Blood cancer diagnosis and classification is highly complex, presenting major challenges for clinicians working in pressured environments. The creation of a decision tree application (DTA) which distils and applies World Health Organization (WHO) criteria proved highly successful – accurately diagnosing 94% of clinical cases tested.
One of the main challenges for accurate diagnosis is navigating and accessing the large amount of available information which is spread across a variety of clinical, morphological, genetic and radiological platforms. This difficulties are exacerbated as clinicians are often trying to make diagnoses during busy multi-disciplinary meetings.
A team including machine learning specialist and HDR UK Fellow Dr Daniel Bean and haematology consultant Dr Thomas Coats of the Royal Devon and Exeter NHS Trust distilled WHO guidelines and criteria into a digital decision tree. Their work, made possible by HDR UK support, resulted in the development of the DTA – which underwent repeated testing (using actual and artificially generated cases) and revision pushing up its accuracy way beyond the levels achieved by clinicians. The results of the multi-centre study were published in the British Society for Haematology journal eJHaem on 26 March, 2021. In testing with artificial cases – designed to include difficult to classify cases with atypical features – the DTA had an accuracy of 85% compared to 53% for individual doctors (range 43%-71%)
An algorithmic approach can be highly accurate and efficient in an especially difficult area of diagnosis by distilling complex data with multiple parameters. It can also help in standardising diagnoses, highlighting the desirability of additional tests, indicating that data points are missing and promoting consistent reporting.
An interdisciplinary approach was invaluable in the creation of the DTA, with clinical and technology experts working together to develop a tool that works in a way that suit clinicians. Dr Coats said: “It’s amazing how well this work has been received in the haematology community.”
Impact and outcomes
The open source DTA is now available. The team envisages that with some further development it could become an important tool for supporting clinicians’ decisions, or flagging up reason why they may wish to review their initial diagnosis.
However, the potential extends much further. It is already being further developed to help generate a clinical guideline in in acute myeloid leukaemia.
Another short-term aim is to extend the DTA’s use into prognosis, something that can be especially challenging in haemato-oncology. As the platform is quite generic they also believe it can be readily adapted for other medical and non-medical applications.
Insights from the Impact Committee
The HDR UK Impact Committee judges were highly impressed with this paper which they said demonstrated high level of potential impact in bringing about better care – one of HDR UK’s national priorities. It was also commended for being leading in its quality and originality, significance and rigour.
Linking health data to understand deaths of homeless patients
27 June 2019
A five-year research project led by Dr Robert Aldridge at the UCL Institute of Health Informatics involved the analysis of medical records for nearly 4,000 homeless people, 600 of whom...
Predicting suicide risk using health data at scale
27 June 2019
More than three-quarters of a million people globally take their own lives every year, to devastating personal, social and economic cost. Whilst past research has shown links between suicide and...