Clinical research technology and CRO services

ERTRIAGE clinical evaluation program.

This is a planned prospective evaluation of AI-assisted emergency department triage. It is under development as a research program. It is not an enrolled study, not a registered trial, and not a performance claim. No hospital is named, no patient has been counted, and no ethics approval is stated here.

What you can use

Questions the protocol would have to fix

Clinically important undertriage and overtriage, agreement with an independent reference standard, consistency of classification, time to complete triage, and differences across sites and subgroups. Those are objectives to design, not results.

What ValidTrials would hold

The protocol version, site records, forms, an independent read, monitoring notes, the analysis snapshot and the figures from that run.

What we would not hold

A marketed accuracy number, an MDR certificate, or a list of hospitals that have not agreed.

Conflict of interest

If a related company develops the evaluated software and also provides research services, that conflict is disclosed and scientific oversight is independent of the people who built the model.

Synthetic demonstration

Any walkthrough uses invented records labeled synthetic. They are not patients.

Your study

A similar design for your own device is a separate protocol and a separate agreement.

How the work moves

Write the intended use

Before any case is scored.

Choose the reference

Independent of the model.

Approve and register

Ethics and, where required, a trial registry. Not done on this page.

Then collect

Only after those approvals exist.

Related

Scientific background

Emergency department triage sorts patients by urgency. Undertriage assigns too low an urgency to someone who needs faster care. Overtriage assigns too high an urgency and consumes capacity. An AI assistant that suggests a triage category can change either error. Studying that suggestion requires a protocol, a frozen software version and a reference that is not the model itself. This program is the design for that kind of study. It is not an enrolled study.

Proposed objectives, not results

Undertriage and overtriage
Clinically important misclassification against the reference, defined before any case is scored.
Agreement
How often the suggestion and the independent reference match, with the uncertainty the plan specifies.
Consistency
Whether the same presentation receives the same category.
Time
How long triage takes with the workflow the protocol allows.
Sites and subgroups
Whether those measures differ across hospitals and the groups named in advance.

How the study would be run

Prospective collection, after ethics approval and any required registration. Those approvals are not claimed here. Staff would follow a written rule about whether they may see the suggestion. The reference standard would be an independent clinical review, with a path for adjudication when readers disagree. Primary safety concern is undertriage: a suggestion that makes a sick patient look less urgent. No patient count on this page is a sample size. A number belongs in a statistical justification that states its assumptions, and that justification has not been published here.

Software version, statistics and governance

Version
The build under evaluation is named. A later build is an amendment, not a silent replacement.
Statistics
The plan would prespecify the estimates, the intervals and the subgroups. Methods that are not implemented would not be promised.
Governance
A sponsor, independent scientific oversight, and a data snapshot that the report must cite.
Conflict of interest
If a related company develops the evaluated software and also provides research services, that is disclosed. The people who built the model do not own the reference standard or the final interpretation.
ValidTrials
Would hold the protocol, sites, forms, adjudication, monitoring notes, the snapshot and the figures. It would not be the ethics committee and not the regulator.
Status
Planned. No hospital, enrollment, ethics approval, registry entry, accuracy result or MDR outcome is stated.

A synthetic walkthrough

In a demonstration, an invented presentation is labeled synthetic. A coordinator opens the approved form, records a suggested category and the independent reference, and a disagreement goes to adjudication. A snapshot is frozen. A figure is produced from that snapshot and marked as demonstration output. None of those rows are patients, and none of them are a performance claim. The same shape is what another AI-device sponsor can ask us to configure for their own intended use, under their own protocol.

Questions

Can I cite an ERTRIAGE accuracy result from this page?

No result is published here.

Is this clinical validation of a device?

No.

Software on this page is what the product can record today. CRO work is a professional service under a separate scope. Neither is a certification, a regulator’s approval, or proof that a study is compliant because the software was used. A qualified person still signs consequential decisions.

See the workflow on your protocol.

The conversation is about your study design and what the software can and cannot do today.