Using natural language processing to identify problem usage of prescription opioids

•Natural language processing and computer assisted manual review can be combined to efficiently and accurately identify conditions such as problem opioid use that are often only documented in unstructured clinical text.•As much as one-third of problem opioid use documented in electronic health records may only be documented in unstructured clinical notes.•Clinicians use hundreds of different ways to describe abuse, addiction, misuse or overuse of prescription opioids.•Discovery of novel and unanticipated ways clinicians describe common health conditions in electronic clinical notes from millions of encounters is possible using full-text indexing and query functions built in to popular relational database systems.