<h3>Purpose</h3> Many ex vivo lung perfusion (EVLP) biomarker studies involve hourly perfusate sampling to predict lung transplant outcomes; however, the diagnostic and predictive value of repeated sampling and a biomarker kinetic profile remains unknown. In this study, we constructed mathematical models of protein biomarkers in EVLP perfusate and assessed the predictive value of the kinetic model features. <h3>Methods</h3> Clinical EVLP cases (n=45) were used in this study. Perfusate samples were collected every 15 minutes from 0-180 minutes of perfusion. Seven protein biomarkers were studied: GM-CSF, IL-10, IL-1β, IL-6, IL-8, sTNFR1, and sTREM1. All concentration data were corrected for circuit dilution using the reported standardized perfusate exchanges. Five mathematical models were tested to fit the biomarker time-series data: linear, quadratic, exponential, four- and five-parametric logistic regression. The model features were then used to predict recipient ICU length of stay using the area under the receiver operating characteristic curve (AUROC). <h3>Results</h3> The time-series data showed that GM-CSF (r<sup>2</sup>=0.92), sTNFR1 (r<sup>2</sup>=0.95), sTREM1 (r<sup>2</sup>=0.96) were best described using a linear model, whereas, IL-10 (r<sup>2</sup>=0.93), IL-1β (r<sup>2</sup>=0.92), IL-6 (r<sup>2</sup>=0.95), and IL-8 (r<sup>2</sup>=0.98) were best described by an exponential growth curve. When biomarker data were corrected for dilution, there was a significant improvement in model fit [p<0.05 (IL-8, GM-CSF); p<0.001 (IL-1β, IL-10, IL-6, sTNFR1, sTREM1)]. For linear models of GM-CSF and sTNFR1, the rate of change (slope) was predictive for ICU length-of-stay (<3days). For exponential models, the rate constant was predictive for IL-1β and IL-8, whereas both the rate constant and starting value were predictive for IL-6. Combining kinetic model features with hourly data resulted in improved diagnostic performance for GM-CSF (AUROC=88% [95%CI:72-99]), IL-10 (AUROC=81% [95%CI:61-99]), and IL-8 (AUROC=72% [95%CI:50-94]). <h3>Conclusion</h3> Kinetic modeling can be utilized to accurately reflect protein biomarker behaviour during EVLP, and features extracted from these models improve the prediction of EVLP outcomes. This study represents an important framework to guide future research to improve the predictive utility of EVLP perfusate-derived biomarkers.