Researchers use AI to predict the toll of concussions on student athletes over time

Researchers at the University of Michigan are using artificial intelligence to predict the health consequences that sport-related concussions might have on student athletes over the course of their college athletic careers. 

The study, which appears in the Annals of Biomedical Engineering, found that five different AI models outperformed a simple benchmark model when predicting the progression of possible health ramifications from concussions, and revealed multiple surprising findings about concussion management.

"To our knowledge, this is the first research that leverages AI to predict the change in key clinical outcomes for a concussed athlete after college play is over," said study first author Lauren Czerniak, who conducted the interdisciplinary research at the Michigan Concussion Center and U-M College of Engineering while earning her Ph.D. in industrial and operations engineering in 2023. 

"This is an important topic as there is a growing concern about the effects of sport-related concussions after college play is over, and there is a lot of uncertainty around when the effects of sport-related concussions begin to present clinically." 

The research was funded by the U.S. Department of Defense, the NCAA, the National Science Foundation Research Fellowship Program, the U-M Summer Undergraduate Research in Engineering Program and U-M's Rackham Graduate School.

For this study, Czerniak and colleagues wanted to know if AI could predict three clinical outcomes: the impact on overall quality of life (symptom burden); cognitive status; and risk of psychological distress. They also wanted to know which AI models predicted those outcomes well, and what factors––such as frequency and intensity of concussion and the sports played (sport exposure)––influenced the progression of those outcomes over an athlete's career. 

The researchers used baseline and exit data from about 3,200 NCAA athletes in the Concussion, Assessment, Research, and Education Consortium, a national concussion research network of NCAA athletes and U.S. military service academy cadets across 30 different institutions. In addition to the CARE Consortium data, researchers fed the models with demographic as well as health information collected during the athlete's college career. 

Key findings: 

• AI helped avoid large prediction errors compared to a simple model that predicted no change. 

• The easy-to-explain AI models outperformed the more complex AI models. 

• An athlete's baseline evaluation––the standard testing at the beginning of an athlete's career––was the most important factor in making accurate predictions. 

• Sport exposure had little to no impact on symptom progression.

• Frequency and intensity of concussions had little to no impact on an athlete’s symptom progression. 

The researchers were surprised by these last two findings––especially that the frequency and intensity of the number of concussions had little to no impact. 

"When initially designing the study, our research team hypothesized that the frequency and intensity of concussions would play the biggest role in an athlete's progression over their collegiate career," Czerniak said. "AI allowed us to tease out the important predictors and demonstrate that our hypothesis was incorrect."

Concussion is one of the most complex injuries in sports medicine, and many high school and college athletes experience one or many concussions while playing. Among the NCAA athletes at CARE, it's estimated that around 10% have experienced one to several sport-related concussions. 

Study co-author Steven Broglio, director of the Michigan Concussion Center and professor at the U-M School of Kinesiology, believes AI can play a big role in concussion diagnosis and management. 

"There is a growing interest in understanding the long-term effects of concussion and head impact exposure, with a particular focus on athletes and military service members," said Broglio, who is also president of the International Concussion in Sport Group. "These analyses give us insight into those effects using sophisticated AI algorithms. These predictions will give insight into the clinical care of athletes and service members to protect their long-term health and safety."

Czerniak says future studies could entail larger and more diverse populations. Given enough data is available, there is potential to embed AI into a software application that uses an athlete's or patient's clinical record to predict areas of progression so that proactive measures or monitoring could be taken. 

Czerniak was inspired to do concussion research after two concussions left her with the sense of "I do not feel right"––along with weathering the typical highs and lows of recovery. 

"It really inspired me to get involved in concussion research to give back to the community that helped me recover from my first and then second concussion," she said. "The knowledge obtained through concussion management research continues to grow, and I think that intersecting concussion management research with AI/operations research will only bring more understanding to this complex field." 

Co-authors include: Max Genthe, Yeuyun Xia and Mariel Lavieri, Georgia Institute of Technology; Michael McCrea, Medical College of Wisconsin; Thomas McAllister, Indiana University School of Medicine; Paul Pasquina and Spencer Liebel, Uniformed Services University. 

Study: Prediction of Symptom Burden, Cognitive Status, and Risk of Psychological Distress in NCAA Athletes with Sport-Related Concussion(s): Findings from the NCAA-DoD CARE Consortium (DOI: 10.1007/s10439-025-03824-w)