SafeMed: A Cross-Platform AI-Integrated Healthcare Application for Adverse Drug Reaction Reporting

Imadeddin Ouahidi (Northwest Missouri State University), Anji Reddy Modugula (Northwest Missouri State University), Lohitha Vodnala (Northwest Missouri State University), Dr. Mark Chai (Northwest Missouri State University)

Adverse drug reactions (ADRs) are a major contributor to preventable morbidity, mortality, and healthcare expenditure worldwide. Beyond direct clinical harm, ADRs impose a substantial economic burden due to additional investigations, prolonged hospital stays, and downstream “prescribing cascades.” Despite the importance of post-marketing surveillance, spontaneous pharmacovigilance systems consistently suffer from severe underreporting. A systematic review of numerical estimates found a median underreporting rate of around 94% across studies, with similarly high underreporting observed even for serious/severe reactions [1]. We present SafeMed, a cross-platform healthcare application designed to reduce reporting friction and improve triage speed through multimodal intake and AI-assisted severity assessment. Patients can submit suspected ADRs via: structured text, voice recordings processed with automatic speech recognition, and photos documenting visible symptoms. This multimodal design aims to lower cognitive burden while capturing richer clinical context. SafeMed’s core pipeline applies LLM-assisted analysis, with clinician review in the loop, to classify severity across four triage levels (Mild, Moderate, Severe, Life-threatening) and to detect “red-flag” symptom patterns that warrant urgent attention. To support real-time responsiveness under variable load, the system uses an event-driven architecture with asynchronous processing, enabling rapid intake while offloading AI analysis to dedicated consumers. To reduce medication-name errors, a common failure mode in patient-generated reports, SafeMed integrates a medication normalization and fuzzy-matching service that combines approximate string matching and similarity metrics to handle typos, brand, and near-duplicate entries [2] [3]. The platform supports duplicate detection using set-similarity to improve report quality and reduce noise. The current SafeMed implementation is a functional prototype evaluated using simulated ADR reports and benchmark datasets to assess triage accuracy and data quality improvements. References [1] Hazell L, Shakir SA. Under-reporting of adverse drug reactions: a systematic review. Drug Safety 29 (2006) 385–396. [2] Tissot H, et al. Combining string and phonetic similarity matching to identify misspelt names of drugs. Journal of Biomedical Semantics 10 (2019). [3] Ding HM. Drug name correction of medication records: comparison of similarity metrics (e.g., Levenshtein and Jaro–Winkler). Technical Report / Thesis (2021).