Characterizing Allegheny County Opioid Overdoses with an Interactive Data Explorer and Synthetic Prediction Tool

TL;DR

This paper explores opioid overdose factors in Allegheny County using autopsy data, introduces an interactive tool for policymakers, and demonstrates how synthetic EMR data can improve predictive models while maintaining privacy.

Contribution

It presents an interactive data exploration tool for policy use and evaluates synthetic EMR data to enhance overdose risk prediction models.

Findings

Autopsy data reveals key overdose risk factors.

Synthetic EMR data improves model accuracy.

Feature extraction methods can enhance predictions without compromising privacy.

Abstract

The United States has an opioid epidemic, and Pennsylvania's Allegheny County is among the worst. This motivates a deeper exploration of what characterizes the epidemic, such as what are risk factors for people who ultimately overdose and die due to opioids. We show that some interesting trends and factors can be identified from openly available autopsy data, and demonstrate the power of building an interactive data exploration tool for policy makers. However, there is still a pressing need to incorporate further demographic factors. We show this by using synthetic Electronic Medical Record (EMR) data to simulate the predictive power of random forests and neural networks when given additional loosely correlated features. In addition, we give examples of useful feature extraction that enable model enhancement without sacrificing privacy.