Bayesian methods for estimating injury rates in sport injury epidemiology

Figures

• 

  Fig. 1. Maxima posterior predictive check. The example shown is for all diagnoses reported during competition events in Division I women’s soccer. Each point represents a specific diagnosis, with the y-axis indicating the proportion of simulations where the injury value exceeded the observed maximum, scaled as a percentage. The figure is faceted by model type to illustrate results from various model configurations. NB indicates the negative binomial model, and ZINB indicates the zero-inflated negative binomial model

• 

  Fig. 2. Observed proportion of zero injury values and the corresponding 95% uncertainty intervals for the simulated data. The example depicted shows all diagnoses reported during competition events in Division I men’s soccer. Each point represents a specific diagnosis, with the y-axis indicating the proportion of zero injuries, scaled as a percentage. The solid black circles denote observed proportions, while the grey lines indicate the 95% credible intervals from the simulated data. The figure is faceted by model type to compare results from various model configurations. NB indicates the negative binomial model, and ZINB indicates the zero-inflated negative binomial model

• 

  Fig. 3. Actual and posterior predicted injury rates for sprain diagnoses reported during competition events in Division I men’s soccer. The y-axis shows the predicted injury rate and the x-axis shows the actual injury rate. The dashed line represents the line of equality, while the solid line is the linear regression fit (fitted by regressing the posterior medians on the actual injury rates). Error bars indicate the 90% credible intervals for both actual and predicted injury rates. The figure is faceted by model type to compare results across various model configurations. NB indicates the negative binomial model, and ZINB indicates the zero-inflated negative binomial model

• 

  Fig. 4. The plot illustrates the empirical cumulative distribution functions (eCDFs) of observed and simulated data for concussion diagnoses reported in Division I women's soccer. The x-axis represents the eCDF of a uniform distribution, and the y-axis represents the eCDF of the PIT-transformed data. Shaded ribbons indicate the 95% credible intervals for the simulated data, while solid lines represent the median PIT values. The dashed line represents the line of equality. The figure is faceted by model type to compare results across various model configurations. NB indicates the negative binomial model, and ZINB indicates the zero-inflated negative binomial model

• 

  Fig. 5. Sample workflow describing analytical approach for Bayesian estimations of injury rate using sport injury surveillance data [18]