Researchers at ETH Zurich and the Inselspital Bern have developed a method that can predict the circulatory failure of patients in intensive care units with high reliability. Medical staff can thus intervene earlier. The approach is based on the evaluation of extensive patient data using machine learning methods.
Patients in the intensive care unit of a hospital are under close observation: vital signs such as pulse, blood pressure and blood oxygen saturation are measured continuously. In this way, the doctors and nurses have a wealth of data at their disposal for assessing the patients' state of health. Nevertheless, it is not easy to use this information to make predictions about the further development of the condition or to identify life-threatening changes far in advance.
Researchers at ETH Zurich and the Inselspital Bern have now developed a method that combines the various vital signs and other medically relevant information about a patient. This allows critical circulatory failure to be predicted several hours before it occurs. The aim is to use the method in the future to evaluate the vital signs in the hospital in real-time and to warn the treating staff in advance. This will enable them to take appropriate measures at an early stage.
Extensive data set
The development of this approach was made possible by a comprehensive data set from the University Clinic for Intensive Care Medicine at the Inselspital. In 2005, the Inselspital became the first large intensive care unit in Switzerland to start storing detailed and temporally high-resolution data of intensive care patients in digital form. For the study, the researchers used data from 36,000 stays in intensive care units in anonymized form and exclusively from patients who agreed to have these data used for research purposes.
On the initiative of Tobias Merz, a researcher, formerly head physician in intensive care at the Inselspital in Bern and now working at Auckland City Hospital, researchers led by ETH professors Gunnar Rätsch and Karsten Borgwardt analyzed this data using machine learning methods. "The algorithms and models thus developed were able to predict 90 percent of all circulatory failures in the data set used. In 82 percent of all cases, the prediction was made at least two hours in advance, which would have left doctors time to intervene," says Gunnar Rätsch, Professor of Biomedical Informatics at ETH Zurich.
Relatively few measured variables are sufficient
For this work, the researchers had several hundred different measurements and medical information per patient at their disposal. "We were able to show, however, that 20 measurements are sufficient for an accurate prediction. These include blood pressure, pulse, various blood values, age and the medication administered," explains Karsten Borgwardt, Professor of Data Mining at ETH Zurich.
To further improve the quality of the forecasts, the researchers plan to include patient data from other large hospitals in future analyses. In addition, the anonymized data set and the algorithms and models will be made available to other scientists.
Few, but highly relevant alarms
"In intensive medical patient care, it is central to prevent circulatory failure. Even short periods of insufficient circulation increase mortality significantly," says intensive care physician Tobias Merz. "Today we have to deal with a multitude of alarms in intensive care units. These are not very precise. Frequent false alarms and short advance warning times lead to delays in circulation-supporting measures". With their approach, the researchers, therefore, want to reduce a large number of alarms to a few, but highly relevant and early alarms. This is possible, as the study showed: with the new method the number of alarms could be reduced to a tenth.
Further development work is needed to ensure that the method can be used as an early warning system. A first prototype already exists, as ETH Professor Rätsch says. Its reliability must now be proven in clinical studies.
This article was originally published on netzwoche.ch