Date of Award

12-12-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Information Science

First Advisor

Ningning Wu

Second Advisor

Wen Zou

Abstract

The opioid crisis in the United States has evolved from a medical response to pain into a nationwide public health emergency. The prescription opioids (POs) – associated cardiovascular adverse events (CVAEs) remain insufficiently studied, especially in relation to sex disparities. This dissertation investigates sex disparities in PO-associated cardiovascular adverse events (CVAEs) through an analytic framework integrating artificial-intelligence-assisted literature mining and real-world pharmacovigilance analytics. In the analytic framework, first, AI powered topic modeling was employed to uncover POs-associated CVAEs risks with a focus on sex differences. A total of 1,837 PubMed abstracts were analyzed using two methods, Latent Dirichlet Allocation (LDA) and the transformer-based BERTopic. Both AI-integrated LDA and BERTopic models generated meaningful topics from the case study on opioid-related cardiovascular risks in women, highlighting key areas of concern and clinical relevance. The study demonstrated the effectiveness of both AI-integrated LDA and BERTopic in the text mining of opioid-related cardiovascular risks in women. The findings highlighted the importance of AI integration with traditional natural language processing (NLP) techniques, which reveal potentially promising directions for future research advancements. Then, real-world adverse event data from the FDA Adverse Event Reporting System (FAERS) were analyzed. More than 18 million adverse event (AE) reports were retrieved and downloaded from FAERS reported over spanning from 2004 Q1 to 2024 Q3. After data pre-processing including drug name normalization and MedDRA PT filtration, 163,356 PO-CVAE pairs were identified for the 17 classes of FDA-approved POs, with 539 unique CVAES. EBGM was applied for the disproportionality analysis (DPA) and identified 79,085 pairs of potential risk signals cross the 17 opioid classes. A comparative analysis was then pursued to provide a global overview of opioid-related CVAEs for all 17 classes of FDA-approved POs. The top 10 CVAE safety signals for each opioid class were then presented. The top 5 CVAEs in detected signals are hypertension, hypotension, cardiac arrest, cardio-respiratory arrest, and myocardial infarction. The major contributors were clarified as well. Results from the network analysis and hierarchical clustering analysis revealed a close association among fentanyl-remifentanil-sufentanyl, and hydrocodone-tramadol-morphine. Additionally, much less commonly reported CVAEs were shared among butorphanol, levorphanol and meperidine. The CVAE-based results were compared with the previous published results which were based on the whole AE analysis. Finally, sex stratified disproportionality analyses were conducted on the same dataset. A total of 79,465 potential safety signals were detected covering 17 POs, including 30,835 associated with males, 40,844 SDRs with females, and 7,786 SDRs without sex information. The statistic comparative analysis revealed that more than half of the detected signals (56.98%) were from women associated mostly with tramadol, fentanyl, oxycodone, codeine, and morphine. Hypertension, hypotension and pericarditis were the top three disproportionate CVAEs, while hypotension, cardiac arrest, and hypertension were the leading CVAEs detected disproportionately in men. For each of the top 10 CVAEs, we observed notable sex differences in the contributing prescription opioids and their relative reporting proportions, suggesting that both the type of prescription opioid and the frequency of associated events differ between men and women. By integrating the three technologies, artificial intelligence, topic modeling, and pharmacovigilance analytics, together, this research establishes a scalable, data-driven framework to identify sex-specific POs - associated safety signals regarding cardiovascular complications. The findings highlight the necessity of incorporating sex as a biological variable in both clinical prescribing and post-marketing surveillance. By elucidating differential cardiovascular risk profiles across opioid ingredients, this work contributes to precision pharmacovigilance and provides actionable evidence for developing safer and more equitable pain-management practices.

Download

Share

COinS