Numerous studies have reported that adverse drug events (ADEs) occur in 5 to 35 percent of patients across all age groups in outpatient settings. With developments in artificial intelligence (AI) and Natural Language Processing (NLP), the authors have employed NLP techniques to design a system that aims to mitigate adverse drug reaction (ADR) levels and facilitate their early identification. This paper presents an analysis and comparative study of traditional and modern techniques for sentiment analysis, focusing on their applicability within the healthcare landscape, using the UCI ML-Drug Review Dataset. Traditional methods, such as Long Short-Term Memory (LSTM) networks, Random Forest, and LightGBM, are presented against advanced language models (LMs) like LLaMA 2-7B, GPT-3.5 Turbo, and Gemma 7B. The experiments demonstrate that LMs, particularly GPT-3.5 Turbo in a 15-shot setting, outperform traditional methods with an impressive accuracy of 94.20%. Based on these findings, the authors developed MediSentBot, a medicine recommendation and review system utilising the GPT-3.5 Turbo model to recommend medicines and aggregate live reviews, ensuring up-to-date and accurate sentiment analysis. This study highlights the enhanced performance and applicability of modern NLP techniques in medical downstream tasks, specifically sentiment analysis.