Subarachnoid Hemorrhage (SAH) affects an estimated 14.5 per 100,000 persons in the United States, and is a substantial burden on health care resources, because it can cause long-term functional and cognitive disability. Much of this is due to delayed cerebral ischemia (DCI) from vasospasm (VSP). VSP refers to the reactive narrowing of cerebral blood vessels due the unusual presence of blood surrounding the vessel. In its extreme, severe VSP precludes blood flow to brain tissue, resulting in stroke. SAH is one of the most common disease entities treated in the Neurointensive Care Unit (NICU). Currently, resource planning is scripted around the Modified Fisher Scale, which predicts the odds ratio of developing DCI based on the volume and pattern of blood on initial brain computed tomography (CT). It does not, however, allow for further individualized risk assessments. The first 14 days are occupied by efforts to detect preclinical or early VSP and arrange timely interventions to prevent permanent injury. The only noninvasive tool supported by guidelines to potentially identify preclinical VSP is the transcrania Doppler (TCD), which has an unreliable range of sensitivity and negative predictive values, and is at the mercy of technician availability. If not identified preclinically, VSP must be detected once it is symptomatic and is then dependent on quality and availability of expertise in the complex and diurnal environment of the ICU. Promisingly, electronic medical record (EMR) data and continuous physiology monitors offer abundant opportunities to risk stratify for future events as well as reveal events in real-time in the acutely brain injured patient. A methodical approach to feature engineering will be performed over a large set of potentially discriminatory data-driven and knowledge-based features. Meta-features representing variations and trends in time series variables will be extracted using a variety of quantitative and symbolic abstraction techniques. Predictive modeling will be performed using Nave Bayes, Logistic Regression, and Support Vector Machine. This project will result in a prediction tool that improves timeliness and precision in VSP classification. It will fill an important gap in the understanding of the potentia of underutilized EMR and physiological data to predict neurological decline. Generating accurate and timely prediction rules from already collected clinical data would be cost effective and have implications not only for SAH patients, but also for almost any monitored patient in any ICU.
This project will explore the optimal methods for creating a prediction tool that improves timeliness and precision of diagnosis. It will fill an important gap in the understanding of the underutilized potential of electronic medical record and high frequency device monitor data. Generating timely and accurate prediction rules from already collected clinical data would be cost effective and have implications for almost any monitored patient in any ICU.
Showing the most recent 10 out of 18 publications
