Mining standardized neurological signs and symptoms data for concussion identification

Area of Science:

• Sports Medicine
• Neurology
• Data Science

Background:

• Millions of concussions occur annually in sports and recreation.
• Concussions elevate risks for subsequent injuries and cognitive impairments.
• Predictors of concussion outcomes and recovery remain unclear.

Purpose of the Study:

• To identify factors most predictive of concussion diagnosis.
• To mitigate physician bias in concussion assessment.
• To develop a robust concussion prediction model.

Main Methods:

• Utilized multivariate logistic regression analysis.
• Analyzed a dataset of 126 subjects aged 12-31.
• Evaluated 322 potential predictive features.

Main Results:

• Selected 27-29 features as highly predictive of concussions.
• Identified features include sports history, prior concussion, drowsiness, nausea, focus issues, and oculomotor function.
• Achieved prediction accuracy >90%, Matthews correlation coefficient >0.8, and AUC >0.95.

Conclusions:

• The multivariate model effectively predicts concussion diagnosis.
• Specific symptoms and history are strong indicators of concussion.
• This approach offers an objective tool for concussion assessment.