Transforming Opioid Poisoning Surveillance Through Novel Technologies: Rationale and Methodological Protocol for...

Area of Science:

• Public Health
• Health Informatics
• Computational Linguistics

Background:

• Opioid-related harm necessitates timely surveillance for effective public health responses and policy evaluation.
• Emergency departments (ED) are crucial for identifying acute opioid poisonings, but current surveillance relies on structured data, missing vital free-text information.
• Underreporting and misclassification in existing systems hinder the identification of emerging trends and policy evaluation.

Purpose of the Study:

• To enhance the accuracy and consistency of opioid poisoning surveillance by applying natural language processing (NLP) to emergency department data.
• To leverage routinely collected ED data, including unstructured free-text fields, for improved identification of opioid poisoning presentations.
• To assess the utility of NLP-enhanced data in evaluating the impact of opioid policy reforms.

Main Methods:

• Development and training of NLP models using 15 years of Victorian Emergency Minimum Dataset (VEMD) records.
• Analysis of both structured and unstructured data fields to identify opioid poisoning cases.
• Validation of NLP models against a manually coded gold standard and incorporation of additional unstructured data (e.g., discharge summaries) for enhanced classification.

Main Results:

• The study pioneers large-scale application of NLP to Australian ED data for drug poisonings.
• Improved accuracy and consistency in identifying opioid poisoning presentations.
• Potential for strengthened routine surveillance without increased clinical workload.

Conclusions:

• NLP offers a robust method to improve the accuracy and consistency of opioid poisoning identification in emergency departments.
• This approach can significantly enhance routine surveillance systems, providing better data for timely policy and health system interventions.
• The study demonstrates the value of NLP in public health surveillance, particularly for drug-related harms.