A new algorithm uses up to 23 years of individual disease history to predict patients’ chances of survival in the intensive care unit.
Determining which treatment is best for each intensive care patient is a great challenge and the existing methods that doctors and nurses use could be much better.
The new algorithm appears in the journal Lancet Digital Health.
“We have used Danish health data in a new way, using an algorithm to analyze file data from the individual patient’s disease history. The Danish National Patient Registry contains data on the disease history of millions of Danes, and in principle the algorithm is able to draw on the history of the individual citizen of benefit to the individual patient in connection with treatment,” says professor Søren Brunak from the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen.
230,000 intensive care patients
To develop the algorithm, the researchers used data on more than 230,000 patients admitted to intensive care units in Denmark in the period 2004-2016. In the study, the algorithm analyzed the individual patient’s disease history, covering as much as 23 years.
The calculations also include measurements and tests from the first 24 hours of the patient’s hospital admission. The result was a significantly more accurate prediction of mortality risk than offered by existing methods.
“Excessive treatment is a serious risk among terminally ill patients treated in Danish intensive care units. Doctors and nurses have lacked a support tool capable of instructing them on who will benefit from intensive care. With these results we have come a significant step closer to testing such tools and directly improving treatment of the sickest patients,” says professor Anders Perner from the departments of clinical medicine and of intensive care at Rigshospitalet.
30- and 90-day predictions
The algorithm makes three predictions: the risk of the patient dying in hospital (which could be any number of days following admission), within 30 days of admission, and within 90 days of admission.
For example, the researchers could tell that up to 10-year-old diagnoses affected predictions, and that young age lowered the risk of dying, even when other values were critical, while old age increased mortality risk. The algorithm is not just a useful tool in everyday practice in ICUs—it can also tell us which factors are significant when it comes to a person’s death or survival.
“We ‘train’ the algorithm to remember which previous diagnoses have had the greatest effect on the patient’s chances of survival. No matter whether they are 1, 5, or 10 years old. This is possible when we also have data from the actual admission, such as heart rate or answers to blood tests. By analyzing the method, we are able to understand the importance it attaches to the various parameters with regard to death and survival,” says Brunak.
The researchers hope to be able to use the algorithm in clinical tests within a couple of years. At the same time, the next step is to try to further develop the algorithm, making it capable of making predictions by the hour.
Innovation Fund Denmark and the Novo Nordisk Foundation supported the work.
Source: University of Copenhagen
