Clinicians working in intensive care units (ICUs) use various metrics to estimate an individual’s chance of survival and to determine the best treatment course. Researchers in Denmark sought to improve the accuracy Previous disease history and acute physiology data from the first 24 h of ICU admission were used separately or in combination as inputs to a neural network algorithm to predict in-hospital, 30-day and 90-day mortality. Source: A. Nielsen et alof such predictive techniques with an algorithm based on neural networks designed to weigh various factors in a patient’s medical history.

To develop the algorithm, which is described in a study published in Lancet Digital Health, researchers used data on more than 230,000 patients admitted to ICUs in Denmark during 2004-2016. Data from each patient’s disease history, covering as much as 23 years, were included, along with measurements and tests made during the first 24 hours of the most recent admission. The result was a more accurate prediction of the patient’s mortality risk than offered by existing methods.

Three mortality predictions are provided by the algorithm: the risk of a patient dying in the hospital, which could be any number of days after admission, risk of the patient dying within 30 days of admission and risk of the patient dying within 90 days of admission. Initial assessments of the tool revealed that age and length of previous hospital visits were variables that markedly affected predictions. A young age at admission lowered the risk of dying, even when other values were critical, while old age increased mortality risk.

The researchers from Daintel, University of Copenhagen and Rigshospitalet plan to further improve the predictive capabilities of the algorithm and evaluate it in clinical tests within a few years.