HARNESSING MACHINE LEARNING IN CLINICAL DECISION SUPPORT: THEORY AND PRACTICE

Abstract

Decision making is a central activity in all clinical professions. Clinical decisions bear wellbeing and economic risks and consequences for patients, families, employers, and national economies. Thus, clinicians should employ sound scientific knowledge to promote optimum decision outcomes. Evidence-based medicine organizes clinical decision-making activities with philosophical, ethical, and methodological foundations to ensure accessibility to the best scientific knowledge to inform clinical decision-making. But high-quality evidence could be lacking or methodically, ethically, or economically unfeasible, compelling clinicians to make risk-bearing decisions without an ideal evidence-base. In uncertain clinical conditions, decisions` outcomes are probabilistic and the clinicians' 'knowledge about outcomes are limited which make the clinical decision in significant need for aid. The advent of electronic health records and the data warehouse technology boosted healthcare industry capacity to generate and store vast amounts of diverse data types. Advanced analytical and computational techniques in artificial intelligence and medical informatics provide unprecedented opportunities to galvanize knowledge deployment in clinical decision, bridging gaps in clinical knowledge. The ability of machine learning approaches to handle the large-scale and diverse data addresses some evidence-based medicine challenges, providing real-time, cost-effective evidence. Nevertheless, despite our beliefs that artificial intelligence and data science may have the potentials to transform the clinical practice, the utilization of artificial intelligence in healthcare is still poor, and the full benefits are not reaped. The adoption of machine learning in healthcare faces several epistemological, methodological, and ethical challenges that make the integration between the machine learning and the evidence-based medicine a hard mission. This dissertation adopted a literature-based reconceptualization to help comprehensively understand the two paradigms, to guide the paradigms reconciliation agenda and to determine the reasons behind the slow adoption of machine learning in healthcare industry. Secondly, we followed an interpretive research design in order to propose a roadmap that aims to enhance the adoption of machine learning in clinical decision support. To validate the theoretical work, we conducted five empirical studies in collaboration with trauma surgery section at Hamad Medical Corporation. In these studies, we developed several ML predictive algorithms to address real-life clinical issues that are faced by the trauma surgery clinicians and predict the prognosis of patients who suffered from Traumatic Brain Injury which include mortality, prolonged mechanical ventilation, ventilator associated pneumonia and prolonged in-hospital length of stay. Subsequently, this dissertation determines how machine learning and evidence-based medicine can be reconciled through proposing a novel pragmatic reconciliation framework that guides the clinicians and the scholars on how to benefit from the synergistic effect of both paradigms. In addition, this dissertation determines the factors that negatively affect the adoption of machine learning in clinical decision support and proposes an original theoretical framework that draws a strategic road map towards the effective adoption of machine learning in healthcare. Furthermore, the dissertation sheds the light on the future research directives that may enhance the compatibility between the data science and the evidence-based medicine paradigms in order to augment the clinicians' capacity to make high-quality informed decisions.