How algorithms can save people from an early death
It’s a scene characteristic for medical series such as House, MD or ER: An alarm goes off and the rapid response team rushes in to revive a patient. But what if the warning had come hours before the life-threatening event actually occurred? The US Food and Drug Administration recently greenlighted a system designed to provide exactly that. “FDA approves crisis-predicting algorithm to save patients from early death” was one typical headline announcing the news.
The reports describe WAVE, a platform developed by Florida-based med-tech company Excel Medical. Created to protect at-risk patients by predicting life-threatening situations, the always-on system monitors key vital signs and calculates the risk of a potentially fatal event, such as heart attack or respiratory failure, within the next six to eight hours – and then immediately warns hospital staff.
Some commentators found it notable that WAVE is driven by a computer program. “This is the first such algorithm to receive FDA approval” is how the US-based tech blog Gizmodo put it. Excel Medical wrote in its press release that the WAVE platform is the “first of its kind to be cleared by the FDA.” Talking to Gizmodo, Mary Baum, chief strategy officer at Excel Medical, said that WAVE saves patients’ lives by recognizing when their condition is deteriorating, thus saving precious time.
Responding to a medical crisis before it happens – thanks to algorithmic monitoring
According to one study, up to 400,000 people die in medical institutions in the United States each year because rapid response teams usually do not have enough time to react to potentially lethal cardiac events.
But can algorithms really predict the chance that a person is going to die, thereby saving lives?
Health authorities are convinced the answer is “yes.” And in contrast to what current stories in the US media suggest, the life-saving system is actually based on software which has been in use for a while and has therefore already been tested: Visensia – The Safety Index. This early warning index, used to monitor patients in hospitals, was originally developed by OBS Medical, a spin-out of Oxford University. It received a CE marking back in 2010 as an approved medical product and was cleared by the FDA in 2011. Since then the software has been used by hospitals and clinics, where it is integrated into existing patient-monitoring systems. In the EU, a number of companies license the index, including Swiss-based Anandic Medical Systems.
Excel Medical now also licenses the algorithm and again received approval by the FDA to market the software as an integral part of WAVE. The clearance is based on a number of clinical studies at the University of Pittsburgh Medical Center, where the platform’s effectiveness and safety were tested. As part of their analysis, researchers compared two groups of elderly patients: One was monitored using the software, the control group was not. Six unexpected deaths occurred in the control group, with no deaths taking place in the treatment group.
The remarkable part is that the algorithmic system uses medical data – such as heart rate, respiratory rate, blood pressure, blood oxygen levels and body temperature – that have long been available to check a patient’s condition. This is what the typical emergency response looks like: A monitoring device sounds the alarm only if one or more vital signs exceeds predetermined limits, at which point the hospital staff must react quickly. WAVE, however, does more than just monitor whether individual parameters are within their normal range. Instead, it compares the various values and analyzes their combined effect. For example, if blood oxygen levels were to fall slightly, a conventional system would not signal that something dangerous is happening. But if blood pressure and pulse also exhibit slight changes, then the WAVE technology sees all of these events collectively as an early indicator that the patient’s condition is deteriorating – several hours before any one vital sign becomes an outlier, thus setting off an alarm.
According to Excel Medical, WAVE could save the lives of thousands of hospital patients by alerting medical staff sooner than conventional systems. If the response team receives the information hours in advance that an adverse event is likely, it can calmly decide which measures to take to stabilize the patient.
Human intuition and real-time algorithmic analysis – a life-saving combination
An overview study by Radboud University Medical Center in the Netherlands, among others, suggests that WAVE and the underlying Visensia Safety Index could indeed provide valuable support for medical staff. The analysis shows that nurses often recognize early on that a patient’s condition is worsening, yet their concern is frequently based on intuition and not on objective information such as heart rate or blood oxygen levels. Instead, their assessment is often influenced by their own subjective, even subconscious perceptions, such as a difference in the patient’s behavior or gaze.
Yet there are no standard procedures for including nurses’ intuitive responses in a patient’s treatment. Surveys reveal that many nurses have difficulty putting their intuition into words – allowing them to explain, for example, why they feel a patient “does not look good.” A nurse with many years of experience will probably be able to trust their gut instinct more than someone just out of nursing school. And they will have a better sense of when and how they should communicate their concern to a doctor.
Algorithms such as the one underlying the WAVE system can thus help doctors and nurses become better at doing their jobs. After all, humans can only register and analyze a limited amount of information about vital signs, while algorithms can spot potentially life-threatening situations in real time, even if individual numbers have only changed minimally. That means algorithms can perhaps confirm what nurses merely suspect based on the patient’s appearance.
This is the first part in a three-part series on using algorithms to predict death.
Part 2 is available here: “Optimizing palliative care: When algorithms predict a patient’s death”
Part 3 is available here: “Using algorithms to predict death: Lessons learned”