Whole-Body Reactive Agility Metrics to Identify Football Players With a Core and Lower Extremity Injury Risk

Abstract

Clinical prediction models are useful in addressing several orthopedic conditions with various cohorts. American football provides a good population for attempting to predict injuries due to their relatively high injury rate. Physical performance can be assessed a variety of ways using an assortment of different tests to assess a diverse set of metrics, which may include reaction time, speed, acceleration, and deceleration. Asymmetry, the difference between right and left performance has been identified as a possible risk factor for injury. The purpose of this study was to determine the whole-body reactive agility metrics that would identify Division I football players who were at elevated risk for core, and lower extremity injuries (CLEI). This cohort study utilized 177 Division I football players with a total of 57 CLEI suffered who were baseline tested prior to the season. Single-task and dual-task whole-body reactive agility movements in lateral and diagonal direction reacting to virtual reality targets were analyzed separately. Receiver operator characteristic (ROC) analyses narrowed the 34 original predictor variables to five variables. Logistic regression analysis determined the three strongest predictors of CLEI for this cohort to be: lateral agility acceleration asymmetry, lateral flanker deceleration asymmetry, and diagonal agility reaction time average. Univariable analysis found odds ratios to range from 1.98 to 2.75 for these predictors of CLEI. ROC analysis had an area under the curve of 0.702 for any combination of two or more risk factors produced an odds ratio of 5.5 for risk of CLEI. These results suggest an asymmetry of 8–15% on two of the identified metrics or a slowed reaction time of ≥0.787 s places someone at increased risk of injury. Sixty-three percent (36/57) of the players who sustained an injury had ≥2 positive predictors In spite of the recognized limitation, these finding support the belief that whole-body reactive agility performance can identify Division I football players who are at elevated risk for CLEI.

Developing a High-Performance Sports Results Prediction Artificial Neural Network: Case Study on World Championship Boxing

“…A variety of models were developed to predict injury or performance using different data analytics tools. Linear regression was utilized by three articles (Bacon et al 2016;Gaudino et al 2015;Guerrero-Calderon et al 2021;), SVM (n=1) (Di Credico et al 2021), Decision Trees/ RF (n=3) (Geurkink et al 2021;Gimenez et al 2020;Rossi et al 2018), ML (n=3) ( He et al 2021;Rommers et al 2020;Rossi et al 2019), logistic regression (n=3) (Klemp et al 202;Philp et al 2020;Zumeta-Olaskoaga et al 2021), three studies combined two methods (Goes et al 2021;Jaspers et al 2018;Oliver et al 2020;Perri et al 2021) and 11 studies utilized more than three tools (Bruce & Wilkerson 2021;Di Cesare et al 2020;Dijkhuis et al 2021;Fanchini et al 2018;Gasparini & Alvaro 2020;Geurkink et al 2019;Jaspers et al 2018;Jauhiainen et al 2019;Mandorino et al 2021;Mandorino et al 2021;Vallance et al 2020) in which no similar predictive tools were included in each methodology. Therefore, although several prediction and modelling strategies have been proposed for injury ✰✪✲ ✸✬✳✻✱✳☎✰✪✹✬ ☎✱✲✬✶✶✩✪✯✂ ✪✱ ✵✯✱✶✲ ✭✮✰✪✲✰✳✲✷ ✰✸✸✳✱✰✹✄ ✹✰✪ ☞✬ ✰✲✫✱✹✰✮✬✲✞ ✭✮✁✲✺ ☞✺ ✟✄ilp (2020) in which the aim was to investigate the effect of using zero-inflated Poisson to improve ✩✪✌✁✳✺ ✸✳✬✲✩✹✮✩✱✪ ☎✱✲✬✶✭ ✩✪ ✭✱✹✹✬✳ ✄✰✲ ✄✱✸✬✲ ✮✱ ✹✱☎✸✰✳✬ ✩✮✷✭ ✳✬✭✁✶✮✭ ✰✯✰✩✪✭✮ ✮✄✬ ✬✁✩✭✮✩✪✯ ✶✩✮✬✳✰✮✁✳✬ available but this was not possible as 1) the datasets had significant differences with regards to independent and dependent variables being investigated, 2) there was variance in injury classification and documentation and 3) modelling methods utilized displayed limited explanatory validity and clinical applicability (Philp et al 2020).…”

“…With regards to injury tracking, studies included previous injury data collected over previous competitive seasons or preseason (Bacon et al 2016;Bruce & Wilkerson et al 2021;Carey et al 2018;Colby et al 2018;Gabbett 2010;Mandorino et al 2021;Mason et al 2021;Rommers et al 2020;Rossi et al 2018;Thornton et al 2017;Vallance et al 2020;Wilkerson et al 2018;Zumeta-Olaskoaga et al 2021). These injuries were usually tracked by a medical expert such as a physiotherapist or athletic trainer with the clubs.…”

“…Two forms of predictive analytics were combined by 13 studies, combinations included: Generalized Estimating Equations (GEE) and Artificial Neural Networks (ANN) (Bartlett et al 2016), linear Regression and GEE (Carey et al 2018), GEE and logistic regression (Colby et al 2017), ROC and RF (Chambers et al 2018), logistic regression andROC (Di Cesare et al 2020;Wilkerson et al 2016;Wilkerson et al 2018), RF and Decision Trees (DT) (Dijkuis et al 2021), GEE and RF (Gastin et al 2019), Support Vector Machine (SVM) and ROC (Jauhiainen et al 2019), logistic regression and ROC (Mason et al 2021), linear regression and SVM (Revie et al 2017) and logistic regression and ANN (Schmid et al 2021). The incorporation of 3 or more predictive analytics models to investigate prediction strategies for performance and injury in team-based field sport was used by 12 studies (Bruce & Wilkerson et al 2021;Carey et al 2016;Carey et al 2018;Colby et al 2018;Fanchini et al 2018;Gasparini & Alvaro 2020;Geurkink et al 2019;Mandorino et al 2021;Mandorino et al 2021;Reyers & Swartz et al 2021, Rowson & Duma 2013Thornton et al 2016; ROC, RF and logistic regressions were the most common models developed for injury and performance prediction among field-based team sports. These models have been used collectively and separately throughout these studies.…”