AI enabled clinical decision support system for breast cancer a peek behind the paper with Danny Ruta

Recently, we interviewed Danny Ruta (Guy’s & St Thomas’ NHS Foundation Trust; London, UK) for a peek behind his recent paper, 'Augmentation of a multidisciplinary team meeting with a clinical decision support system to triage breast cancer patients in the United Kingdom.' Find out more about how an AI enabled clinical decision support system can streamline cancer treatment and reduce clinical burden below.

Please provide a short summary of your paper

We set out to try and answer a simple but very important question for cancer care, particularly in the UK: is it possible to use an AI enabled clinical decision support system (CDSS) to streamline the multidisciplinary team (MDT) pathway for cancer treatment and reduce the clinical burden. It is an important question for the UK National Health Service because the volume of cancer patients requiring treatment is overwhelming current MDT pathways. Using a retrospective sample of breast cancer patients who had been referred to our breast MDT meeting (MDTM) at Guy’s Hospital (London, UK), we compared three approaches to triaging less complex breast cancer patients away from the MDTM and straight to an agreed standard of care: an AI-enabled CDSS acting ‘alone’; an AI supported two-person triage ‘team’ (one oncologist and one surgeon); and a two-person triage team acting alone. We found a slight variation in compliance with best practice between two independent triage teams. Providing the teams with AI support made a very small difference to their compliance. However, the most exciting finding was that the AI-enabled CDSS, acting alone, was able to triage 40% of patients to a standard of care, with 100% compliance with best practice.

Read the full paper here

What was the inspiration behind this research?

Our interest in AI-enabled clinical decision support began back in around 2014, after we talked to oncologists at Memorial Sloane Kettering (NY, USA) who had co-developed Watson for Oncology (WFO) and later met with IBM Watson Health (MA, USA). Our clinicians could not see significant benefit from the original use case for WFO to support MDTM decision making, but as the clinical workload on cancer MDTMs continued to grow, and waiting times for treatment increased, we collaborated with IBM Watson Health to identify a different use case that would meet this pressing clinical need.

What are the implications of AI-enabled clinical decision support systems for the NHS?

Since we completed our study, we have been evaluating the use of rule based expert system CDSSs to streamline cancer MDT pathways. To our surprise, this ‘old fashioned’ AI performs just as well, and in some cases better than a machine learning CDSS, especially where robust evidence based clinical guidelines exist. One can envisage the wide proliferation of expert system and machine learning CDSSs, both to automate less complex clinical decisions, and to drive compliance with best practice and reduce variation in clinical outcomes when supporting clinicians to make more complex decisions. The recent advances in large language models and generative AI are likely to dramatically improve the performance of AI enabled CDSSs in the next few years. As voice enabled and ‘listening’ AI systems evolve, we will see AI enabled CDSSs become seamlessly integrated within clinical workflows and the clinician-patient interaction. They will also move ‘upstream’ as patient self-management and health maintenance mobile apps. These developments are less than 10 years away, and the implications for the NHS and other health systems around the world will be truly transformational.

Could this technology be used outside of breast cancer?

The use case we have evaluated – as a tool to automate the streamlining of the breast cancer MDT pathway – can be applied to all cancer pathways. We have recently completed a prospective evaluation of an expert system CDSS, developed by a UK company called Deontics (London, UK), in our prostate MDT. It was able to triage 30% of all our prostate cancer patients straight to standard of care. On the strength of these findings, we are deploying the system in the prostate pathway, and over the next 12 months we plan to deploy the system across all 58 cancer MDT pathways. AI-enabled CDSSs can be used beyond cancer care: essentially wherever scientifically valid clinical guidelines exist.

What are some of the challenges you came across during this study?

The two biggest challenges were around IT and freeing up clinical time to engage in AI clinical evaluation. NHS IT departments are under resourced and risk averse, with many layers of information governance that take a great deal of time and effort to navigate. Clinicians are under so much pressure that it is extremely difficult for them to create space in their work plan to engage with AI clinical evaluation and with the re-design of clinical workflows for AI deployment.

If you had unlimited funding and resources, what would you research next?

If you mean unlimited funding and resources within our organization, then I am honestly not sure if I would spend the money on research per se. Whether you would call it research or R&D, or implementation, I think I would spend the money on building the capacity and capability for AI clinical evaluation and rapid deployment of AI. That would include: expanding our existing Clinical Scientific Computing team of AI scientists, evaluators and clinicians with protected time; creating a dedicated IT team for AI integration; and expanding the investment in our existing ‘Cogstack’ natural language processing program to convert all the unstructured data in our record systems to structured data. The latter is absolutely critical for AI research and development: it converts our ‘crude oil’ into ‘jet fuel’. If I were pushed to identify one area for research, it would be to use that refined data to train deep learning algorithms on real world evidence of the outcomes from the myriad of real time healthcare interventions, to create a new kind of data-driven evidence-based medicine.

About the author:

Danny Ruta is the AI Clinical Lead of Guy’s Cancer Centre at Guy’s & St Thomas’ NHS Foundation Trust (London, UK) and an Honorary Senior Lecturer at Kings College London (UK). The Guy’s Cancer Centre is partnering with the King’s Technology Evaluation Centre (KiTEC) at King’s College London, on the clinical evaluation of AI technology in cancer care. Danny currently leads on several AI clinical evaluations in cancer care, including evaluations of AI for clinical decision support, clinical trials matching and histopathology diagnosis. Danny has formerly held posts as Director of Public health for the London Borough of Lewisham and the City of Newcastle, and as Senior Lecturer in Epidemiology and Health Services Research at Newcastle University (UK) and Dundee University (UK). Danny developed the first evidence based national clinical guideline for epilepsy and evaluated its implementation in a large cluster randomised trial across Scotland. He has developed Patient Reported Outcome Measures recommended by the US FDA, used with patients across the NHS in England and throughout the world. Disclaimers: The opinions expressed in this interview are those of the author and do not necessarily reflect the views of Future Medicine AI Hub or Future Science Group.