Health Workforce Forecasting Models

A health workforce forecasting model projects how many providers of each type a health system will have, and how many it will need, over a planning horizon of typically three to ten years. It does this by modeling the workforce as a moving system: a current stock of providers, inflows that add to it (graduates, recruitment, migration, return from leave), outflows that reduce it (retirement, attrition, emigration, reduced hours), and a demand side driven by the population and the services it uses. The output is a gap, by role, place, and year, with the assumptions that produced it.

Most health systems in Canada and the United States do not have one. Health Workforce Canada's 2024 review of provincial modelling and forecasting found capacity uneven and, in many jurisdictions, immature: models built for a single policy question, maintained by one analyst, or commissioned once and never refreshed. The consequence is visible in the news cycle: ratio mandates, agency-nursing audits, and restructurings that all arrive as surprises to systems that had the data to see them coming.

A forecast is not a prediction of one number. It is a documented argument about direction, magnitude, and which decisions change the outcome.

The three model families

Supply models (stock-flow)

The workhorse. Start with the current headcount or FTE by provider type, age band, location, and employment status. Apply annual inflow and outflow rates by cohort. Project forward. Because the rates are applied by cohort, an aging workforce produces a retirement wave the model can see years out. The method is transparent, auditable, and buildable in a spreadsheet for a first version, which is why it is where most systems should start.

Demand models: utilization-based and needs-based

Utilization-based demand projects forward the services a population currently uses, adjusted for demographic change. It is simple and defensible but bakes in today's access problems. Needs-based demand estimates the services a population should receive given its health status and evidence-based levels of care, then derives the workforce required to deliver them. It is harder, requires clinical input on productivity and models of care, and is the approach recommended in the Canadian literature. The two are compared in detail in Needs-Based vs. Utilization-Based Planning.

Gap and scenario models

Supply minus demand, by year, under alternative futures. Scenarios are where the model earns its keep: what happens to the nursing gap if a ratio mandate lands in 2027, if retirement accelerates by two years, if a new tower opens, if international recruitment halves? Each scenario is a changed assumption with a traceable effect, which is what a board or a ministry needs in order to choose.

What goes into a credible model

Provider taxonomy that matches how the organization actually manages people, not just the regulatory categories. Nurse practitioner and RN are different lines; casual and full-time are different lines.
Cohort structure by age or tenure, because attrition and retirement are not uniform.
Flow rates from the organization's own history, validated by clinical and HR leads. National averages are a placeholder, not a basis.
Geographic resolution sufficient to see where the gap is, since a provincial surplus and a regional shortage can coexist.
Explicit assumptions, each in its own cell or parameter, dated, sourced, and owned.
A refresh cadence and a process for comparing last year's forecast with what actually happened.

How we build one

Frame the decisions

List the decisions the forecast must inform in the next 24 months (hiring targets, seat expansion, ratio implementation, capital plans). The model is scoped to those, not to everything.

Establish the data foundation

Inventory sources (HR, payroll, scheduling, registry, ADT), reconcile definitions, and build the extracts. See what data a workforce plan needs.

Build the supply baseline and flows

Current stock by cohort; three to five years of inflow and outflow history; rates validated with operational leads who know why last year looked the way it did.

Build demand

Utilization-based first (fast, defensible), needs-based for the priority services where access is the policy question.

Run the base case and scenarios

Base, plus three to six scenarios tied to the decisions from step 1. Present ranges, not points.

Operationalize

Automate the extracts, document the parameters, assign an owner, set the refresh cadence, and schedule the forecast-versus-actual review.

Reading the output

A useful forecast answers four questions for each provider group: in which year does the gap open or close; how large is it relative to the workforce (a 4 percent gap is a recruitment problem, a 20 percent gap is a model-of-care problem); which assumptions is the result most sensitive to; and which decisions available now change the trajectory. A forecast that produces a single number without those four answers will be argued with, and should be.

Sources

Health Workforce Canada. State of Health Workforce Modelling and Forecasting in Canada. 2024. healthworkforce.ca
Methods for health workforce projection model: systematic review and recommended good practice reporting guideline. Human Resources for Health, 2024. biomedcentral.com
McMaster Health Forum. Rapid Synthesis: Exploring Models for Health Workforce Planning. 2019. mcmasterforum.org
Canadian Health Workforce Network. Health Workforce Planning. hhr-rhs.ca

Questions leaders ask about forecasting models

What horizon should a health workforce forecast cover?

Three to ten years. Under three years the forecast is mostly a budget; beyond ten the assumptions dominate the result. Most organizations run a five-year base case with annual refresh.

Can this be built in Excel?

A first stock-flow model, yes, and it should be, because a spreadsheet is auditable by the people who have to trust it. The data pipeline that feeds it and the scenario layer usually move to a proper analytics environment in the second year.

How accurate are these models?

Accuracy is measured by comparing last year's projection with actuals, which most organizations never do. Well-maintained supply models are typically within a few percent at one year and improve as flow rates are recalibrated. Demand models carry more uncertainty, which is why scenarios matter more than point estimates.

Who should own the model inside the organization?

A named analyst in workforce planning or HR analytics, with a clinical and a finance counterpart who validate assumptions annually. If nobody owns it, it stops being run within two years.

How Health Workforce Forecasting Models Work