Machine learning (ML) greatly facilitates processes of health evaluation, mainly prognosis and premature diagnosis of long-term diseases. This is a review on the k-nearest neighbors (KNN) algorithm and its optimized versions with a comparison of how effective they have been in making disease predictions such as chronic kidney disease (CKD). While there are a variety of ML algorithms, such as decision trees, support vector machines, and neural networks, that hold potential, the focus here in this paper is the performance of the various KNN variants on prediction accuracy, hardness, and ease of computability. We use publicly available datasets on the Kaggle and UCI Machine Learning Repository platforms to examine the application of KNN on data of various types. Along with that, we also discuss model explainability and heterogeneity of data concerns and look at integration with real-time health monitoring based on IoT. We also refer to privacy-preserving methods like federated learning to enable secure data management. We identify KNN’s competitive advantage in medical diagnosis, especially for structured data, and determine where hybrid or ensemble models can offer value add, in our study. This is an important critique of KNN-based health assessment models and a basis for future possible clinical applications and research.