PhD Proposal: Building Context-Aware Systems for Toxicovigilance

Traditionally, the thought of an opioid overdose has conjured images of an unconscious patient on the floor with a syringe and a quantity of drugs. Due to recent epidemic of opioid abuse and deaths, many patients have begun to appear in the emergency room with injuries differing from those traditionally associated with opioid abuse. Common injuries associated with opioid abuse have begun to include head trauma, rhabdomyolysis, and even car accidents. Sometimes, the injuries caused by the opioid overdose are so severe, that the medical professionals do not even notice that an opioid overdose was the root cause of the injuries. This is particularly common in cases where the patient has been involved in a car accident due to driving under the influence of opioids.
Despite the plethora of data and tools available to identify potential opioid abuse and begin early intervention, politics, data-formatting, and access to data sources present large obstacles for researchers and medical professionals alike. Additionally, due to the medical nature of the problem, opioid abuse has traditionally been thought of as a public health problem. This mindset has possibly contributed to the fact that scale of research on this topic has remained relatively stagnant in contrast to the increase in opioid-related abuse, overdoses, and deaths. Many public health research groups lack the technical expertise to fully utilize current data-mining and machine learning techniques. As a consequence, many of these groups still perform data mining by hand, which not only increases the time required to identify trends in opioid abuse, but also introduces possibilities for error, and increases cost.
To address these problems, we frame the issue of opioid abuse detection as a context-aware systems problem, and present a context for addressing the problem at a national, community and patient level. The success of this work relies not only on the ability to process large amounts of data, but also to impose relevant domain knowledge. We rely on the expertise of medical doctors, EMTs, and public health researchers to help give the data a context for analysis.
Our goal is three fold. First, we wish to apply current data-mining and machine learning techniques within the domain of public health. Second, we wish to build a context aware system for performing toxicovigilance using both publicly available and privately held data sources, by laying down a frame work for combining multiple disparate data sources, and inferring the relative strengths and weaknesses of each source. Finally, we wish to use this context aware system to augment existing medical and law-enforcement practices, in order to help identify persons and communities in need of early intervention, without any change to existing medical or law-enforcement practices.
Examining Committee:
Chair: Dr. Ashok Agrawala
Dept rep: Dr. Atif Memon
Member: Dr. Jay Unick