Early Detection of Opioid Over-Procurement: A Semisupervised Machine Learning Approach

Problem definition: The misuse of prescription opioids over the past two decades has emerged as a leading factor contributing to the decline in life expectancy in the United States. This study introduces a machine learning model designed to identify instances of illegal diversion within the prescription drug supply chain in the United States. Methodology/results: We introduce a semisupervised machine learning model aimed at identifying pharmacies involved in excessive procurement of prescription opioids. The model incorporates spatio-temporal and contextual factors, including geographic heterogeneity, supply chain information, and order transactions, to classify whether a retail pharmacy is engaged in opioid overprocurement. To train and evaluate the model, we utilize the Automation of Reports and Consolidated Orders System, a data set maintained by the Drug Enforcement Administration (DEA) that tracks the movement of controlled substances from manufacturing to retail sale. Ground-truth labels indicating improper opioid purchasing behavior are obtained from various sources, including legal proceedings and investigative reports. By comparing our model with three established DEA rules for detecting suspicious orders, we reveal the limitations of these rules, characterized by high rates of false positives and false negatives. In contrast, our semisupervised model achieves accurate identification of suspicious opioid buyers engaged in improper purchasing behavior, with an area under the curve of 0.87. Managerial implications: Our analysis reveals that the buyers identified by our model tend to cluster geographically and have common upstream distributors/manufacturers. This finding underscores the importance of local supply chain networks in facilitating the illicit distribution of prescription opioids. In terms of implementation, models similar to the one proposed in this paper can leverage the DEA’s existing infrastructure to identify and alert unusual opioid distribution patterns.