Conceptual Design Principles for Data-Driven Clinical Decision Support Systems (CDSS): Developing Useful and Relevant CDSS

Introduction

Appropriate use of clinical data is the cornerstone for accurate and safe clinical decisions and to improve patient outcomes and overall organizational performance (Nicholson et al., 2017). This is especially important when health systems are challenged by extreme situations, such as the Covid-19 pandemic. Several types of health-related information, from various resources, are accessed and reviewed together by clinical decision-makers who then make informed clinical decisions (Greenes, 2014). When hospitals are filled to capacity and the care demands are high, the clinical decision-makers need to have available tools which facilitate improved access to information and decision-making.

For decades now, researchers have been seeking ways to provide information tools to clinicians and therefore assist them in making better decisions. Before the decade of 2000s, most of these systems were based on hardcoded knowledge associations. They were not data-driven, but instead knowledge-based expert systems and were incorporating medical field knowledge and complex clinical pathways. The accuracy of their recommendations heavily relied on how knowledge was pre-coded and whether all clinical judgment considerations were successfully incorporated. Any practice change would require the recommender system to be updated. Since health analytics is gaining momentum quickly, data scientists have been challenged with several opportunities to utilize large clinical and administrative datasets and develop data-driven recommender systems to support clinical decisions. Contrary to the older generation expert systems, these data-driven approaches are not hard coded, but instead, provide algorithm-based recommendations calculated directly from the data. Data science has gained significant momentum during the past decades, and it contributes to understanding clinical care, and public health concerns, like in the case of the recent Covid-19 pandemic.

In this era, it is widely accepted that data-driven methods can improve the delivery of care in a hospital setting. These data-driven methods are usually incorporated into end-user front ends which are referred to as Clinical Decision Support Systems (CDSS) and are widely recognized as an effective tool to improve patient safety (Jao et al., 2010) and reduce medical and medication error rates (Smith et al., 2006). To warrant the success of CDSS, their development needs to involve a multidisciplinary effort. Designing contextually relevant CDSS requires a continued partnership with domain experts since they are the ones who have a practical understanding of what attributes are used together, and how those parameters need to be modeled successfully, on a high level first, before the technical implementation. One reason for this, is, that any algorithms that are specifically developed for CDSS require, not just clinical knowledge integration, but also decision science to generate decisions that are in line with the clinical workflows, hospital practice, and the ways that clinical information is used, recycled, reevaluated, and reconfirmed, during the various stages of clinical decision making (Ruland et al., 2002). During the conceptual design of CDSS, it needs to be communicated that these systems are meant to provide clinicians with information that is intelligently filtered and presented at appropriate times, in line with, and always tailored around the different phases of care (HealthIT.gov, 2013). In other words, CDSS provides information that reflects not only decision-making, in general, but also the intellectual effort of physicians, themselves. CDSS cannot, therefore, be successful if they are designed with a ‘data staticity’ mindset. Instead, CDSS should provide dynamic recommendations, with fluidity, providing interactions with clinicians which are in line with each phase of decision-making and considering the longitudinal nature of health and disease.