How Early Is Early Enough? What Advance Warning Is Really Worth in Patient Monitoring

If a system warned you a week before a patient got worse, would you believe it?

Probably not. A week is too far ahead to feel real, and you would have little idea what to do with the warning. Now picture the opposite: a warning two minutes before a patient collapses. That you would believe. It would also be almost too late to help.

Predictive patient monitoring sits between those two extremes. It uses the patterns in a patient's readings to estimate whether they are heading towards trouble before it becomes obvious. The interesting question is not only how early the warning comes, but whether it comes early enough for something useful to happen.

Advance warning has to pass two tests

The gap between a warning and the event it predicts is often called lead time. It sounds like a number where bigger is always better. In practice it has to pass two tests.

The first is action. The warning has to arrive while the care team still has options: time to see the patient, run tests, change treatment, move them somewhere they can be watched more closely, or call a senior doctor.

The second is trust. The warning has to be reliable enough that busy staff act on it. A system that warns very early but is often wrong will soon be ignored, however much notice it gives.

What a few hours can change

For many patients on a hospital ward, a few hours is a meaningful window. It is long enough for a doctor to come and see the patient, for blood test results to come back, for fluids or antibiotics to start, or for the team to decide the patient needs more intensive care. It is short enough that the warning still feels connected to the patient in front of them.

That is why a warning measured in hours has real practical value, even if it sounds modest next to bigger claims.

What the evidence shows, and where it comes from

Advance-warning figures are easy to misread, so it helps to be clear about where each one comes from.

At Lifesigns, predictions come from the AI engine built into Lifesigns Nexus, the company's patient monitoring platform. It works from the way each patient's vital signs change over time. In the groups of patients where Lifesigns has measured it, this has typically given three to six hours of warning, and one pilot recorded an average of about 172 minutes.

Longer warnings become possible with more information. PICTURE is a set of AI models developed at the University of Michigan's Weil Institute that uses the patient's electronic health record, including routine blood test results, to predict deterioration in adults on general hospital wards. The university reports that PICTURE gave an average of 30 hours of advance warning, and 17 per cent fewer false alarms for each true alarm than the widely used Epic Deterioration Index when tested on COVID-19 patients. PICTURE is incorporated under licence into the AI engine in Nexus, for hospitals where that richer record is available.

Each figure belongs to its own setting. A result from general-ward patients with full electronic records does not automatically apply to an ambulance or an ICU. That is how evidence should be read.

A warning needs a reason

Even a well-timed warning can be wasted if the person receiving it does not understand it. An alert that shows which changes drove it, such as a rising heart rate alongside faster breathing and a worsening blood test, gives the team somewhere to start. The advance notice creates the opportunity, and the explanation helps the team use it.

So the most useful way to judge predictive monitoring is a practical question: early enough for what? Early enough to see the patient, test, treat or call for help. When a warning arrives in time for those things, and clearly enough to be trusted, it has done its job.