An Ontology-Driven Approach to Clinical Evidence Modelling Implementing Clinical Prediction Rules

Introduction

Diagnostic error has been shown to be a major threat to patient safety, particularly within the context of the primary care setting (Elder & Dovey, 2002). Where diagnostic errors occur there is the potential for an adverse event to take place that may undermine the safety of the patient, sometimes with disastrous results (Fisseni, Pentzek, & Abholz, 2008). Proponents of evidence-based medicine have analysed the diagnostic process itself and described the flaws they see as existing in traditional diagnostic approaches that can contribute to the possibility of adverse events taking place (Kostopoulou, Delaney, & Munro, 2008). Steps have been proposed that describe the diagnostic process and are seen as requirements for implementing a more rigorous diagnostic process (Sackett, Rosenberg, Gray, Haynes, & Richardson, 1996a). The formal steps describe a methodology which advocates obtaining, evaluating and utilising the best available clinical evidence that has been previously demonstrated to be effective and pertains to each clinical case as it arises (Knottnerus & Buntinx, 2002).

If we are to practice evidence-based medicine effectively there is a need to efficiently and easily access the latest validated clinical guidelines that have been shown to be applicable to the diagnostic context and circumstances of the particular case being considered. The use of information technology and the internet for this purpose has greatly assisted in this task and its use has been formalised as part of the steps of the evidence-based medicine philosophy (Greenhalgh, 2010). Whilst this approach has been useful it can still be manually intensive and not without its problems.

There are a number of widely proposed models of clinical knowledge but acceptance of these models has not been without question and more generic ontology driven approaches advocated as possible improvements (Smitha & Ceustersc, 2006). Dissemination of accepted clinical guidelines is often done in the form of peer-reviewed hard-copy texts, soft-copy textual documents or textual web pages. It is difficult for consistent interpretation and representation of those guidelines that allow for their subsequent incorporation into integrated clinical information systems providing more sophisticated functionality such as clinical decision support. This problem can therefore by rephrased as follows: how can we more formally define and represent what constitutes recognised ‘clinical knowledge’ with a view to disseminating that knowledge more effectively for use by both clinical practitioners and information systems designed to support their use? The use of clinical prediction rules (CPRs) has long been accepted as a valuable means of deriving and disseminating clinical guidelines in respected evidence based journals such as the JAMA series (Ebell, Smith, Barry, Ives, & Carey, 2000). Because of their well defined structure they can be more easily incorporated and utilised in clinical decision support systems. CPRs can also be derived using well understood statistical methods such as logistic regression that could be applied to large sources of epidemiological data that are currently available in the form of both public and privately owned electronic research repositories.

Another problem is the time lag and delay that exists between clinical research outputs and their translation into frontline primary care practice (Lenfant, 2003). The ability of primary care practitioners to effectively practice evidence-based medicine is undermined by the delays that exist in disseminating accepted clinical best practice that has been derived from epidemiological based clinical research. This second question can be stated as: how do we expose our formal models of clinical knowledge with a view to easily integrating them with other systems that are utilised to bridge the gaps that exist between the research and primary care processes.