Designing and Evaluating an Ensemble Reasoning-based Clinical Decision Support System

A clinical decision support system (CDSS) can help physicians make clinical decisions, such as differential diagnosis, therapy planning, or plan critiquing. To make such informed decisions, a physician may need to keep track of a large amount of medical data and literature, such as new research articles, pharmacological therapies, and updates in Clinical Practice Guidelines. Therefore, a CDSS can be designed to assist physicians by providing relevant evidence-based clinical recommendations, subsequently reducing the cognitive overhead required to stay up-to-date with an evolving body of literature. We designed a CDSS by leveraging Semantic Web technologies to create an AI system that reasons in a way similar to physicians. We base our abstraction of human reasoning on the Select and Test Model (ST-Model), which combines multiple forms of reasoning, such as abstraction, deduction, abduction, and induction, to arrive at and test hypotheses. Based on this framework, we perform ensemble reasoning, the integration and interaction of multiple types of reasoning. We apply our CDSS to the treatment of type 2 diabetes mellitus by designing a domain ontology, the Diabetes Pharmacology Ontology (DPO), that supports both deductive and abductive reasoning. DPO is also used to provide a schema for our knowledge representation of hypothetical patients, where each patient is encoded in RDF as a Personalized Health Knowledge Graph (PHKG). We use the Whyis knowledge graph framework to implement our CDSS. This is achieved by writing software agents to perform custom deductive reasoning and integrating abduction using an existing reasoning engine, the AAA Abduction Solver. We apply our approach to perform therapy planning on hypothetical patients.