An AI that forecasts vital signs could get patients off the IV drip sooner

01The question

Serious infections are often treated with antibiotics dripped straight into a vein (IV). Once patients improve, switching them to antibiotic pills can shorten their hospital stay, lower the risk of infections from the IV line and save staff time. Yet in England about one in five patients stay on IV antibiotics even after they meet the rules for switching.

Part of the problem is sheer numbers. A typical English hospital has around 550 inpatients at any moment, and about a third of them are on antibiotics, far more than the small infection teams can check every day. The researchers asked whether a computer could point those teams to the patients most likely to be ready.

02What they did

A team at University College London built a system that looks at the last 48 hours of a patient's five basic vital signs: heart rate, breathing rate, oxygen level, blood pressure and temperature. It then forecasts how those signs are likely to change over the next 12 hours, including how uncertain each forecast is.

Here's the clever part. Instead of learning from what doctors did in the past, which would copy their delays, the system checks its forecasts against the standard clinical rules for "stable." Patients most likely to stay in the healthy range get ranked highest on a daily 9 am review list. The team tested it on past records from a London hospital group (10,584 hospital stays) and a Boston intensive care unit (6,333 stays).

03What they found

Among the top five patients it flagged each day, the system found 2.2 times as many truly ready patients as picking at random in the London hospitals, and 3.2 times as many in the sicker intensive care patients. It forecast vital signs more accurately than simpler methods on most measures, and it worked similarly well across age, gender and ethnic groups.

Other kinds of AI models ranked patients about as well. The team favors this design because doctors can see why a patient was flagged: the system shows the predicted vital signs themselves, not just a score. The rules can also be adjusted without retraining it, for example ignoring breathing rate for a patient on a ventilator.

In one example from the records, the system rated a patient as ready to switch before their hospital stay ended, even though that patient was never actually switched to pills.

04Why it matters

Getting people off IV lines sooner is safer, frees up nurses and saves hospitals money. It also fits the broader push for antibiotic stewardship, using antibiotics only as long and as intensively as needed.

The system is designed to help doctors, not replace them. It only ranks who to look at first, and the final decision about pills stays with the medical team.