Nice to meet you.

Enter your email to receive our weekly G2 Tea newsletter with the hottest marketing news, trends, and expert opinions.

AI or the Grim Reaper? Predicting Mortality Rates With AI

December 19, 2019

The evolution of artificial intelligence has seen its impact in various industries. 

AI can be seen from the medical field all the way to the fast food industry. In its simplest form, AI has been implemented in many industries as a chatbot for customer communications such as online help centers. In its more complicated form, AI has been used to actually learn from collected data and make decisions or predictions based on that data. This is called machine learning, a subset of AI. 

Predicting mortality rates with AI

Machine learning has been implemented in many creative ways, but one of it most recent implementations may just send a shiver down your spine. With the right information, AI software can predict the mortality rate for a set of people with accuracy unmatched by doctors and the most commonly used algorithms alike. 

See the Easiest-to-Use Artificial Intelligence Software...

A recent study attempted to use AI to predict premature deaths with the health data of more than 500,000 people from 2006 to 2016, nearly 14,500 who died sooner than the nation average mainly from cancer, heart disease, and respiratory diseases. The data came from the UK Biobank, an open-access database of genetic, physical, and health data. After providing the AI system with said data, it was asked to compute the likelihood of each individuals’ premature mortality, specifically from chronic disease. Within the study, two types of AI were tested: “deep learning” and “random forest”. 

RELATED: See how chatbots can provide accessible healthcare support, thanks to AI.

How deep learning impacts AI predictions 

Deep learning, a subset of machine learning, attempts to replicate the human processing of data by using algorithms inspired by the human brain. This algorithm reports the outcomes of “tasks” or scenarios that it has run over and over again, altering the scenario each time to further improve its outcome, which in this case is its mortality prediction. 

Random forest, another subset of machine learning, consists of a cluster of what are called decision trees. Decision trees distribute the collected data into specific classifications by using a yes/no branching system. In our case, these “classifications” are whether or not an individual is likely to die due to chronic illness. The classifications are collected from all the decision trees in the random forest and the most common classification is deemed the algorithm’s prediction.

Researchers compared predictions gathered from these deep learning and random forest algorithms and compared them with one of the most commonly used early mortality predictors: the Cox model. 

Deep learning, random forest, and the Cox model

Deep learning, random forest, and the Cox model all focused heavily on variables such as age, gender, and smoking history, but the models diverged on variables such as ethnicity and physical activity – things the Cox model put a heavy emphasis on. Deep learning alone put a heavier emphasis on variables such as the use of certain medications, alcohol intake, and exposure to job related hazards and air pollution. Random forest focused on variables such as body fat percentage, waist circumference, fruit/vegetable intake, and skin tone.

The Cox model proved to be no match for the AI methods as it was only able to predict a meger 44 percent of the premature deaths that occured between 2006 and 2016 while the deep learning and random forest algorithms correctly identified 76 percent and 64 percent of subjects respectively that died during the same time period.

While these statistics are quite the accomplishments for AI, this is hardly the first time that it has been used to make medical predictions and assumptions. In 2017, researchers developed an AI algorithm that was able to analyze PET scans of potential Alzheimer’s candidates and predict with 84 percent accuracy who would go on to develop the memory impairment disease. In other studies, AI has also been used to predict onset autism, likelihood of heart attack or stroke, and has even been able to identify signs of encroaching diabetes.

Predicting mortality via AI – less scary than you'd think 

Although AI’s ability to predict early mortality appears to be rather grim reaper-esque, it is quite the opposite. Identifying people who are at risk of dying early can allow physicians to point to this likelihood and provide patients with ways to lower their chances of dying young, using AI to effectively fight off the grim reaper rather than being it.

The current state of AI is amazing and can be incredibly useful to just about every industry, but as technology continues to advance, the capabilities of AI are assuredly going to grow along with it. 

Curious about the historical advancements AI has made throughout the years? Check out the complete history of AI guide on G2. 

Related content: History of Artificial Intelligence →

AI or the Grim Reaper? Predicting Mortality Rates With AI Learn how artificial intelligence software uses deep learning, machine learning, and more to predict mortality rates in humans based on various factors.
Tucker Sutlive Tucker Sutlive is a is an AI enthusiast and writer for with a specific interest in conversational design and writing natural dialogue for chatbots. Tucker graduated with his Bachelor's degree in liberal arts from the Grady College of Journalism and Mass Communication at the University of Georgia. In his free time, he enjoys fishing, talking politics and keeping his dog Byrdie away from the mailman.

Never miss a post.

Subscribe to keep your fingers on the tech pulse.

By submitting this form, you are agreeing to receive marketing communications from G2.