PREDICTION OF ADVERSE DRUG REACTIONS AND DRUG-DRUG INTERACTIONS FOR ENHANCED PATIENT SAFETY

Adverse Drug Reactions (ADRs) and Drug-Drug Interactions (DDIs) represent two of the most critical threats to
patient safety in modern healthcare, frequently leading to hospitalizations, treatment failures, and, in severe
cases, fatalities. The increasing prevalence of polypharmacy, where patients are simultaneously prescribed
multiple medications for chronic conditions such as diabetes, hypertension, and cardiovascular disease, has
intensified the risk of undetected drug interactions. This study presents the design, development, and evaluation
of an intelligent machine learning-based system capable of predicting ADRs and DDIs in real time from clinical
prescription inputs. The proposed system integrates a multi-source pharmacovigilance database combining
SIDER, OFFSIDES, TWOSIDES, and DrugBank, covering over 1,430 drugs and their documented adverse
effects. A stacked ensemble model comprising Random Forest, XGBoost, and LightGBM is trained on this
consolidated dataset and augmented by a Natural Language Processing (NLP) pipeline using SciSpacy
(BC5CDR model) for automated drug name extraction from free-text prescriptions and clinical notes. A
pharmacovigilance noise-filtering mechanism employing Proportional Reporting Ratio (PRR) thresholds (≥ 2.0
for ADR, ≥ 3.0 for DDI) and an 11-category noise blacklist eliminates confounding signals such as self-harm
events, procedural artifacts, and socioeconomic confounders, ensuring clinically meaningful predictions. The
system achieved an accuracy of 98.94%, precision of 99.7%, recall of 55.46%, and an F1-score of 71.27%. A
Streamlit-based web interface provides an accessible, real-time prediction environment for clinicians and
pharmacists. The system was validated against 20 real-world clinical test cases and demonstrated high
agreement with established pharmacological safety literature. This tool has significant potential to enhance
clinical decision-making, reduce adverse drug events, and facilitate the transition towards safer, personalized
medicine