David L. Carey

David L. Carey

Sports Analytics Lecturer, La Trobe University

Research

This is an abbreviated list of publications - you can find all of my research papers on my Google Scholar profile.

In most cases I will reply to full text requests sent via Researchgate (or other channels).

Purpose: To evaluate common modeling strategies in training load and injury risk research when modeling continuous variables and interpreting continuous risk estimates; and present improved modeling strategies. Method: Workload data were pooled from Australian football (n = 2550) and soccer (n = 23,742) populations to create a representative sample of acute:chronic workload ratio observations for team sports. Injuries were simulated in the data using three predefined risk profiles (U-shaped, flat and S-shaped). One-hundred data sets were simulated with sample sizes of 1000 and 5000 observations. Discrete modeling methods were compared with continuous methods (spline regression and fractional polynomials) for their ability to fit the defined risk profiles. Models were evaluated using measures of discrimination (area under receiver operator characteristic [ROC] curve) and calibration (Brier score, logarithmic scoring). Results: Discrete models were inferior to continuous methods for fitting the true injury risk profiles in the data. Discrete methods had higher false discovery rates (16%-21%) than continuous methods (3%-7%). Evaluating models using the area under the ROC curve incorrectly identified discrete models as superior in over 30% of simulations. Brier and logarithmic scoring was more suited to assessing model performance with less than 6% discrete model selection rate. Conclusions: Many studies on the relationship between training loads and injury that have used regression modeling have significant limitations due to improper discretization of continuous variables and risk estimates. Continuous methods are more suited to modeling the relationship between training load and injury. Comparing injury risk models using ROC curves can lead to inferior model selection. Measures of calibration are more informative judging the utility of injury risk models.

Recommended citation: Carey, D. L., Crossley, K. M., Whiteley, R., Mosler, A., Ong, K. L., Crow, J., & Morris, M. E. (2018). Modeling Training Loads and Injuries: The Dangers of Discretization. Medicine and science in sports and exercise, 50(11), 2267-2276.

To investigate whether training load monitoring data could be used to predict injuries in elite Australian football players, data were collected from athletes over 3 seasons at an Australian football club. Loads were quantified using GPS devices, accelerometers and player perceived exertion ratings. Absolute and relative training load metrics were calculated for each player each day. Injury prediction models (regularised logistic regression, generalised estimating equations, random forests and support vector machines) were built for non-contact, non-contact time-loss and hamstring specific injuries using the first two seasons of data. Injury predictions were then generated for the third season and evaluated using the area under the receiver operator characteristic (AUC). Predictive performance was only marginally better than chance for models of non-contact and non-contact time-loss injuries (AUC<0.65). The best performing model was a multivariate logistic regression for hamstring injuries (best AUC=0.76). Injury prediction models built using training load data from a single club showed poor ability to predict injuries when tested on previously unseen data, suggesting limited application as a daily decision tool for practitioners. Focusing the modelling approach on specific injury types and increasing the amount of training observations may improve predictive models for injury prevention

Recommended citation: Carey, D. L., Ong, K., Whiteley, R., Crossley, K. M., Crow, J., & Morris, M. E. (2018). Predictive modelling of training loads and injury in Australian football. International Journal of Computer Science in Sport, 17(1), 49-66.

Purpose: To investigate whether preseason training plans for Australian football can be computer generated using current training-load guidelines to optimize injury-risk reduction and performance improvement. Methods: A constrained optimization problem was defined for daily total and sprint distance, using the preseason schedule of an elite Australian football team as a template. Maximizing total training volume and maximizing Banister-model-projected performance were both considered optimization objectives. Cumulative workload and acute:chronic workload-ratio constraints were placed on training programs to reflect current guidelines on relative and absolute training loads for injury-risk reduction. Optimization software was then used to generate preseason training plans. Results: The optimization framework was able to generate training plans that satisfied relative and absolute workload constraints. Increasing the off-season chronic training loads enabled the optimization algorithm to prescribe higher amounts of “safe” training and attain higher projected performance levels. Simulations showed that using a Banister-model objective led to plans that included a taper in training load prior to competition to minimize fatigue and maximize projected performance. In contrast, when the objective was to maximize total training volume, more frequent training was prescribed to accumulate as much load as possible. Conclusions: Feasible training plans that maximize projected performance and satisfy injury-risk constraints can be automatically generated by an optimization problem for Australian football. The optimization methods allow for individualized training-plan design and the ability to adapt to changing training objectives and different training-load metrics.

