Applying medical knowledge to medical practice at scale is non-trivial. National clinical guidelines serve as a backbone of our health practice and are used daily in thousands of general practice surgeries and hospitals nation-wide. Yet, these guidelines are captured in large text documents ill-suited to being deployed in digital tools. Using text guidelines with generative AI techniques is not feasible in scenarios where a patient may be suffering from multiple conditions (multi-morbidities). Transforming these guidelines into computable structures, updated from latest evidence in a structured, tractable manner, would support shared decision making, and facilitate dealing with multi-morbidities, enabling multiple guidelines need to be taken into account simultaneously.
The project will prototype a platform for developing guideline-based decision support tools in UK general practice, combining generative AI to create symbolic representations which can be then used in tractable explainable algorithms, allowing human validation. This approach will be supported by an evaluation framework for assessing the design and performance of the resulting tools to enable them to be certified as medical devices and deployed in practice. The platform will be evaluated on combinations of most common multi-morbidities in UK general practice, in collaboration with NICE.
There are three main challenges to be addressed: 1) how to automate knowledge acquisition from narrative guidelines at national and international scale; 2) how to represent this knowledge in computable, symbolic form; 3) how to provide transparent reasoning engines that can provide clinical decision support while preserving provenance and auditability.