Opioid Overdose Death Prediction with Graph Neural Networks
Xianhui Chen1,2, Zishan Gu1, John Myers3
1Computer Science and Engineering, The Ohio State University, Columbus, Ohio, USA.
Medrxiv : the Preprint Server for Health Sciences
|March 27, 2026
Summary
Predicting opioid overdose deaths in Ohio is crucial for intervention. A novel Spatial-Temporal Graph Neural Network (ST-GNN) framework improves prediction accuracy, especially for larger counties, aiding public health efforts.
Area of Science:
- Public Health
- Data Science
- Epidemiology
Background:
- The opioid crisis significantly impacts Ohio, with overdose death rates exceeding national averages.
- Rural and Appalachian regions are disproportionately affected by opioid overdose deaths.
- Accurate county-level prediction of opioid overdose deaths is essential for timely public health interventions.
Purpose of the Study:
- To develop and evaluate a Spatial-Temporal Graph Neural Network (ST-GNN) framework for predicting county-level opioid overdose deaths in Ohio.
- To improve the accuracy and reliability of opioid overdose death predictions, addressing challenges posed by variations between large and small counties.
- To integrate spatial and temporal dynamics with socio-economic factors for enhanced public health decision-making.
Main Methods:
- Utilized a Spatial-Temporal Graph Neural Network (ST-GNN) framework combining Graph Neural Networks (GNNs) for spatial relationships and Long Short-Term Memory (LSTM) networks for temporal dynamics.
- Incorporated quarterly opioid overdose death data from Q1 2017 to Q2 2023 for 88 Ohio counties.
- Integrated a nine-dimensional dynamic feature set (e.g., naloxone administration, high-risk prescribing) and a static Social Determinants of Health (SDoH) index.
Main Results:
- The proposed ST-GNN framework demonstrated superior predictive performance compared to traditional statistical models and temporal deep learning baselines.
- The model showed enhanced accuracy particularly in predicting overdose deaths for larger counties.
- A supplementary classification-based strategy significantly improved prediction stability and reliability for smaller counties.
Conclusions:
- Spatial-temporal modeling is critical for accurately predicting opioid overdose deaths at the county level.
- Customized training strategies, differentiating between county sizes, are necessary for robust public health interventions.
- The findings support the use of advanced machine learning techniques to inform and enhance strategies addressing the ongoing opioid crisis.
Related Concept Videos
01:20Opioid Analgesics: Morphine and Other Natural Cogeners
1.5K
Opioids are a class of drugs that mimic endogenous opioid peptides and act on opioid receptors, and help in pain relief. These compounds are classified as natural, synthetic, or semi-synthetic. Natural opioids, like morphine, codeine, and thebaine, are derived from the opium poppy plant (Papaver somniferum or Papaver album) and are termed opiates. Synthetic opioids are artificial, while semi-synthetic opioids combine natural and synthetic compounds. Morphine, a prototypical opioid, possesses a...
1.5K
01:15Opioid Analgesics: Synthetic and Semisynthetic Opioids
1.4K
Synthetic and semisynthetic opioids are pivotal in pain management and tackling opioid addiction. Semisynthetic opioids, including morphinans (morphine derivatives), oxycodone, oxymorphone, hydrocodone, and hydromorphone, have improved pharmacokinetic profiles compared to morphine. Additionally, heroin and 6-MAM (6-Monoacetylmorphine) show better CNS penetration than morphine due to heightened lipid solubility. Hydromorphone, a potent opioid, undergoes hepatic metabolism to form the active...
1.4K
01:22Opioid Receptors: Overview
6.4K
Opioid receptors, including the mu (μ, MOR), delta (δ, DOR), and kappa (κ, KOR) types, belong to the rhodopsin family of G protein-coupled receptors. These receptors are located throughout the central and peripheral nervous systems and in non-neuronal tissues such as macrophages and astrocytes. Opioid receptor ligands can be categorized into agonists or antagonists. Highly selective agonists include [d-Ala2, MePhe4, Gly(ol)5]-enkephalin or DAMGO for MOR, [D-Pen2,...
6.4K
01:25Analgesia and Pain Management
2.9K
Pain is critical to various clinical pathologies, provoking an urgent need for effective management. Pain, whether acute or chronic, is a complex neurochemical process. Its alleviation depends on the type, with nonopioid analgesics effective for mild to moderate pain, such as musculoskeletal or inflammatory pain, while neuropathic pain responds best to anticonvulsants, tricyclic antidepressants, or serotonin/norepinephrine reuptake inhibitors. For severe acute or chronic pain, opioids may be...
2.9K
01:07End Point Prediction: Gran Plot
1.4K
A Gran plot is used to predict the equivalence volume or endpoint of a potentiometric or acid-base titration without reaching the endpoint. Typically, titration data is collected as a function of the titrant's volume up to a point less than the equivalence volume and then transformed into a linear format. The straight line is extended to the x-axis, indicating the necessary titrant volume to achieve the equivalence point.
For potentiometric titration, the Gran plot is created by plotting...
For potentiometric titration, the Gran plot is created by plotting...
1.4K
01:24Drug Toxicity: Risk factors
167
Adverse Drug Reactions (ADRs) are potential complications that arise during pharmacotherapy, influenced by multiple risk factors. Age plays a significant role; both neonates and the elderly are at heightened risk due to their respective immature and diminished metabolic and elimination processes. Gender also impacts ADRs, with females experiencing a 1.5 to 1.7-fold greater risk than males, which may be linked to pharmacokinetic, pharmacodynamic, and hormonal differences. Notably, neonates, the...
167