AI Methods for Large Scale Epidemiological Studies using Patient Reports of Medication Adherence and Tolerability

About This Grant

Project Summary Adherence to prescribed medications is a critical aspect of effective medical treatment, especially for chronic con- ditions; however, the World Health Organization (WHO) estimates that more than 50% of patients with chronic conditions in the United States do not take their medications as prescribed. Medication non-adherence is associ- ated with worsening health conditions and increased comorbidities, and is estimated to annually account for 25% of hospitalizations, more than 100,000 preventable deaths, and up to $500 billion in healthcare costs in the United States. The WHO affirms that “increasing the effectiveness of adherence interventions may have a far greater impact on the health of the population than any improvement in specific medical treatments.” The challenge of increasing the effectiveness of adherence interventions has likely been due in part to the fact that the majority of medication non-adherence is intentional (in contrast to unintentional, such as forgetfulness) and sources of data for understanding the factors that influence intentional non-adherence remain limited. As encouraged by the United States Food and Drug Administration and Centers for Disease Control and Prevention, our prior work—funded for the past 10 years by the National Library of Medicine (R01LM011176)—has demonstrated that patient reports in real-world, non-traditional sources of data can be used as a novel, complementary approach to post-marketing pharmacovigilance. Because online patient reports are not biased by survey questions or interviewers, are avail- able on a large scale, and may include participants who are under-represented in traditional sources of data, in a preliminary qualitative content analysis, we were able to gain novel insights about medication non-adherence that were not well-represented in other studies, such as patients reporting dechallenge (i.e., an adverse effect stopping when the medication was stopped) and rechallenge (i.e., the adverse effect resuming when the medication was started again). Validating our approach through five disease-specific case studies in collaboration with domain experts in cardiovascular diseases, gastrointestinal diseases, cancer, HIV, and diabetes (Aim 3), we propose to develop novel natural language processing and artificial intelligence methods (an intelligent agent with a special- ized large language model at its core) to capture medication use narratives from online patient reports (Aim 1) and to elucidate additional factors contributing to medication non-adherence from spontaneous reporting systems (SRS, e.g., FAERS), such as drug indications, the magnitude of adverse events, and drug-drug interactions (Aim 2). We incorporate findings from other sources into our case studies through systematic reviews (Aim 3), synthe- sizing and comparing published studies to what we learn from patient reports posted online and in SRS. This is the most comprehensive study of its kind ever attempted, bringing the voice of the patients directly to researchers in a reproducible, cost-effective manner. This can inform adherence interventions and reduce the morbidity, mortality, and financial burden associated with intentional non-adherence, aligning with the NLM’s goal of creating a future in which data and information transform and accelerate biomedical discovery and improve health and healthcare.