Yan Chu1, Gregory Knell2, Riley P Brayton2, Scott O Burkhart3, Xiaoqian Jiang1, Shayan Shams4. 1. School of Biomedical Informatics, The University of Texas Health Science Center at Houston (UTHealth), Houston, TX, USA. 2. Department of Epidemiology, Human Genetics, and Environmental Sciences, The University of Texas Health Science Center at Houston (UTHealth), Dallas, TX, USA; Children's Health and The University of Texas Health Science Center at Houston (UTHealth), Dallas, TX, USA; Children's Health Andrews Institute for Orthopaedics and Sports Medicine, Plano, TX, USA. 3. Children's Health Andrews Institute for Orthopaedics and Sports Medicine, Plano, TX, USA. 4. School of Biomedical Informatics, The University of Texas Health Science Center at Houston (UTHealth), Houston, TX, USA; Department of Applied Data Science, San Jose State University, San Jose, CA, USA. Electronic address: Shayan.Shams@sjsu.edu.
Abstract
OBJECTIVES: Sport-related concussions (SRCs) are a concern for high school athletes. Understanding factors contributing to SRC recovery time may improve clinical management. However, the complexity of the many clinical measures of concussion data precludes many traditional methods. This study aimed to answer the question, what is the utility of modeling clinical concussion data using machine-learning algorithms for predicting SRC recovery time and protracted recovery? METHODS: This was a retrospective case series of participants aged 8 to 18 years with a diagnosis of SRC. A 6-part measure was administered to assess pre-injury risk factors, initial injury severity, and post-concussion symptoms, including the Vestibular Ocular Motor Screening (VOMS) measure, King-Devick Test and C3 Logix Trails Test data. These measures were used to predict recovery time (days from injury to full medical clearance) and binary protracted recovery (recovery time > 21 days) according to several sex-stratified machine-learning models. The ability of the models to discriminate protracted recovery was compared to a human-driven model according to the area under the receiver operating characteristic curve (AUC). RESULTS: For 293 males (mean age 14.0 years) and 362 females (mean age 13.7 years), the median (interquartile range) time to recover from an SRC was 26 (18-39) and 21 (14-31) days, respectively. Among 9 machine-learning models trained, the gradient boosting on decision-tree algorithms achieved the best performance to predict recovery time and protracted recovery in males and females. The models' performance improved when VOMS data were used in conjunction with the King-Devick Test and C3 Logix Trails Test data. For males and females, the AUC was 0.84 and 0.78 versus 0.74 and 0.73, respectively, for statistical models for predicting protracted recovery. CONCLUSIONS: Machine-learning models were able to manage the complexity of the vestibular-ocular motor system data. These results demonstrate the clinical utility of machine-learning models to inform prognostic evaluation for SRC recovery time and protracted recovery.
OBJECTIVES: Sport-related concussions (SRCs) are a concern for high school athletes. Understanding factors contributing to SRC recovery time may improve clinical management. However, the complexity of the many clinical measures of concussion data precludes many traditional methods. This study aimed to answer the question, what is the utility of modeling clinical concussion data using machine-learning algorithms for predicting SRC recovery time and protracted recovery? METHODS: This was a retrospective case series of participants aged 8 to 18 years with a diagnosis of SRC. A 6-part measure was administered to assess pre-injury risk factors, initial injury severity, and post-concussion symptoms, including the Vestibular Ocular Motor Screening (VOMS) measure, King-Devick Test and C3 Logix Trails Test data. These measures were used to predict recovery time (days from injury to full medical clearance) and binary protracted recovery (recovery time > 21 days) according to several sex-stratified machine-learning models. The ability of the models to discriminate protracted recovery was compared to a human-driven model according to the area under the receiver operating characteristic curve (AUC). RESULTS: For 293 males (mean age 14.0 years) and 362 females (mean age 13.7 years), the median (interquartile range) time to recover from an SRC was 26 (18-39) and 21 (14-31) days, respectively. Among 9 machine-learning models trained, the gradient boosting on decision-tree algorithms achieved the best performance to predict recovery time and protracted recovery in males and females. The models' performance improved when VOMS data were used in conjunction with the King-Devick Test and C3 Logix Trails Test data. For males and females, the AUC was 0.84 and 0.78 versus 0.74 and 0.73, respectively, for statistical models for predicting protracted recovery. CONCLUSIONS: Machine-learning models were able to manage the complexity of the vestibular-ocular motor system data. These results demonstrate the clinical utility of machine-learning models to inform prognostic evaluation for SRC recovery time and protracted recovery.
Authors: Deborah Jacob; Ingunn S Unnsteinsdóttir Kristensen; Romain Aubonnet; Marco Recenti; Leandro Donisi; Carlo Ricciardi; Halldór Á R Svansson; Sólveig Agnarsdóttir; Andrea Colacino; María K Jónsdóttir; Hafrún Kristjánsdóttir; Helga Á Sigurjónsdóttir; Mario Cesarelli; Lára Ósk Eggertsdóttir Claessen; Mahmoud Hassan; Hannes Petersen; Paolo Gargiulo Journal: Sci Rep Date: 2022-05-30 Impact factor: 4.996