Where every number comes from

The five evidence streams

Our research dossier has five parts, each addressing one of the four critiques we level at HRSA's headline:

1. Alternative projections (Part A)— parallel workforce models showing HRSA's number is on the low end of the reasonable range.
2. Need vs. utilization (Part B) — the clinical literature showing that observed care use is a rationed floor, not a ceiling.
3. AI & technology (Part C) — empirical studies of whether AI actually expands clinician capacity, and whether it drives new demand.
4. Staffing ratios & care models (Part D) — policy evidence on how ratio mandates, scope expansion, and team-based care change the FTE denominator.
5. Financial impact (Part E) — unit economics of shortage: revenue loss, premium labor, turnover, preventable admissions, GDP drag.

Part A — Alternative projections

Four headline numbers you should know:

• HRSA (the baseline we're critiquing): 187,130 physician shortfall by 2037, 87,150 in primary care. HRSA NCHWA Physician Projections Factsheet.
• AAMC: 13,500 to 86,000 physician shortfall by 2036. Similar methodology (microsimulation, utilization-anchored demand), different assumptions about GME expansion and retirement timing. AAMC 2024 Complexities of Physician Supply and Demand.
• Mercer:100,000+ healthcare worker shortfall by 2028 — only two years out. A near-term projection already at the 100k mark suggests HRSA's 2037 number compounds understatement. Mercer Future of US Healthcare.
• McKinsey:200,000 to 450,000 RN gap by 2025 using intent-to-leave survey methodology instead of HRSA's trend-based attrition model. McKinsey Nursing Workforce.

What this tells us:there is no methodological consensus that HRSA's headline is the right answer. The reasonable range across serious projections spans roughly 1x to 5x HRSA. Our platform lets you pick your own assumptions and see where that lands.

Part B — Need is not utilization

HRSA's HWSM Technical Documentation Ch. I says this directly:

“Workforce demand is defined as the number of health care workers required to provide a level of services that will be utilized... As discussed later, demand is different from need. Demand reflects the level of care that people are likely to use, while need is usually a clinical definition.”

Now ask what the clinical literature says about how much care people actually need:

• Preventive services: Only 8% of US adults receive all high-priority, appropriate clinical preventive services (Borsky et al., Health Affairs 2018).
• Hypertension: 20.7% of hypertensive adults have their BP controlled; ~100 million Americans walk around with uncontrolled hypertension (CDC NCHS Data Brief 511, 2024).
• Mental illness: 22.8% of adults had any mental illness; only 23% received treatment (SAMHSA NSDUH 2023).
• Underinsurance: 23% of insured adults are underinsured; 57% of underinsured skipped care due to cost (Commonwealth Fund 2024 Biennial Survey).
• Maternity care: 35% of US counties are maternity care deserts; 1,104 counties have zero obstetric clinicians (March of Dimes 2024).

How we use this:extension E1 (Need-Based Baseline) multiplies HRSA's utilization-anchored demand by a per-profession factor derived from the best source for that profession. Primary Care defaults to 2.5×; Behavioral Health to 4×; Oral Health to 1.8×; Women's Health to 1.6×. Child & Adolescent Psychiatry is at 5× — the worst need gap in the entire model.

Part C — Does AI actually expand supply?

This is the most contested part of the dossier, and we try to be honest about it:

• Kaiser Permanente: deployed ambient AI scribes to 7,260 physicians; saved ~16,000 documentation hours per year; 88% of physicians reported positive impact (JAMIA 2024).
• NEJM AI (null finding): a longitudinal study of DAX at a major academic medical center found that ambient AI scribes did not make clinicians as a group more efficient on measurable EHR or financial metrics — even as physicians reported reduced cognitive burden (NEJM AI 2024).
• Jevons paradox: efficiency gains in diagnostic imaging have historically been absorbed by demand growth; CT use has roughly doubled in 15 years as scanners got faster (Lancet Digital Health 2025).

How we use this: extension S1 (AI Productivity) defaults to +10% effective supply — the midpoint of the literature range — and goes down to 0 in the slider to reflect the NEJM null finding. Extension E4 (AI Demand Multiplier) lets you add a demand uplift from AI-enabled access, modeling the Jevons effect. Turn both on and the net impact on the gap can be neutral or even negative (supply expansion offset by induced demand).

Part D — Staffing ratios and care models

• California AB 394: mandatory minimum RN:patient ratios (1:5 med-surg, 1:2 ICU) since 2004. Spetz et al., Health Affairs 2013 estimates nationwide adoption would lift RN demand ~17%.
• Oregon HB 2697: 2023 follow-up legislation, now in effect. Evidence that ratio laws can pass in non-CA political environments.
• Full-practice authority: 27 states + DC grant NPs full practice authority; the AANP state-by-state tracker is the canonical source.
• Team-based care: Bodenheimer & Smith, Health Affairs 2013, estimates 10-25% reduction in physician hours per panel member with full team leverage (MAs, scribes, RN care managers, clinical pharmacists).

How we use this: extension S3 defaults RN demand to +17% at 2030 with a linear ramp from 2023. S2 defaults NP absorption to 15% of physician visits. S4 defaults team leverage to 8% reduction in physician demand. Every slider has a source tooltip.

Part E — Financial impact

• Hospital labor share: Labor represents 56-60% of hospital operating expense; Kaufman Hall 2024 reported labor at 83.9% of total expense in Q2 (Kaufman Hall National Hospital Flash Report).
• RN turnover: 16.4% annual rate; fully-loaded replacement cost $61,110 per RN; average hospital loses $4.75M per year (NSI 2025 National Health Care Retention Report).
• Physician subsidy: Average hospital subsidy per employed physician: $306,792 (Kaufman Hall 2024).
• Preventable admissions: AHRQ Prevention Quality Indicators: ~4.8 million avoidable hospitalizations per year in the US, costing $25-34 billion.
• GDP drag: McKinsey Global Institute 2020 estimates poor health reduces global GDP by ~15% per year and equity improvements could add $12 trillion by 2040 (MGI Prioritizing Health).
• Deloitte health inequity: US health inequity costs $320B today, projected to exceed $1T by 2040; equity improvements could add $2.8T to GDP by 2040 (Deloitte 2024).

How we use this: our Layer 3 financial model uses per-profession-group revenue per FTE (from MGMA + Kaufman Hall), 6% operating margin (blended), +$75k/year premium labor per shortage FTE (from Staffing Industry Analysts), AHRQ PQI rates for preventable hospitalizations, and a 2.2x Keynesian multiplier on direct expenditure + 1.8x on outcome costs for GDP drag (bracketed by Milken and Deloitte estimates).

What the evidence does NOT say

A few honest caveats:

• The need-based multipliers are not consensus. Clinical societies like ACP and USPSTF publish guidelines, but they don't translate those directly into FTE requirements. Our multipliers are our best reading of the literature; a peer-reviewed methodology paper would need to derive them more formally.
• AI productivity is actively contested. The NEJM AI null finding is a real result that deserves weight. Our 10% default is a midpoint, not a consensus.
• The GDP drag multiplier is not a causal estimate. It's a Keynesian-style aggregation that sits within the Milken and Deloitte ranges. For policy advocacy, the direct health-system P&L numbers are more defensible than the macro figure.
• State-level is noisier than national.HRSA's state projections carry more uncertainty than the national numbers, and our extensions inherit that uncertainty.

How to engage with the evidence

Every extension knob in the scenario builder has a citation link that goes directly to the primary source. If you disagree with a default, dial it to what you believe — the permalink is self-documenting about which assumptions produced which number.

← Back to the one-page brief