Dynamic Profiling: Modeling the Dynamics of Inflammation and Predicting Outcomes in Traumatic Brain Injury Patients

Gregory Constantine, Marius Buliga, Qi Mi, Florica Constantine, Andrew Abboud, Ruben Zamora, Ava Puccio, David Okonkwo, Yoram Vodovotz

Inflammation induced by traumatic brain injury (TBI) is complex, individual-specific, and associated with morbidity and mortality. We sought to develop dynamic, data-driven, predictive computational models of TBI-induced inflammation based on cerebrospinal fluid (CSF) biomarkers. Thirteen inflammatory mediators were determined in serial CSF samples from 27 severe TBI patients. The Glasgow Coma Scale (GCS) score quantifies the initial severity of the neurological status of the patient on a numerical scale from 3 to 15. The 6-month Glasgow Outcome Scale (GOS) score, the outcome variable, was taken as the variable to express and predict as a function of the other input variables. Data on each subject consisting of ten clinical (one-dimensional) variables, such as age, gender, and presence of infection, along with inflammatory biomarker time series were used to generate both multinomial logistic as well as probit models that predict low (poor outcome) or high (favorable outcome) levels of the GOS score. To determine if CSF inflammation biomarkers could predict TBI outcome, a logistic model for low (≤3; poor neurological outcome) or high levels (≥4; favorable neurological outcome) of the GOS score involving a full effect of the pro-inflammatory cytokine tumor necrosis factor-α and both linear and quadratic effects of the anti-inflammatory cytokine interleukin-10 was obtained. To better stratify patients as their pathology progresses over time, a technique called "Dynamic Profiling" was developed in which patients were clustered, using the spectral Laplacian and Hartigan's k-means method, into disjoint groups at different stages. Initial clustering was based on GCS score; subsequent clustering was performed based on clinical and demographic information and then further, sequential clustering based on the levels of individual inflammatory mediators over time. These clusters assess the risk of mortality of a new patient after each inflammatory mediator reading, based on the existing information in the previous data in the cluster to which the new patient belongs at the time, in essence acting as a "virtual clinician." Using the Dynamic Profiling method, we show examples that suggest that severe TBI patient neurological outcomes could be predicted as a function of time post-TBI using CSF inflammatory mediators.