Artificial Intelligence Predicting Which Patients Will Die Prematurely

A group of healthcare specialists have recently created an AI-based system that can predict the risk of premature death caused by chronic disease. This AI platform was found to be very accurate and outperformed current means of prediction by human experts in a study published in PLOS ONE.

This system was trained using healthcare data from over 500,000 people between ages 40 and 69 from the UK Biobank between 2006 and 2010, followed up through 2016. At the helm of the research was Dr. Stephen Wang, assistant professor of Epidemiology and Data Science at the University of Nottingham.

“Preventative healthcare is a growing priority in the fight against serious diseases so we have been working for a number of years to improve the accuracy of computerized health risk assessment in the general population,” said Wang. “Most applications focus on a single disease area but predicting death due to several different disease outcomes is highly complex, especially given environmental and individual factors that may affect them.”

To test the participants likelihood for premature death, the researchers evaluated both deep learning AI, which layered information-processing networks to train the computer from examples, and a “random forest” consisting of a simpler AI that combined multiple tree-like models to account for possible outcomes. These AI systems were then compared to results generated from the Cox model, a standard algorithm.

READ MORE: Using Robotics and AR to Help Those With Motor Impairments

In the ten-year span that these systems were evaluated, nearly 14,500 of the pool of over half a million patients died, primarily due to cancer, heart disease, and respiratory disease. Each of the models found age, gender, smoking, and prior cancer history to be top variables for predicting likelihood of early death, but each emphasized different factors.

The Cox model found ethnicity and physical activity to be key contributors, whereas the AI models did not. The random forest model also placed more emphasis on body fat percentage, fruit and vegetable consumption, and skin tone according to the study. The deep learning model listed job-related hazards and substance intake as top factors.

The researchers noted that the deep-learning algorithm was determined the most accurate of the three, identifying 76 percent of the premature deaths accurately. The random forest model correctly predicted 64 percent of these deaths, and the Cox model only 44 percent.

Joe Kai, professor and clinical researcher involved in the project said: “There is currently intense interest in the potential to use ‘AI’ or ‘machine-learning’ to better predict health outcomes. In some situations we may find it helps, in others it may not. In this particular case, we have shown that with careful tuning, these algorithms can usefully improve prediction.

READ MORE: Patients Telling Digital Healthcare Apps of Suicidal Thoughts

“These techniques can be new to many in health research, and difficult to follow. We believe that by clearly reporting these methods in a transparent way, this could help with scientific verification and future development of this exciting field for health care.”

This AI study follows the Nottingham team’s previous work in which four different algorithms, were found to be significantly better in predicting cardiovascular disease than algorithms commonly used in cardiology today. The team feels AI will play a pivotal role in creation of future systems to customize risk management to the specific patient.

Sources: Science Daily, Live Science