Optimizing drug selection through sentiment-based machine learning models presents an innovative approach that incorporates patient feedback to improve therapeutic decision-making. This research investigates the use of Long Short-Term Memory (LSTM) networks to analyze patient reviews and clinician input, with a focus on extracting sentiments related to drug effectiveness, side effects, and overall satisfaction. The LSTM model is trained using labeled sentiment data from patients, allowing it to classify reviews as positive, negative, or neutral, while also identifying aspect-based sentiments, such as relief and adverse reactions. This classification can guide healthcare providers in prioritizing drug options that align closely with patient preferences and experiences, contributing to personalized care. The model development involves data preprocessing (text cleaning, tokenization, and embedding), feature extraction through word embeddings, and fine-tuning with dropout layers to enhance generalization. Sentiment scores are then integrated into an optimization framework, combining clinical efficacy and patient sentiment. Evaluations using metrics like accuracy, F1-score, and ROC-AUC demonstrate the model's potential to accurately capture sentiment and improve drug recommendations. This approach underscores the importance of sentiment analysis in healthcare, allowing for a data-driven method to optimize drug choices and potentially improve patient adherence and satisfaction.