do, which is precisely why these models are built to be interactively adjustable rather than static reports.
Workforce planners generally group interventions into four categories:
• Expand the pipeline: add Medicare-funded GME slots, open new medical schools in underserved regions, streamline IMG licensure and visa pathways (e.g., Conrad 30 J-1 waiver programs) • Improve retention: loan repayment and scholarship programs tied to service in shortage areas, burnout-reduction investment, flexible/part-time practice models to retain late-career physicians longer • Redistribute the existing supply: rural residency tracks, telehealth to extend specialist reach into shortage areas, locum tenens and regional staffing pools • Expand the care team: broaden scope of practice for nurse practitioners and physician assistants, team-based care models that shift lower-acuity visits off physician panels
Each lever has a different cost-per-physician-year-equivalent and a different time-to-impact — GME expansion is slow but structural; scope-of-practice and telehealth changes can shift effective capacity within a single budget cycle.
Modeling residency-slot expansion and scope-of-practice offsets together typically closes a projected regional gap faster than either lever alone — the two act on different timescales (10+ years vs. under 2 years), making them complements rather than substitutes in a policy portfolio.
No physician workforce forecast should be read as a precise prediction. Key sources of irreducible uncertainty include: unmodeled shocks (pandemics, sudden immigration-policy changes), the pace of technology substitution (AI-assisted diagnostics, expanded telehealth), unpredictable shifts in medical graduates' specialty and geographic preferences, and the fact that "demand" itself is partly a policy choice — expanding insurance coverage or Medicaid eligibility can increase realized demand independent of any demographic trend.
Good practice, followed by HRSA and AAMC alike, is to publish a range, update the model annually against realized data, and treat the forecast as a decision-support tool for where to direct GME funding, loan-repayment dollars, and scope-of-practice reform — not as a fixed prophecy of the year 2036.