In a recently published study, Effects of Individualized Oxygenation Targets on Mortality in Critically Ill Adults: A Machine Learning Analysis, researchers from Aotearoa New Zealand collaborating with scientists in the United States have unlocked a new frontier in medicine using advanced machine learning techniques.
Traditionally, randomised clinical trials determine the average effect of treatments on patient outcomes. Such clinical trials have revolutionised medicine by providing high-quality evidence on the effect of treatments on patient outcomes; however, one key problem with these trials is that they assume every patient responds to treatment in the same way.
In their recent study, a collaborative research team used machine learning to generate individualised predictions about the effect of higher or lower amounts of oxygen on mortality in critically ill adults receiving life support in the intensive care unit (ICU).
Specifically, they used the data from one study, the Pragmatic Investigation of Optimal Oxygen Targets (PILOT) trial, to generate a model using machine learning, and then they tested the model using data from another trial, the Intensive Care Unit Randomized Trial Comparing Two Approaches to Oxygen Therapy (ICU-ROX) trial.
Their results were remarkable. One of the study's senior investigators, Professor Paul Young, Deputy Director and Intensive Care Medicine lead at the Medical Research Institute of New Zealand (MRINZ), explains, “When we applied the machine-learning model we found that the individual responses to higher or lower levels of oxygen varied dramatically. At one extreme, the model predicted a 27.2% absolute decrease in the risk of death by using a lower oxygen target, and at the other extreme, the model predicted a 34.4% absolute decrease in the risk of death by using a higher oxygen target.”
“Overall, if patients in the ICU-ROX trial had received the oxygen therapy regimen recommended by the machine learning model, the mortality rate would have been 6.4 percentage points lower,” says Professor Young.
Dr Alex Psirides, Chair, Critical Care Advisory Group to Te Whatu Ora, states, “Around 40% of the 24,000 patients admitted to New Zealand ICUs each year receive life support. If these findings are confirmed, they would be expected to save the lives of around 600 New Zealanders every year.”
In the accompanying editorial, Dr Derek Angus, the Deputy Editor of the Journal of the American Medical Association, commented, “If the results are true and generalisable, then the consequences are staggering.” “If one could instantly assign every patient into their appropriate group of predicted
benefit or harm and assign their oxygen target accordingly, the intervention would theoretically yield the greatest single improvement in lives saved from critical illness in the history of the field,” states Dr Angus.
Professor Richard Beasley, MRINZ Director, notes that the implications of this research extend well beyond oxygen therapy for patients in the ICU, stating, “This technique of using machine learning to predict individualised treatment responses for patients using data from clinical trials is likely the greatest advance in the generation of medical evidence in decades.”
Professor Beasley says, “It is difficult to overstate the degree to which research of this kind could change medicine.”
“For decades, the practice of medicine has experienced the tension of choosing between care that is personalised but not evidence-based and care that is evidence-based but not personalised. Paul Young and his collaborators have shown machine learning methods can predict individualised treatment effects allowing care that is both evidence-based and personalised,” says Professor Beasley.
This study’s innovative use of machine learning in determining individualised oxygenation targets marks a significant step forward in healthcare. With the potential to bridge the gap between evidence-based and personalised care, it could offer significantly improved patient outcomes. This approach represents a transformative shift in medical research and practice.
