Sonar-AI discovers first comprehensively described cases of GLP-1 receptor agonist associated thrombocytopenia: Faster, cheaper, and better than the 20th century approach of RADAR/sonar

Abstract

Abstract Background/Introduction. The 21st century Southern Network on Adverse Reactions-Artificial Intelligence (SONAR-AI) is a team of multidisciplinary experts specializing in cutting-edge innovations aiming to identify first cases of novel hematology-associated adverse drug reactions (ADRs). SONAR-AI is the only National Institutes of Health (NIH)-funded pharmacovigilance network. The 20th century RADAR/SONAR program reported >50 serious adverse drugs reactions (ADRs) between 1998- and 2020 (pre-pandemic), including the first 10 comprehensively described cases of clopidogrel-associated thrombotic thrombocytopenic purpura (TTP) (NEJM 2000), and were the first to report this comprehensive case information to the Food and Drug Administration (FDA). (Wood AJJ, NEJM 2000). An international platelet expert and 3 hematologists identified, to our knowledge, the first four patients with new onset severe thrombocytopenia after initiating GLP-1 receptor agonists. These cases sparked a SONAR-AI investigation of GLP-1 receptor agonist-associated thrombocytopenia. Methods. SONAR-AI applies novel methods to identify the first cases of hematologic ADRs. Several methods facilitated a comprehensive review of multiple sources of GLP-1 receptor agonist-associated thrombocytopenia: detailed case histories and patient studies; FDA Adverse Event Reporting System (FAERS) review; social media reviews of various platforms; and publication searches. We evaluated their effectiveness serving as an early warning system to identify the first reports of a novel hematologic ADR. Results. Four patients at two large medical centers all presented with (immune) thrombocytopenia following GLP-1 receptor agonist initiation. Platelet counts were 106,000, 45,000 31,000, and 2,000 platelets per microliter. Communication with treating clinicians indicated that SONAR-AI facilitated the first FAERS reporting of these cases. A search of FAERS captured 54 other users of semaglutide, liraglutide, and tirzepatide with (immune) thrombocytopenia. These reports were incomplete with respect to data on age, gender, race/ethnicity, comorbidity, concomitant illnesses, and long-term outcome. The number of these incompletely reported FAERS cases has increased annually since 2018. Social media reviews included incompletely described reports from Reddit and Twitter with reports of low platelet counts, excessive bruising, gastrointestinal and/or abnormal gynecological bleeding. Data synthesis was attempted, prompting ChatGPT with keywords and phrases, which created summary reports for Reddit, but NOT Twitter, TikTok, Instagram, or Facebook. (ChatGPT has direct access to Reddit but not other social media platforms.) The platform eHealthMe, a program that asserts it relies on AI algorithms and models as an early detection system to identify ADRs, found 77 very incompletely described reports of semaglutide-associated thrombocytopenia, with information including duration of use, gender, age, other drugs taken, and comorbid illnesses, but not outcome. TriNetX Global Collaborative Database Review showed that among both patients with diabetes and patients with obesity but without diabetes, semaglutide use was associated with a lower risk of thrombocytopenia compared with older diabetes and weight loss medications. Pre-print sites and a PubMed literature review each did not identify any reports at all of GLP-1 receptor agonist-associated thrombocytopenia. Conclusion: The SONAR-AI approach facilitated the first comprehensive description of four cases of GLP-1 receptor agonist-associated thrombocytopenia. This effort was completed in four weeks and cost $500. The four cases have been the first comprehensively described cases submitted to FAERS. This effort mirrors the prior RADAR/SONAR effort that comprehensively described the first 10 cases of clopidogrel-associated TTP, though that NIH-funded effort cost $250,000 and took 1 year to complete. SONAR-AI is a marked upgrade to our prior RADAR/SONAR- 95% faster, 96% cheaper, and 90% more comprehensive.