HUS participates in the development of an algorithm improving traumatic brain injury care in ICU

The development of the algorithm started in 2017 by the two lead authors associate professors Dr. Rahul Raj and Dr. Miikka Korja. The study was a collaboration between the Finnish university hospitals in Helsinki, Kuopio, Tampere, Turku, and Karolinska Hospital in Stockholm, Sweden.

It is challenging to predict the outcome and effects of different interventions when treating patients with traumatic brain injury (TBI) in the intensive care unit, especially when the treatment is prolonged. Patients with severe TBI are often unconscious and the brain is monitored by several indirect markers, such as intracranial pressure, systemic blood pressure, and cerebral perfusion pressure. These parameters give over a million data points per day making it virtually impossible for the human brain to analyze all data. Thus, treatment decisions will often be based on incomplete information, which can lead to suboptimal treatment decisions. That is why the researchers wanted to develop a prediction model that can use all the gathered data and give real-time predictions to improve the decision-making process.

"We aimed to develop a big-data algorithm that can be used all over the world using a minimal number of different parameters. The used parameters must be objective so that they are not subject to human interpretation, as this could bias the algorithm. We could have developed an algorithm based upon a more extensive set of different parameters but then it could not be used in settings unable to perform such extensive monitoring", says one of the first authors Dr. Raj.

The algorithm is based upon three parameters that are most frequently monitored in intensive care units around the world, i.e., intracranial pressure, systemic blood pressure, and cerebral perfusion pressure. Several new so-called features were engineered from these parameters to capture the three parameters’ complex relationship with outcome after brain injury.

"We did not want the algorithm to be too accurate during the first 1-2 days as this could lead to hastened treatment decisions. The algorithm gets exposed to more data as the treatment goes on; thus, the predictions become more accurate. Consequently, the main benefit of the algorithm is that it provides us with a tool to analyze trends in prognosis, especially when the treatment is prolonged, and the clinical decision-making processes are the most difficult. Another benefit of the algorithm, is that it also can be used to assess the effects of different interventions given", says one of the first authors Dr. Wennervirta.

The algorithm was developed in a cohort consisting of 686 patients from Finland with 62,000 hours of monitoring data and validated in two international cohorts consisting of 638 patients with 60,000 hours of monitoring data coming from Sweden and the United States. The developed algorithm gives a new prediction for every 8 hours of treatment and predicts the risk of 30-day mortality.

"More importantly, the number of false positives, i.e., the algorithm predicts death, when the patient survives, was no higher than 2.5%. Furthermore, the majority of patients with false positive predictions had poor one-year functional outcomes. This is very important as it minimizes the risk of erroneously withdrawing care in patients that could have survived", says Dr. Korja, one of the lead authors.

"The algorithm is meant to inform the treating physician and aid decision-making. It was not our goal nor is it reasonable that this or future algorithms would make independent life-or-death treatment decisions. However, with the use of the algorithm, physicians can make more data-driven informed treatment decisions, which could improve outcomes", Dr. Korja continues.

The study can be found here: https://www.nature.com/articles/s41746-022-00652-3

Corresponding author: Rahul Raj