Predicting overdose among individuals prescribed opioids using routinely collected healthcare utilization data

Area of Science:

• Public Health
• Health Informatics
• Data Science

Background:

• Rising opioid overdose rates in the US necessitate improved surveillance tools.
• Early identification of high-risk individuals is crucial for effective intervention.

Purpose of the Study:

• To develop and validate a predictive algorithm for opioid overdose using routinely collected healthcare data.
• To assess the algorithm's performance in identifying patients at risk of overdose.

Main Methods:

• A commercial claims database (2011-2015) was used, including over 5 million patients prescribed opioids.
• An elastic net logistic regression model was employed for prediction, with patients randomly assigned to training, validation, and test sets.
• Model performance was evaluated using metrics like c-statistic, sensitivity, and specificity, and compared to simpler models.

Main Results:

• The elastic net model demonstrated strong predictive performance (c-statistic 0.887), outperforming simpler models.
• Key predictors for overdose included younger age (18-25), prior suicide attempt, and opioid dependence.
• The model identified 2,682 overdoses among 5,293,880 individuals, with a sensitivity of 80.2% and specificity of 80.1%.

Conclusions:

• Sophisticated algorithms utilizing healthcare databases can effectively predict overdose risk.
• This predictive capability enables opportunities for active patient monitoring and timely intervention.
• The developed algorithm offers a valuable tool for public health initiatives aimed at mitigating the opioid crisis.