This study proposes a patient-centric drug recommendation model leveraging Electronic Health Records (EHRs) to aid in patient-centered decision-making, particularly for those with multiple comorbidities and in remote medical emergencies. The proposed model capitalizes on the rich tapestry of information contained within Electronic Health Records (EHR) data such as medical history and past medications. The study involves construction of a labeled dataset for drug recommendation from the unstructured MIMIC-III database. Using this labeled dataset a patient information graph is created, which is then processed by a Graph Neural Network (GNN) model to recommend drugs. Focusing on patients with neurological and nervous system disorders, the model demonstrates a micro-averaged recall of 60%, precision of 50%, and F1-score of 54%. The study underscores the significance of incorporating diverse information present in EHRs such as patient diagnoses, phenotypes, demographics, and medical history in drug recommendation systems.