LINCR

AI in healthcare reads notes or predicts numbers. Mechanism is the missing layer.

The current AI-in-healthcare stack splits two ways. The first wave — Abridge, Nuance DAX, Hippocratic AI, Glass — runs language models over clinical notes for scribing, billing, and summary generation. Useful documentation products. They never see physiology. The second wave — gradient-boosted and transformer time-series predictors like Edwards's HPI, the only FDA-cleared example in intraoperative care — has two structural problems. First, it only runs on Edwards's proprietary HemoSphere monitor + Acumen IQ sensor: capital purchase per OR, per-case disposables, and gated to invasive-arterial-line cases — a small fraction of US ORs and zero ambulatory surgery centers. Second, even where it does run, it pattern-matches on waveform surface signals to predict BP will drop in 15 minutes — without modeling the cardiovascular physics underneath. It can warn that something is coming, but it can't say why, and a warning without a cause doesn't change what the clinician does.

Each year ~30 million U.S. patients undergo surgery, and roughly half develop intraoperative hypotension at clinically meaningful thresholds (MAP <65 mmHg). The AKI, myocardial injury, prolonged stays, and SNF discharges that follow drive tens of billions of dollars in direct cost and a large share of anesthesia malpractice claims. The largest randomized trial to date — IMPROVE-multi, n=8,520 — just confirmed that better blood-pressure targeting alone doesn't reduce any of those outcomes. The number isn't enough.

The reason is structural. The same low BP number is driven by four physiologically distinct causes, each requiring a different clinical intervention. Clinicians already reason in mechanism, not in numbers — case-by-case, at the bedside. What's missing is the comprehensive record: retrospective detection of every intraoperative event from the lead-up vital-sign changes, the interventions delivered in response, and the surrounding labs that captured the downstream impact — attributed to the cause that drove it. That is the quality-improvement, safety, and care-optimization layer this product is.

Sources: Sessler et al., IMPROVE-multi (2026). Walsh et al., Anesthesiology 2013. Bijker et al., Anesthesiology 2009.

One product: the retrospective causal audit.

Anesthesia groups upload existing case data; we return the comprehensive record of what drove every intraoperative event in their last quarter — the QI, safety, and care-optimization documentation quality committees, malpractice carriers, and value-based-care contracts actually need.

The audit relationship and the dataset it produces unlock a pre-op risk stratifier downstream — also under the FD&C Act §520(o)(1)(E) CDS carve-out, no 510(k) required. Same QI / safety / care-optimization category. Ships after the wedge wins, not before.

The audit covers the full intraoperative emergency catalog — not just hypotension. Events from the Stanford EPIC manual the audit identifies:

Value-based care turned anesthesia complications into a line item on the hospital P&L.

Healthcare reimbursement is moving off volume and onto outcomes. Hospitals and the anesthesia groups that staff them now carry direct financial exposure for the post-op complications that intraoperative hypotension drives — AKI, MI, prolonged length of stay, SNF discharges, 30-day readmissions. Three CMS programs already put that exposure on the books:

The audit converts that exposure into something a quality committee can act on: which cases drove which complications, which causal pattern clustered where, which protocol changes would have prevented the events that incurred the penalty. That is why anesthesia groups and hospitals will pay for the wedge product today — the audit is a budget defense, not a research tool.

Sources: CMS TEAM Final Rule (CMS-1808-F, August 2024). HAC Reduction Program (Section 3008, ACA). BPCI-A program documentation. Anesthesia Quality Institute (AQI) closed-claims analysis 2022.

We train a model that produces per-subject estimates of human intraoperative physiology — the substrate behind every clinical decision-making tool we ship.

That substrate powers the retrospective audit shipping today, the prospective stratifier downstream, and the broader QI / safety / care-optimization tools the audit relationship and dataset unlock. CF2 is the architecture: a 63,975-parameter compartmentalized neural-ODE estimator with three compartments — cardio (27 params), respiratory (8), pharmacokinetic (14) — that encode the actual physics of preload, afterload, contractility, ventilation, and drug PK. The encoder is trained once on population data; at inference, it ingests each patient's streaming signals (waveforms, vitals, drug record, labs) and produces per-subject parameter estimates that fit that patient's physiology. Those parameter estimates feed the audit's cause-attribution layer alongside clinical-rule triangulation across vitals, interventions, and labs — CF2's contractility / SVR / preload outputs combined with drug-record timing, SpO2 drift, EKG morphology, and lab markers to assign each event to its driving cause. The model learns physiology. Notes-LLMs and time-series predictors pattern-match on the output of physiology.

On retrospective benchmarks, CF2 outperforms the only FDA-cleared incumbent. HPI remains the relevant scientific benchmark — it's the only FDA-cleared AI predictor in this space — even though it failed to change practice. CF2 reads what the OR already produces — arterial waveform or NIBP cuff, ECG, pleth, capnography, drug record. No new hardware, no proprietary sensor, no per-case disposable. The same trained CF2 model produces per-subject estimates across both signal regimes without retraining; HPI is structurally locked to Edwards's HemoSphere + Acumen IQ in invasive-arterial-line cases. The same model carries from inpatient hospital cases into ambulatory surgery centers — into a regime HPI cannot operate in.

Methodology disclosure: prevalence is window-level (samples around hypotension events), not patient-level — standard in the published HPI literature and the rest of the IOH-prediction field. Patient-level intraoperative hypotension base rate is roughly 10–30% per Walsh et al. and Bijker et al. Brier Skill Score >0.7 is publication-grade calibration. Patent: USPTO Provisional App #64/011,899, "Compartmentalized Neural Ordinary Differential Equation System for Dynamical State Prediction," filed 2026-03-20, 109 clauses, sole inventor Anish Joseph.

First paid pilot Q4 2026. Wedge first; growth follows the audit relationship.

Built by someone who administers the drugs the model classifies the response to.