In intensive care units, some monitoring device or other is always sounding the alarm. However, many of these are in fact false alarms, which means there is a risk that real alarms will go under. It is therefore in the interest of nurses and doctors to greatly reduce the number of false alarms, the Swiss Federal Institute of Technology in Zurich (ETH) explained in a press release.
Working with scientists of the Neurocritical Care Unit at the University Hospital Zurich, ETH researchers have now developed a machine learning method that aims to achieve just that. “Usually, before a computer can start learning, humans first need to have categorised a certain number of alarms as relevant or non-relevant,” says Walter Karlen, Professor of Mobile Health Systems at ETH. Having someone classify alarms in intensive care is something of a never-ending task, so a system was needed that could teach itself even if nurses or doctors have classified only a small number of alarms.
In the method developed by ETH scientists, it took just 25 or 50 manual classifications for the system to flag a large number of alarms as false. “Particularly in situations where there has been very little manual help, the new method is much more effective than existing machine learning methods,” ETH reported. The effectiveness of this algorithm will now be tested in a prospective clinical study.