A new algorithm developed thanks to the University of Copenhagen uses 23 years of medical records to predict the survival chances of patients undergoing intensive care.
Determining the best treatment for each critically ill patient is a major challenge and the methods existing can be improved greatly with the use of AI and machine learning.
The new algorithm, presented in Lancet magazine, leverages Danish patient health data from the Danish National Patient Registry, an endless database that collects millions of patients, “and is able to define for each patient the benefit connected to the care received,” explains the professor Soren Brunak of the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen.
230,000 times thanks
To develop the algorithm, researchers used data from more than 230.000 patients admitted to the unit of intensive therapy in Denmark during the period 2004-2016. The study includes the medical history of patients they had illnesses dating back up to 23 years.
The calculations also included measurements and tests of the first 24 hours of hospitalization, to improve the accuracy of the mortality risks associated with the first care applied.
“Overtreatment presents collateral risks that require analysis in order to calibrate treatments well in such delicate moments of a patient's life,” adds the professor Anders Perner of the department of intensive care and clinical medicine at Rigshospitalet.
30- and 90-day forecasts
The algorithm, like a modern oracle, offers three predictions: the risk that the patient dies in hospital (and within how many days), the risk of dying within 30 days of admission to the unit, the risk let him die in 90 days.
“We 'instructed' the algorithm to remember which diagnoses had the greatest effect on the patient's chances of survival, net of his age (young patients risk less on average than elderly ones): by analyzing not only the statistics but also the methods adopted we are able to establish not only how many chances of life there are, but also what the best treatments to administer are,” says Brunak.
The researchers hope to be able to use the algorithm in clinical trials within a couple of years. Then the refinement of the algorithm (with the collection of more data) will allow it to detail the forecasts up to understand also how many hours of life a patient arriving at the hospital has left, and what treatments can save his life immediately.
Source: University of Copenhagen
