Monitor alarms and decision support: what does an anaesthesia co-pilot do?

Often it is an alarm that makes us look back at the monitor. In my experience, this happens in the middle of other tasks: answering the surgeon, filling in records or following another part of the procedure.

The alarm does an important job. But once we look at the screen, other questions come up. Is the change real? Is the signal credible? What has changed in the patient? What do we need to check?

Decision support belongs in that gap, between noticing the warning and understanding the case. Its aim is to help link the information, while assessment and decisions stay with the clinical team.

What alarms already do

A threshold alarm tells you that a measurement has met a set condition. Say a lower heart-rate limit is set: the warning then depends on that condition being reached.

Still, it would be wrong to reduce every monitor to a minimum and a maximum. The Mindray ePM 12M Vet manual describes physiological and technical alarms with different priorities ¹. Functions vary from one device to another.

So showing an advantage over monitors takes more than several severity levels in software. A useful comparison looks at the specific equipment and at what each system adds in practice.

Then there is what happens before the alarm. Heart rate may fall over several observations before it crosses a limit. Spotting that trend may prompt an assessment before the warning.

Some changes are abrupt, though. There are acute events, and there are times when the available data show no earlier sign. Vigilance cannot rest on a promise that everything can be foreseen.

Projecting a falling line into the future also falls short. The trend may level off or reverse, or it may go with an expected change in the anaesthetic context. What it means has to be interpreted.

Interpreting before deciding

Decision support tries to organise the relevant data and relate them to context, so whoever is monitoring can work out what to check. In anaesthesia, that may include the sequence of values and the drugs given. It may also include events during the procedure and the quality of the measurements.

AAHA recommends combining monitor data with clinical assessment of the patient ². That still applies once software is added.

A proposed explanation is a hypothesis that clinical assessment may confirm or rule out, and a recommendation stays a recommendation until someone carries it out. The team has to judge whether each suggestion is relevant, and record what it actually did.

One difficulty I run into is telling a clinical change from an unreliable measurement. A system should help make that doubt explicit, without settling it with more certainty than the data justify.

Take a mean arterial pressure below the diastolic pressure in the same set of measurements. That is an inconsistency, and the information needs checking. The error could lie in the reading, the channel identification or how values were matched, and finding out which takes more than the set alone.

Low SpO₂ with a low perfusion index is a different case. It does not show that the sensor is the only problem. The change may involve the patient, the measurement or both. Perfusion index should not be used alone to dismiss a worrying value ³.

Trends and prediction

A trend describes the change seen so far. A predictive system tries to estimate a future event within a defined time horizon.

Trends may be relevant data for research into prediction, but showing that a system can predict takes more. You have to define the event and judge how much advance warning is useful. You also have to measure errors, including false warnings and missed events.

Predictive validation takes more than creating alert levels or spotting a rise. I want to move in that direction, as a research goal, but that is not evidence that VetNoa already predicts deterioration.

How we describe VetNoa

VetNoa is a platform under development to support anaesthetic monitoring. It uses a phone camera to take photos of existing monitors, and it reads only the numbers shown on the screen. It does not analyse waveforms, such as ECG, capnography or spirometry.

Capture is automatic and adapts to the alert level assigned by the system to the available data. At the system's Stable level it runs roughly every 20 seconds, at Watch roughly every 10 seconds and at High roughly every 5 seconds. At Critical it runs as often as the system can process a reading, which is still periodic, not instantaneous or continuous. The user can set a fixed interval instead, or take a reading by hand.

VetNoa organises this information over the course of the case. VetNoa assigns an alert level to the available data — Stable, Watch, High or Critical — and shows which signals to watch. It suggests treatments by name, without doses, as suggestions for the team to assess.

The current product does not predict or forecast. Prediction is a separate research and validation goal. Readings are periodic and may contain errors or gaps, so VetNoa does not replace the monitor's own continuous alarms. The monitor and the patient both still need direct assessment.

Frequently asked questions

Does a co-pilot replace the professional responsible for the anaesthetic?

No. Assessment, decisions and intervention remain the team's responsibility. Watching the patient and delivering care still fall to the same people, with or without software.

Should VetNoa replace monitor alarms?

No. It should be presented as complementary support. VetNoa reads the monitor periodically, so it does not replace the monitor's continuous alarms. The equipment's own features still need to be used and set up carefully, whatever other system is in place.

Does a software warning confirm a diagnosis?

No. It may point to a change, an inconsistency or a hypothesis to check. How it is read depends on data quality and clinical assessment.

If there is no warning, can we conclude that everything is fine?

No. Data may be missing, capture may be limited or the system may have missed an event, all without any warning. The absence of a warning does not confirm normality.

Does VetNoa predict complications?

No. The current product has no prediction or forecasting. Predictive capability is a separate research and validation goal. Showing that complications can be predicted takes more than organising data and analysing trends.

How should we judge whether a co-pilot is useful?

By the quality of its data and the relevance and clarity of what it presents, and by its proven performance in the setting where it is used. Benefits such as a lighter workload or better outcomes need to be evaluated before anyone assumes them.