R. C. Cockrell1, G. An1 1University Of Chicago,Surgery,Chicago, IL, USA
Introduction: Current predictive models for sepsis generally use correlative methods, and as such are limited in their individual precision due to patient heterogeneity and data sparseness. In particular, the inability of these methods to capture the clinical dynamics and mechanisms of sepsis limits achieving true prediction. Other fields of science have used computational modeling and simulation to help contextualize multi-dimensional data arising from complex systems in order to describe their behavior. Advances in supercomputer-aided modeling can provide this same capability to biomedical research. Herein we begin this process by defining the behavioral possibility space for sepsis across a range of physiological, microbial and environmental parameters, and apply advanced mathematical analysis to identify predictive metrics for system outcome.
Methods: 80 million microbial sepsis patients were simulated representing a 28-day hospital course using a previously validated agent-based model (ABM) of sepsis implemented on a Cray XE6 supercomputer. Parameter space was examined regarding system response with and without antibiotics regarding the following parameters: cardio-respiratory-metabolic resilience; two properties of microbial virulence, invasiveness and toxigenesis; and degree of contamination from the environment. Simulation data was analyzed using varied tiers of multi-dimensional space, which was then subjected to mathematical analysis using deterministic chaos and bifurcation theory.
Results: The identified parameter space had a clear structure with plausible boundaries reflecting the range of possible human behavior in response to microbial sepsis (Fig 1). The central region of highest outcome uncertainty corresponds to the critically ill population. Analysis of this region using chaos and bifurcation theory identified basins of attraction leading to three outcomes – complete recovery, hyperinflammatory system failure, and overwhelming infection – in a 32-dimensional phase space defined by various chemokine concentrations and their conjugate generalized momenta. These basins of attraction were used to develop metrics for prediction of individual outcomes based on networks of mediator and cell population levels.
Conclusion: Supercomputing simulations of sepsis can play a vital role in the contextualization of both big data output and mechanistic basic science research. The representation of the aggregate set of individual trajectories facilitates formal mathematical analysis and is a necessary step towards developing truly predictive models and precision therapies.
