The overarching objectives of the research of the Medical Knowledge and Decision Support group are to accelerate translation and integration of evidence into the clinic and to ensure that technological progress is aligned with the needs of clinicians.
Semantic AI: Our research aims to develop and apply novel artificial intelligence approaches that combine prior knowledge, often formally represented as ontologies, with large-scale machine learning from data in order to make knowledge-informed predictions. We also work on transfer learning strategies that use large-scale pre-training on general-purpose datasets with fine-tuning for particular applications to enable powerful applied machine learning from very small datasets, as may be the case for many biomedical applications. We are developing these methods currently for applications relating to metabolism and physical activity, however, they are broadly applicable to many different biomedical data types and application areas.
Clinican Experiences, Interpretability and Trust: Our research aims to understand and amplify clinician experiences and expectations of artificial intelligence systems and decision support systems, and match those to available technological capabilities and opportunities in order to strategically steer future developments towards greater clinical impact. In particular, we develop and evaluate approaches for enhanced interpretability of predictive models and conduct qualitative research with users and clinicians in order to explore experiences with, and trust in, artificial intelligence as mediated by different forms of visualisation and explanation.
Evidence Synthesis and Translation: We apply machine learning and other artificial intelligence strategies such as knowledge representation and reasoning in order to advance methods for evidence management, including evidence discovery, information extraction, evidence synthesis, and methods for the translation of research in animal models into applicable prioritised hypotheses for human investigations. These approaches aim to advance knowledge-driven approaches to support partial automation of biomedical evidence synthesis and thereby to accelerate the translation of evidence into implementation in the clinics.
Medical Knowledge and Decision Support – at the interface between technologies and humans
Digitalisation is rapidly advancing into the clinic. But clinician perspectives and insights have at times been neglected in technological innovations and implementations, leading to solutions that do not work well within existing workflows and processes, and add additional burdens within already strained clinical workloads. Can clinician and patient perspectives be considered more closely in the design and development of digital tools?
Medical knowledge and decision support face challenges in integration and translation of evidence into practice through all the steps from primary research to clinical implementation. The medical evidence base grows exponentially and can be fragmented. It can be difficult to identify, assess, evaluate and gain an overview of all the evidence relevant to a given topic. The process of manual synthesis of evidence and translation into practice is slow and laborious. Can synthesis, translation and implementation be accelerated through digital tools?
While there have been many recent advances in the algorithms and technologies used in clinical decision support, these have been slow to translate into usable systems that are implemented in practice and have a measurable impact on health outcomes. This may be in part because of gaps between what the technologies can offer and the contexts in which successful implementations need to be delivered. Can design and implementation considerations be harnessed to develop a new generation of decision support tools that bridge the gap between technology and clinical need?