Cross-state validation of a novel prescription-based model to predict new long-term opioid use

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Opioid use disorder and overdose remain significant public health concerns in the United States. The CDC has identified reducing the number of previously opioid-naïve patients who transition to long-term (>90 days) opioid use, an outcome associated with both overdose and opioid use disorder, as a key strategy for preventing these opioid-related harms. Clinical tools to predict patient risk of transition to long-term use can be an important component of this strategy. Prescription drug monitoring programs (PDMPs) are ideal implementation platforms because they contain: a) statewide, population-level controlled substance prescription data, b) web interfaces that clinicians in most states are required to check before prescribing opioids, and c) basic support for clinical decision-making (e.g., identifying patients receiving prescriptions from multiple prescribers). Key barriers to equipping PDMPs with these tools are that state populations differ and PDMPs operate independently; building 50 different prediction models from scratch is inefficient and impractical. States need accurate and generalizable PDMP-based prediction models that can be turned into clinical tools to help clinicians avoid inappropriate transitions to long-term opioid use and, by extension, prevent opioid-related harms. The goal of this project is to produce a scientifically transparent, generalizable, and clinically useful prediction model that PDMP administrators can use as a foundation for future tool development. Using 2016-2018 California PDMP data, we have developed and validated a novel risk prediction model that predicts an opioid-naïve patients’ likelihood of transitioning to long-term opioid use with high discrimination (concordance statistic = 0.91). Our proposed project has two objectives. First, to assess how a California-based risk prediction model generalizes to Kentucky, a state with substantially different demographics and higher rates of opioid prescribing and overdose than California, by updating our existing 2018 model to predict incident long-term use in Kentucky. Second, to compare accuracy of an updated form of the existing California model versus new state-specific models developed using Kentucky PDMP data. This project will provide PDMP administrators information about the trade-offs between using the product of this proposal as a “foundational model” versus investing resources to develop their own state-specific models. Study findings will set the stage for future implementation of prediction tools into the PDMPs of Kentucky and California, and also provide critical information to PDMP agencies from other states interested in implementing their own prediction tools. The proposed study is a necessary step in a research program to build and implement PDMP-based tools that promote safe opioid prescribing and reduce the incidence of opioid overdose, opioid use disorder, and other opioid-related harms in the United States.