Optimization
Test patient flow before you pay for more beds.
We think a health system should test how patients move through the beds and staff it already has before it adds more. A full hospital may be short of discharges rather than beds, and hiring cannot close every gap when health workers are scarce. WHO projects a shortfall of 11 million health workers by 2030, mostly in low- and lower-middle-income countries, and WHO and the World Bank estimate that 4.6 billion people still lack access to essential health services.
How can AI help hospitals plan capacity?
AI helps when it forecasts demand by the hour and the hospital tests its response in a model before acting. We would forecast admissions and emergency visits, simulate how patients move through wards and operating rooms, and use optimization to build staff rosters and surgery schedules that meet the forecast at the lowest cost. Clinicians and managers review every plan, and a named official signs off.
Crowding is treated as a bed shortage
When a hospital runs out of beds, planned operations are postponed and emergency patients wait. When it staffs for a busy day that does not come, money goes to idle shifts instead of care. A patient who is ready to leave but waiting for a community care place keeps a bed from the next admission.
Health systems can get more from what they already spend. The IMF estimates that public health spending has fallen short of the best results possible with the same resources by about 26% in advanced economies, 28% in emerging markets and 32% in low-income developing countries.
What we would model, and with which data
Discharge data matters as much as bed counts.
- Open health statistics, such as WHO's Global Health Observatory.
- Population projections for each catchment area, from national statistics offices.
- Hospital records of admissions, discharges, transfers, bed occupancy and operating room use, with timestamps.
- Staff rosters, skill mix and leave.
- Community care and discharge data, which show why beds stay occupied.
FHIR is a standard for exchanging health records, published by HL7, and DHIS2 is a health information platform that its developers say is used in more than 80 low- and middle-income countries.
How the model is built and checked
Demand forecasts give expected admissions by day and hour, as a range. A queueing simulation, a model that replays how patients arrive, wait and move between units, tests a change such as an extra discharge round before anyone makes it. Optimization then builds rosters and surgery schedules that meet the forecast within labor rules and skill mix. Synthetic patient records, which look realistic but belong to no one, let us build and test the model before any real data is shared. Back-testing replays a past year and checks that the model reproduces its known occupancy and cancellations.
What decision-makers get
- Health ministries get capacity plans by area, with the beds, staff and services each demand scenario needs.
- Hospital managers get weekly forecasts with recommended staffing and bed plans, and the expected effect on waits.
- Private hospital groups and insurers get the same view across their networks, including where a new clinic would cut waits most.
A first project picks one hospital and one pathway, such as emergency admissions or planned surgery, where the effect of a change can be measured. We would build the model from synthetic data and a sample of records, back-test it on a past year and hand over a scorecard. The hospital then decides whether to continue.
How it runs safely
A capacity model plans beds and shifts. It never admits, discharges or triages a patient.
- Clinicians make clinical decisions, and a named manager signs off capacity plans.
- Patient data stays in the provider's systems, and the model uses synthetic data wherever it answers the question.
- Fairness checks compare forecast errors and waits across areas and patient groups.
- Health data law applies. HIPAA governs health information privacy in the United States, and the GDPR treats data concerning health as a special category in the EU. The EU AI Act also classes AI for emergency patient triage as high-risk. This page is not legal advice.
Open-source building blocks
Open tools let a health system keep its model and let others check it.
- DHIS2, a platform for collecting, managing and visualizing health data.
- HAPI FHIR, a Java implementation of FHIR for health data clients and servers.
- Ciw, a Python library for simulating networks of queues.
- OR-Tools, Google's optimization toolkit, which includes a worked nurse scheduling example.
- Synthea, which generates synthetic patient populations.
- StatsForecast, for demand forecasting.
Has extendfuture built this for a hospital or health ministry?
No. Optimization is a new extendfuture offering. Our production work so far is AI for companies, with accuracy measured as a number, human review and an audit trail.
Do you need patient records to start?
No. A first model can start from published statistics and synthetic patients. Real records come later, inside the provider's systems, limited to the fields the question needs.
Can private hospitals and insurers use this?
Yes. A hospital group faces the same questions across its sites, and an insurer can test where added capacity would cut waits for its members.
How does this connect to staff shortages?
A capacity plan states how many staff with which skills it needs. That number feeds workforce and training plans. WHO names the mismatch between education and employment strategies as one cause of shortages.
Talk to the people who build.
One call. An honest read on what AI can do for this, and the number it has to beat.