Effective architecture for clinical decision support system

Abstract

The Healthcare domain is a very sensitive domain where the direct stakeholders are patients in the public, which makes a healthcare system directly dealing with patients’ lives. There could be heard of several cases in a year where it leads to critical damage or even loss of life, because of misdiagnosis or delays in the diagnosis process or the delay or ignorance of new treatment methods. A clinical decision support system facilitating support for diagnosis and therapeutic decisions could greatly help healthcare professionals when identifying diseases by going through patients’ biometrics, life cycle, and symptoms, and when deciding on necessary clinical tests to arrive at confirmation of the diagnosis and decide on treatment methods. The main focus of this study is mapping the real-world diagnosis process to a digitalized system. Senior clinicians use their own experience to derive medical diagnoses accurately. This study proposes an architecture for an evidence-based clinical decision support system, where the system infers knowledge from past knowledge, using machine learning algorithms, and use for future predictions, which could infer and use the medical incidents of the past for future diagnosis, just like experienced doctors. In a practical scenario, diagnosis of disease happens step by step, going through several stages starting from an initial level and digging deeper. To incorporate this behavior, a layered knowledge modeling system is proposed with an ensemble classifier of Random Forest classifier, Support Vector Machine, and Naïve Bayes classifier, and organized into a tree structure based on disease classification hierarchy. Additionally, the proposed system provides feedback and suggestions for clinical tests using feature selection, and a rationale for the diagnosis derived by incorporating explainable machine learning concepts.