A decision tree framework for understanding blast-induced mild Traumatic Brain Injury in a military medical database. (October 2017)

• Record Type: Journal Article
• Authors: Walker, Peter B; Mehalick, Melissa L; Glueck, Amanda C; Tschiffely, Anna E; Cunningham, Craig A; Norris, Jacob N; Davidson, Ian N
• Is Part Of: Journal of defense modeling and simulation. Volume 14:Number 4(2017:Oct.)
• Page Start: 389
• Page End: 398
• Publication Date: 2017-10
• Subjects: Biomedical informatics -- graph theory -- classification algorithms -- clustering algorithms; Military art and science -- Computer simulation -- Periodicals; 355.0011305
• DOI: 10.1177/1548512916683841
• Languages: English
• ISSNs: 1548-5129

Abstract

Personalized medicine is a ubiquitous term that has come to be used to describe a medical model that proposes the customization of healthcare, such that decisions and/or treatments are tailored to each individual patient. Under this type of clinical practice model, diagnostic and prognostic decisions are often based upon selecting the most appropriate therapy based on a patient's genetic, demographic, and/or other pertinent information. The primary aim of this paper is to use a personalized medicine framework to better understand the relationship between neuropsychological testing and the progression of symptoms in a blast-induced mild Traumatic Brain Injury (mTBI) patient population. In this paper, we extended our earlier work on Constrained Spectral Partitioning (CSP), a graph-based approach that incorporates additional information from separate graphs to help improve the clustering quality on both graphs simultaneously. While our previous work demonstrated the effectiveness of this algorithm in its ability to accurately classify whether symptoms improved or declined over time, that work did not provide any insights into the progression of symptoms. Therefore, this paper sought to identify, from a clinical perspective, whether symptoms increased/decreased over time. To accomplish this, we developed a decision tree classifier to classify symptom progression based on the outputs from our CSP algorithm. We present results from four separate decision tree classifiers that illustrate the adaptability of these algorithms for utilization as decision rules for the treatment of patients following blast-induced mTBI. Decision tree classifier models are useful in the healthcare setting because patient health data (e.g., diagnosis of a condition or a type of treatment) can be imput into the model and, based on the health data variables, a resulting outcome can be suggested, and providers can use this outcome as information to direct their clinical treatment.