Bayesian ACCESS for Understanding Latent Epidemic Trajectories from Publicly Released Suppressed Data: Application to U.S. Opioid-related Overdose Mortality

2026-08-10

AI Summary

Publicly available health statistics often omit critical information due to cell suppression for small counts (e.g., in CDC WONDER), hindering accurate inference of rare events and subgroup-specific epidemiological dynamics. This work proposes the Bayesian ACCESS framework, which explicitly models the suppression mechanism within a unified hierarchical structure. By integrating a suppression-aware observation model, autoregressive change-point detection, and Bayesian nonparametric clustering, ACCESS jointly infers multiple temporal change points, reconstructs latent epidemic trajectories, and intelligently shares information across geographic regions and demographic groups—balancing accuracy with heterogeneity. Applied to U.S. state-level opioid overdose mortality data from 1999 to 2024, the method successfully identifies subpopulation-specific epidemic shifts and dynamic patterns undetectable by conventional approaches.

Abstract

Publicly released health statistics play a central role in characterizing temporal trends and identifying structural changes in population health. However, disclosure limitation through suppression of small cell counts, as implemented in systems such as the Centers for Disease Control and Prevention Wide-ranging ONline Data for Epidemiologic Research (CDC WONDER), produces partially observed count data that complicate statistical inference. These challenges are particularly acute for rare outcomes and subgroup analyses, where suppression is widespread and varies across geographic regions, demographic populations, and time. We propose Bayesian ACCESS (Autoregressive Change-point and Clustering Estimation for Suppressed Count Series), a Bayesian hierarchical framework for inference on latent epidemic trajectories and their structural changes from disclosure-limited health statistics. The proposed model directly represents suppressed count data through a suppression-aware observation model, jointly infers multiple temporal change points and latent trajectories, and borrows information across related geographic and demographic populations through Bayesian nonparametric clustering while preserving meaningful heterogeneity. We apply Bayesian ACCESS to opioid-related overdose mortality data from CDC WONDER for U.S. states from 1999 to 2024. The analysis identifies distinct subgroup-specific epidemic trajectories and structural changes that would be difficult to characterize using publicly released health statistics without explicitly accounting for data suppression.

Problem

• data suppression
• epidemic trajectories
• disclosure limitation
• overdose mortality
• latent inference

Innovation

• Bayesian hierarchical model
• data suppression
• change-point detection
• Bayesian nonparametric clustering
• latent epidemic trajectory