Fig. 5: Our new validated thrombin concentration time-history model has better predictive performance compared to the previous model.

a Sorted correlation of coagulation factor concentrations to model parameters, from highest to lowest. Activated PC is the main driver for three of five parameters in the improved model. b Thrombin concentration time-history prediction is much improved compared to an older model⁴⁸, shown here for four edge cases of minimum peak, maximum peak, minimum peak-time, and maximum peak-time. c Mean percent error of three CAT parameters estimated using the old model and the new model for 40 trauma patient samples, dataset 5. Percent errors are calculated for each sample by comparing each CAT parameter estimated using dynamic models to the actual CAT parameters from fits of experimental data, and then the mean and standard deviation of all sample relative errors reported. For example, \(\,{{\mbox{CAT Peak percent error}}}=| \frac{{{{\mbox{Peak}}}}_{{{\rm{model}}}}-{{{\mbox{Peak}}}}_{{{\rm{experiment}}}}}{{{{\mbox{Peak}}}}_{{{\rm{experiment}}}}}| \times 100\). d Five-fold cross-validation bootstraps the data (datasets 4 and 5) and confirms that model predictions are valid with acceptable mean percent error. e CATs predicted with our improved model using an additional experimental dataset⁴⁸, dataset 8, that was not harnessed for learning. Our model is able to accurately capture both trends and magnitudes of actual CATs. Numbers in the legend indicate coagulation factor concentration reported as percent activity.