Panel Paper: Using Data Science to Identify Individuals at High Risk of Opioid Overdose: A Multiyear Data Linkage Project in Maryland

Friday, November 4, 2016 : 11:15 AM
Embassy (Washington Hilton)

*Names in bold indicate Presenter

Brendan Saloner, Johns Hopkins University

Drug overdose is now the most common cause of injury death in the United States, and opioid analgesics are involved in almost half of these deaths. Overdose deaths have quadrupled since the late 1990s, corresponding to a surge of prescriptions for opioid analgesics. In recent years, there has also been a spike in heroin overdoses. To curb overdoses, states have implemented prescription drug monitoring programs (PMDPs). PDMPs collect all prescriptions written for controlled substances within a state. Analysis of PDMP data can identify patients exhibiting patterns of harmful prescription opioid use such as rapid escalations of dosage or frequent visits to multiple prescribers (“doctor shopping”), which can be valuable for systematic intervention within patient populations. However, under the status quo PDMPs are used in an ad-hoc manner and often requires burdensome queries of the database, limiting opportunities for timely intervention.

Under a grant from the US Department of Justice, our team has partnered with the Maryland Department of Health and Mental Hygiene to develop a predictive risk model (PRM) that will provide an analytical basis to identify individuals at high risk of an opioid overdose. The ultimate goal is dissemination of risk scores into clinical practice to enable better referral to treatment and supportive resources. In the first stage of the project, we are developing and validating a method for case identification with PDMP records linked to fatal overdoses and piloting measures to detect high-risk prescription opioid use. The model will initially be trained on data from 2013-2014.

Our team is also pursuing person-level linkages of PDMP records to clinical and social databases. We hypothesize that databases beyond the PDMP will reveal important clinical or social risk factors (such as hospitalizations and arrests) beyond what can be learned from the PDMP alone, ultimately improve case identification and timely intervention of high-risk individuals. Likely databases for linkage to the PDMP include all-payer hospitalization records, service claims from insurance programs (especially Medicaid and commercial insurance), records from addiction treatment programs, and emergency responder data from ambulance services. Social data may include arrest and community supervision data and social services receipt. Data use agreements are currently being developed and implemented with several state agencies. Linkages across databases will be performed by a regional data information exchange that uses validated probabilistic matching.

Analyses conducted in spring-summer 2016 focus on developing the PRM first with the PDMP data alone (following prior studies on PDMPs) and, second, with PDMP data linked to clinical and social databases (e.g., hospitalization and arrest records). The presentation at APPAM will provide a detailed profile of the PDMP population and describe the multiple and overlapping risk factors associated with fatal overdoses and other adverse outcomes. For example, I will provide a detailed examination of the relative contribution of different clinical risk factors that occur in the year before an overdose. Because of the novelty of the linkage process, I will also discuss some of the challenges and opportunities of large-scale linkage processes, including lessons for practitioners and researchers.