The evidence

We didn't wait for a pilot.
We ran the population.

A live pilot measures one building on one good month. Before we ever touch a real ward, we run hundreds of de-identified synthetic patients — generated by our own engine and validated against the openly published 1-million-patient synthetic set — straight through the OS. The output is a gap assessment: what the system would have caught, prioritized, with the action beside it.

No PHI, no permission slip

Synthetic patients carry no real identities — so we can demonstrate the OS end-to-end without exposing a single person's data.

Reproducible & shareable

The same cohort runs anywhere, every time. A finding you can re-run is a finding you can trust — and hand to a reviewer.

A population, not a building

Hundreds of trajectories across conditions and settings — breadth a single-site pilot structurally cannot give you.


The patient generator

Validated against a million.

Our generator engine builds clinically coherent synthetic encounters, calibrated and validated against Synthea's openly published 1-million synthetic-patient set — the same lineage of data peer-reviewed and used across the research community. We then map those encounters into the OS data fabric and let the bundles reason over them exactly as they would over a live hospital.

See the literature
1M
synthetic-patient set used for validation
100s
de-identified patients run through the OS
0
real patients exposed to demonstrate it
re-runs — the cohort is fully reproducible

Sample output

A gap assessment, generated.

An excerpt of what the OS returns after a synthetic-cohort run: findings the fabric surfaced, ranked by priority, each with the suggested action and the bundle that owns it.

Gap Assessment · Synthetic Cohort v3
n = 420 synthetic encounters · medical-surgical + ED · generated by nOS · read-only
P1 criticalP2 elevatedP3 monitor
P1

Indwelling catheters retained past evidence-based duration

Owl · governance

Finding. Foley catheters held beyond a documented indication in 18% of qualifying synthetic encounters — the leading modifiable CAUTI driver in the cohort.

Suggested action. Enable the day-2 Foley necessity review gate; auto-flag retained lines to the bedside nurse. Impact: fewer catheter-days, lower CAUTI risk, avoidable-harm cost recovered.
P1

Sepsis bundle elements completed late

Eagle · Flock

Finding. One or more 3-hour bundle elements (lactate, cultures, antibiotics) completed outside target in 23% of modeled sepsis presentations.

Suggested action. Surface a live bundle clock in Eagle and a gated task in The Flock at recognition. Impact: shorter time-to-antibiotics; mortality is time-dependent.
P2

Admitted patients boarding in the ED

Eagle · throughput

Finding. Admitted patients boarded > 4 hours in 31% of ED-to-inpatient transitions — concentrated at the discharge-lounge hand-off.

Suggested action. Route the bottleneck to Eagle's locator; trigger the discharge-lounge workflow earlier. Impact: recoverable bed-hours, located and priced.
P2

Medication reconciliation incomplete at admission

Flock · safety

Finding. Admission med-rec unverified at the ordering step in 14% of encounters — an adverse-drug-event exposure.

Suggested action. Gate the first inpatient order on a completed med-rec task. Impact: fewer reconciliation errors and ADEs.
P3

Modeled shifts below the resilience floor

Flock · the slack is sacred

Finding. 9% of modeled shift configurations fell below the computed unit safety floor — the conditions that precede error and turnover.

Suggested action. Enforce the staffing refusal gate; a trade that breaches the floor is refused, with evidence. Impact: safer floors, lower turnover cost.
Illustrative excerpt from a synthetic-cohort run. Percentages describe this synthetic population, not any real facility. On a partner engagement, the same assessment runs against your own streams — every figure carrying its source and confidence.

Outcomes & Opportunities

Calibrated by clinicians, in the open.

Once a year, clinicians from our member facilities sit down and peer-review the evidence base behind the engine — challenging the assumptions, checking them against the floor, and recalibrating the digital twin that the gap assessments run on. The model isn't a black box we defend; it's a shared instrument the field keeps honest.

We name the meeting the way the product names everything: not "incident review." Outcomes and opportunities.

👥

Clinician peer review

Member-facility clinicians challenge and validate the evidence base — annually, on the record.

🧬

Digital-twin calibration

Their input tunes the model the synthetic cohorts and gap assessments run against.

🔍

Governed, not asserted

The evidence base earns its authority from the people who practice it.

When the guidance changes

New standard drops. We update. Your gap assessment already knows.

When a governing body releases new guidance, we fold it into the model — and the next time you open your gap assessment, it shows you exactly what's now out of compliance, the action to close it, and the deadline you're working against. Keeping current stops being a project you staff and becomes a state the system maintains.

Owl owns this
New guidance ingested · gap delta
Updated sepsis bundle timing3 units now non-compliant · action assigned
due 14d
Revised restraint renewal intervalrenewal gate retuned · documentation flagged
due 30d
Hand-hygiene audit cadencealready met — no action
on track
Illustrative. Each item carries the standard, the action, the owner, and the deadline.

Beyond our own walls

A control population the whole field could share.

Most clinical AI is trained and validated on the same handful of real datasets — MIMIC above all, a single academic medical center's critical-care records. It's invaluable, but when everyone benchmarks on one hospital's data, the field inherits one hospital's blind spots.

Our engine offers something complementary: a reproducible, privacy-free synthetic control population that any prospective study can hold constant — to baseline an intervention, stress-test a model, or compare across sites without moving a byte of PHI. Synthetic cohorts already have a peer-reviewed precedent as research and benchmarking instruments; we make one purpose-built for operational and coordination questions.

🧪

A constant control arm

Hold a known synthetic population fixed while you vary the intervention.

🔁

Break the data monoculture

An alternative to everyone validating on the same single-center records.

🤝

Shareable across sites

No IRB tangle to exchange a synthetic cohort — collaboration without exposure.

The literature

Synthetic patients are an established research instrument.

We didn't invent the idea of synthetic clinical data — we built a purpose-tuned engine on a peer-reviewed foundation. A selection of references on the generation, validation, and research use of synthetic patient data is below; the full Digital Pilot white paper carries the complete reference list, the reduction- and baseline-rate sources, and the full metric appendicesdownload the PDF ↓.

  1. Walonoski J, Kramer M, Nichols J, et al. Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record. Journal of the American Medical Informatics Association. 2018;25(3):230–238. doi:10.1093/jamia/ocx079
  2. Chen J, Chun D, Patel M, Chiang E, James J. The validity of synthetic clinical data: a validation study of a leading synthetic data generator (Synthea) using clinical quality measures. BMC Medical Informatics and Decision Making. 2019;19:44. doi:10.1186/s12911-019-0793-0
  3. Walonoski J, Klaus S, Granger E, et al. Synthea™ Novel coronavirus (COVID-19) model and synthetic data set. Intelligence-Based Medicine. 2020;1–2:100007. doi:10.1016/j.ibmed.2020.100007
  4. MITRE Corporation. SyntheticMass — a publicly available synthetic dataset of ~1 million synthetic residents of Massachusetts, generated with Synthea. synthea.mitre.org
  5. Johnson AEW, Pollard TJ, Shen L, et al. MIMIC-III, a freely accessible critical care database. Scientific Data. 2016;3:160035. (referenced for contrast — widely-used single-center real data)

References support the generation, validation, and research use of synthetic patient data generally. They do not constitute an endorsement of Nightingale OS. The complete citation list and metric appendices appear in the full Digital Pilot white paper (PDF) ↓. Every projected figure is modeled, not observed.

Run your own cohort

Bring your data. Or borrow our population.

We'll generate a gap assessment on a synthetic cohort, then on yours — and show our work at every step.

Request access

Want the macro case first? Read The Unsustainable System — our investor market analysis →