Clinical AI (Comprehensive)

Predictive models

Rules answer "is this happening now?". The predictive layer answers "is this about to happen?" — every model fitted on your hospital's own historical cohort, never a black-box vendor model, with the validation cohort and results visible to your clinical informatics team.

Mortality risk

A continuous risk score and tier for every ICU patient from the first day of stay, recomputed daily — surfaced as a Live Monitor tile colour and a chart trend curve.

Sepsis trajectory

Watches the live observation stream for the quiet early signatures — small persistent rises in heart rate, drifts in mean arterial pressure, lactate trends — and composes them into a risk band refreshed every fifteen minutes through the critical first day, surfacing sepsis 4–6 hours ahead of clinical declaration.

Deterioration & instability

Continuous early-warning scoring for every inpatient, recomputing as new data arrives, feeding escalation rules and the unit-wide view.

30-day readmission

Scores every discharge plan and surfaces inline on the discharge-planning view, flagging the patients who need a structured intervention before they leave.

Demand & occupancy forecasting

The same analytics layer extends past the patient to the unit — census and capacity foresight so leadership plans ahead instead of reacting.

Waveform analytics — real-time signal analysis

The bedside delivers continuous waveforms at the device's native sample rate — ECG, plethysmograph, capnograph, ventilator pressure/volume/flow, invasive arterial pressure. A curated detector library analyses every channel in real time — available as configurable clinical intelligence modules, subject to customer validation and the applicable regulatory pathway.

Lethal-rhythm detection

Ventricular fibrillation, ventricular tachycardia and asystole detected within seconds of waveform receipt — with built-in self-cooldown so the same arrhythmia does not page the clinician forty times in a minute.

ECG morphology drift

QRS-width trends, onset of conduction disease, ST-segment elevation and depression trends — caught as drift, not as a crisis.

Ventilator waveform analysis

Tidal volume, plateau and driving pressure, PEEP creep, patient–ventilator asynchrony, double-triggering and breath-stacking — with respiratory severity scoring reading directly from the analytics rather than a manually charted entry.

Capnograph & perfusion analysis

End-tidal CO₂ waveform shape and trend, rebreathing patterns, perfusion-index trend and perfusion-failure surveillance from the plethysmograph.

Every detection lands in the flowsheet

Each finding is written into the chart as an observation with its timestamp, detector identity and the contributing waveform window attached — consumed by the same rules, scores and predictions as any hand-charted value. One flowsheet, one source of truth.

AI-drafted words

Where the predictive layer surfaces risk, the drafting layer surfaces words. It drafts; the clinician edits and signs.

Rationale drafts

When a rule fires, the AI drafts the one-line rationale citing the underlying observations and the relevant guideline — a written explanation, not just a flag.

Shift & discharge drafting

Structured handover summaries of the last twelve hours — events, score trends, bundle adherence, outstanding tasks — plus discharge summaries, medication reconciliation and patient-facing instructions.

Clinical narratives

Structured delirium narratives when an assessment turns positive, and one-line trend descriptions when a flowsheet row bends in a concerning direction.

What it never does

It places no orders, signs no documents, and never skips the clinician's review. Every output is timestamped, attributed to the model revision, and stored alongside the clinician's edits in the audit log.

Trust & governance

Explainable by construction

Every suggestion shows the rule or model that produced it, the contributing observations, and the citation when present — the clinician can always independently review the basis.

Advisory only, always

zMed makes no diagnosis, prescribes no medication, orders no procedure, and discharges no patient. Every output is one tap from override, and the chain of clinical responsibility flows through the human clinician at every step.

Regulatory posture

Designed with advisory, explainable, human-in-the-loop CDS principles. Final regulatory classification depends on intended use, deployment configuration, clinical claims, and jurisdiction. The audit log is the inspector's primary artefact.

Your data, your boundary

Models run in a secure environment and can run fully on-premises; patient data is never used to train shared or third-party models, and protected health information is redacted before any cloud model is invoked.