C. Cockrell1, G. An1 1University Of Chicago,Surgery,Chicago, IL, USA
Introduction: There remain considerable challenges in the ability to reconstruct the behavior of human systemic inflammation and multiple organ dysfunction syndrome (MODS) from available human biomarker/mediator data,: this reconstruction is a necessary step to be able to rationally engineer putative precision therapies for sepsis/MODS. Alternatively, the increasing ability to capture physiological data in the clinical setting can provide depictions of physiological phenotypes, but offer no insight or pathway toward the design of mediator-based interventions. Other fields of science have used computational modeling and simulation to help contextualize multi-dimensional data in order to describe system function and link multiples scales of behavior. Advances in supercomputer-aided modeling can provide this capability to biomedical research, including in the area of sepsis.
Methods: We analyzed an extension of a prior agent-based model (ABM) of multiple organ dysfunction (MODS) (An, Theoretical Biology and Medical Modelling 2008, 5:11) using our developed methods of Probabilistic Basins of Attraction (PBoAs) and Stochastic Trajectory Analysis (STAs) (Cockrell and An, Journal of Theoretical Biology 2017 (430):157-168). 2 billion patients were simulated representing a 90 day hospital course due to microbial sepsis arising from the following initial insults: 1) pneumonia, 2) urosepsis and 3) primary septicemia. Parameter space was coarse-grained for pulmonary, cardiac, renal, hepatic and gastrointestinal physiology, and linked to SOFA scores when applicable. Linear classifiers were applied to determine if simulated patients could be classified on biomarker state and physiological score within the 1st 72 h with samples q6 h.
Results: Boundary conditions for parameter sets were identified corresponding to plausible patterns of MODS arising from different sources of initial infection, but with a probability distribution of atypical MODS patterns for each type of initial infection. PBoAs of specific organ systems demonstrated tipping points at which individual organ system dysfunction extended to MODS, and displayed a hierarchical effect separating implemented organ physiology/function from blood-borne cytokine mediated endothelial dysfunction. Characterization of biomarker system state was unable to separate survivors/non-survivors (AUROC .61), but the addition of physiological state improved this (AUROC .91).
Conclusion: Supercomputing simulations of MODS can be used to generate higher-order physiological behavioral manifolds that provide a path towards linking cellular-molecular mechanistic models with clinically-available physiological data. Modeling and simulation of this type can be used to provide a bridge between sparse biomarker/mediator data and more readily accessible physiological signal data, and aid in the engineering of truly predictive models and precision therapies.