Automatable algorithms to identify nonmedical opioid use using electronic data: a systematic review.

Area of Science:

• Health Informatics
• Pharmacovigilance
• Public Health Surveillance

Background:

• Identifying nonmedical opioid use is crucial for resource allocation.
• Automated algorithms using electronic health records (EHRs) offer a potential solution for large-scale surveillance.
• Systematic review of existing algorithms for detecting nonmedical opioid use.

Purpose of the Study:

• To review the utility, validation, and application of automated algorithms for detecting nonmedical opioid use.
• To assess the performance characteristics and settings of these algorithms.
• To understand the potential of automated methods in identifying at-risk individuals and prescribers.

Main Methods:

• Systematic literature search of PubMed and Embase for automatable algorithms.
• Assessment of algorithm development, validation methods, and performance metrics.
• Categorization of algorithms by target (patients, providers, or medications) and methodology (e.g., regression modeling, NLP).

Main Results:

• Fifteen algorithms were included: 10 for patients, 2 for providers, 2 for both, and 1 for high-abuse potential medications.
• Most patient-focused algorithms utilized prescription and medical claims, with substance abuse codes as the reference.
• Regression modeling was common (11 algorithms), with others using NLP, data mining, or factor analysis.

Conclusions:

• Automated algorithms can aid population-level surveillance for nonmedical opioid use.
• No definitive "gold standard" exists for validation; users must consider false positives and negatives.
• Limited real-world application exists, necessitating further implementation research to clarify utility.