Recommended citation: Carey, D. L., Crow, J., Ong, K. L., Blanch, P., Morris, M. E., Dascombe, B. J., & Crossley, K. M. (2018). Optimizing preseason training loads in Australian Football. Int J Sports Physiol Perform, 13(2), 194-199.

Aims (1) To investigate whether a daily acute:chronic workload ratio informs injury risk in Australian football players; (2) to identify which combination of workload variable, acute and chronic time window best explains injury likelihood. Methods Workload and injury data were collected from 53 athletes over 2 seasons in a professional Australian football club. Acute:chronic workload ratios were calculated daily for each athlete, and modelled against non-contact injury likelihood using a quadratic relationship. 6 workload variables, 8 acute time windows (2–9 days) and 7 chronic time windows (14–35 days) were considered (336 combinations). Each parameter combination was compared for injury likelihood fit (using R2). Results The ratio of moderate speed running workload (18–24 km/h) in the previous 3 days (acute time window) compared with the previous 21 days (chronic time window) best explained the injury likelihood in matches (R2=0.79) and in the immediate 2 or 5 days following matches (R2=0.76–0.82). The 3:21 acute:chronic workload ratio discriminated between high-risk and low-risk athletes (relative risk=1.98–2.43). Using the previous 6 days to calculate the acute workload time window yielded similar results. The choice of acute time window significantly influenced model performance and appeared to reflect the competition and training schedule. Conclusions Daily workload ratios can inform injury risk in Australian football. Clinicians and conditioning coaches should consider the sport-specific schedule of competition and training when choosing acute and chronic time windows. For Australian football, the ratio of moderate speed running in a 3-day or 6-day acute time window and a 21-day chronic time window best explained injury risk.

Recommended citation: Carey, D. L., Blanch, P., Ong, K. L., Crossley, K. M., Crow, J., & Morris, M. E. (2017). Training loads and injury risk in Australian football—differing acute: chronic workload ratios influence match injury risk. British journal of sports medicine, 51(16), 1215-1220.

The ability of machine learning techniques to predict athlete ratings of perceived exertion (RPE) was investigated in professional Australian football players. RPE is commonly used to quantifying internal training loads and manage injury risk in team sports. Data from global positioning systems, heart-rate monitors, accelerometers and wellness questionnaires were recorded for each training session (n=3398) from 45 professional Australian football players across a full season. A variety of modelling approaches were considered to investigate the ability of objective data to predict RPE. Models were compared using nested cross validation and root mean square error (RMSE) on RPE predictions. A random forest model using player normalised running and heart rate variables provided the most accurate predictions (RMSE ± SD = 0.96 ± 0.08 au). A simplification of the model using only total distance, distance covered at speeds between 18-24 km·h−1, and the product of total distance and mean speed provided similarly accurate predictions (RMSE ± SD = 1.09 ± 0.05 au), suggesting that running distances and speeds are the strongest predictors of RPE in Australian football players. The ability of non-linear machine learning models to accurately predict athlete RPE has applications in live player monitoring and training load planning.

Recommended citation: Carey, D. L., Ong, K., Morris, M. E., Crow, J., & Crossley, K. M. (2016). Predicting ratings of perceived exertion in Australian football players: methods for live estimation. International Journal of Computer Science in Sport, 15(2), 64-77.

We study the buckling instability of a colloidal particle layer adhered to an elastic substrate using an integrated experimental and theoretical approach. Experiments using monodisperse colloid-scale spherical particles made of polystyrene and silica, show that the wavelength of the initial (critical) buckling mode is independent of particle modulus and linearly dependent on particle radius—in contradiction with the predictions of the prevailing continuum model. We developed a granular model of the particle layer using structural mechanics techniques. The granular model predicts the observed wavelength of the initial, critical buckling mode within the estimated range of parameter values for the experiment. The evolution of this mode into the post-buckling regime is examined. Results highlight the crucial role of material discreteness in the mechanical response, and the need for accurate methods of estimating parameters for the particle-scale resistances against buckling.

Recommended citation: Tordesillas, A., Carey, D., Croll, A. B., Shi, J., & Gurmessa, B. (2014). Micromechanics of elastic buckling of a colloidal polymer layer on a soft substrate: experiment and theory. Granular Matter, 16(2), 249-258.