Methodology
How Pare decides what to open.
The engine under the autonomous pre-op agent — deterministic, explainable, and audit-ready. It recommends supplies only; it never advises on clinical care.
Scope of use. Pare is a supply-chain analytics tool, not a clinical decision support system. It generates supply-cart setup recommendations for the scrub technician. It does not advise surgeons or modify clinical judgment.
Simulated overnight run on real engine output · supply staging only · every plan reviewed by your team
What the agent produces
For every case, each item on the procedure's candidate card is placed in exactly one of three tiers. The scrub tech works from that plan at 6 a.m. — a tier and a plain-language reason for every item.
Opened and prepared at the start of the case.
Brought to the room, kept sealed, opened only if it is asked for.
Not brought to the room for this case.
Items on the emergency list are always placed in Open Now, regardless of anything else the data suggests. See the safety layer below.
How a recommendation is formed
The starting point for each item is the operating surgeon's own history for that procedure — how often they have actually opened it. When that personal history is sparse, it is shrunk toward the cross-surgeon average for the procedure (Bayesian shrinkage), so a surgeon with only a handful of cases inherits the procedure's norm until their own pattern is established.
The patient's clinical profile then shifts individual items up where a risk pattern calls for a contingency. Direction only — no item is ever staged down by a patient factor:
- Elevated BMI raises the likelihood of larger-exposure and high-volume suction items.
- Higher ASA physical status raises the likelihood of anesthesia contingency items — monitoring and vasoactive support.
- Reduced ejection fraction raises the likelihood of hemodynamic-support items.
- Prior cardiac surgery (redo) raises the likelihood of re-entry and hemostasis items.
Every recommendation is traceable — a reviewer can see the surgeon history and any patient factors that produced an item's tier. The durable advantage is the longitudinal case-data asset, not the framework.
The safety layer
Emergency items are never dropped
A defined set of emergency cardiothoracic items is always placed in Open Now, no matter what the data would otherwise suggest. That set was defined with cardiothoracic attending surgeons, is never derived from data, and cannot be changed without an explicit attending sign-off. A hard backstop runs last: nothing on the emergency set can ever land in Leave Off — if it would, the run fails rather than ship an unsafe plan. The safety layer only ever escalates an item, never stages one down.
Monitored for false negatives
The primary safety metric is the false-negative rate — a needed item the engine placed in Leave Off. It is monitored continuously; a result above the safety bound triggers manual review before the engine is used further. Per-case savings and adoption are measured conservatively in the pilot and are never presented as a guaranteed target.
Data handling and HIPAA
The retroactive pilot analysis operates exclusively on de-identified data. Under the HIPAA Safe Harbor method (45 CFR §164.514(b)(2)), de-identified data is not protected health information and does not require a Business Associate Agreement.
For live deployment, full Business Associate Agreements are executed with the hospital and relevant subprocessors. Standard safeguards are in place: encryption at rest and in transit, per-request access logging with correlation IDs, role-based access control, and audit trails. Every query is scoped to the deploying hospital's tenant; cross-tenant data access is architecturally prevented.
Accepted de-identified fields
- –Surgeon identifier (anonymized or coded)
- –Procedure name
- –Case date (year and month only, or suppressed for small cells)
- –BMI (numeric)
- –ASA physical status class (numeric)
- –Ejection fraction (numeric, if collected)
- –Redo surgery flag (boolean)
- –Diabetes flag (boolean)
- –Anticoagulation flag (boolean)
- –Items opened / items used (item identifiers)
- –Documented waste cost (if available)
Roadmap
The engine is rules-based today. As the multi-hospital dataset grows, the probability estimator is upgraded — the three-tier output and the safety layer never change.
Today
Deterministic, rules-based engine. Every recommendation traces to a surgeon's history, the patient's profile, and the safety layer.
At data scale
The probability estimator is upgraded to statistical models once per-surgeon case volume supports it. Same three-tier output, same safety layer.
Autonomous run loop
The agent runs the engine across tomorrow's full OR schedule overnight, human-overridable on every screen. Same output, same safety layer.
See the savings on your own data.
Send six months of de-identified cardiothoracic case data and receive a CFO-grade waste analysis. No commitment, no BAA required.