Estimating Weekly National Opioid Overdose Deaths in Near Real Time Using Multiple Proxy Data Sources

Area of Science:

• Public Health
• Data Science
• Epidemiology

Background:

• Opioid overdose is a critical public health issue in the US.
• National data on overdose deaths have significant reporting delays.

Purpose of the Study:

• To develop and validate a statistical model for near real-time estimation of national opioid overdose deaths.
• To provide timely insights into opioid overdose mortality trends.

Main Methods:

• A LASSO regression model was employed, integrating 5 proxy data sources (health, law enforcement, online data) from 2014-2019.
• Weekly predictions for 2018-2019 were generated and validated against actual National Vital Statistics System data.
• Model performance was compared to a baseline seasonal autoregressive integrated moving average (SARIMA) model.

Main Results:

• The machine learning model achieved a low error rate (1.01% in 2018, -1.05% in 2019) in estimating national opioid overdose deaths.
• Weekly prediction accuracy showed a substantially lower root mean squared error (60.3 in 2018, 67.2 in 2019) compared to the SARIMA model (310.2 in 2018, 83.3 in 2019).

Conclusions:

• Proxy administrative data sources can effectively estimate national opioid overdose mortality.
• This approach offers a more timely understanding of the opioid overdose public health problem.
• The developed model shows superior accuracy for real-time surveillance of opioid overdose deaths.