Semantic and Perceptual Models for Clinical Decision Support Systems

Abstract

The current vision of healthcare is evolving in directions towards the secondary use of health data for producing new evidence, rapidly assimilating new knowledge, including the patient as an active component in decision-making and developing communication strategies to coordinate different areas of health care, among others. The work in these directions heavily relies on the appropriate use of different technologies. Among these technologies, Clinical Decision Support Systems (CDSS) implement validated evidence as computable artifacts that enable access to medical knowledge at the point in time when it is needed to make a decision about a person’s health. During the last two decades CDSS standards and technologies have progressed significantly to develop them as more robust and scalable systems. However, the current context of medicine sets high demands in aspects such as interoperability to enable the use of EHR data in CDSS, the need to establish communication challenges to include the patient as an active component in decision making, collaborative learning and sharing CDSS across institutional borders, to name a few. In this thesis I tackle some of these challenges. In particular, I evolve previous conceptual computerized decision support frameworks and I postulate a CDSS environment where different models interact to enable:

• Secondary use of data for CDSS: The dissertation presents a model to leverage different developments in data access and standardization of medical information. The result is an openEHR-based Data Warehouse architecture that enables access, standardization and abstraction of clinical data for CDSS. The architecture allows: a) to access heterogeneous data sources; b) to standardize data into openEHR to grant interoperability of data; and c) to exploit an openEHR repository as a Data Warehouse that allows querying data in a technology-independent format (the Archetype Query Language).

• CDSS semantic specification: The semantic model proposed exploits the paradigm of Linked Services to unambiguously describe CDSS in a machine- understandable fashion. This grants ontological descriptions of functional, non- functional and data semantics. These descriptions facilitate to overcome some of the barriers in CDS functionality sharing. In particular, the semantic model proposed allows using expressive queries to discover CDS services in health networks, and analyzing CDSS interfaces to understand how to interoperate with them.

• Effective patient-CDSS interaction: the dissertation proposes a method to evaluate the communication process between patients and consumer-oriented CDSS. The method aims for detecting if important human-computer interaction barriers that could lead to negative outcomes are present in CDSS user interfaces.

The methods and developments presented are framed in the context of the CDSS er du syk. Er du syk (in English, are you ill) is a symptom checker that allows users to record data regarding their symptoms and demography. These data are combined with epidemiology information from regional Laboratory Information Systems to provide patients a list with the likelihoods of the diseases that may be affecting them.