The Effectiveness and Challenges of Data Oriented Patient Assessment: A Case Study in Head and Neck Cancer
Session type: Poster / e-Poster / Silent Theatre session
Theme: Diagnosis and therapy
The continued advancement in the digitisation of knowledge has meant the collection and use of data for predictive purposes can assist in the delivery of health care, specifically with referral and management of suspected cancer patients. Most patients referred from primary to secondary care are done so using the existing '2 week wait' (2ww) pathway.
Data relating to 2ww referral of patients with suspected head & neck cancer referred to 2 large tertiary referral centres in the UK were collated both prospectively and retrospectively. The combined dataset was subjected to analysis via a variety of machine learning techniques using k-fold cross validation to faciliate risk stratification. Type II errors were minimised using class weighting and manipulation of posterior probabilities.
Over 5000 patient entries were analysed. On a 10% validation corpus we repeatably demonstrate that 26% of patients referred with signs or symptoms of suspected head & neck cancer could have been safely managed out with the 2 week wait referral pathway, with minimal (<1/1000 chance) risk of a false negative ascribed non-cancer diagnosis.
Our techniques in this setting demonstrate the rationale for utilising a data rich, meachine learning based tool to provide safe and accurate decision support in the management of patients with potential head and neck cancer diagnosis. It follows that use of such a tool could potentially lead to significant cost savings in the management of these referrals from primary to secondary care, however its implementation will need to be very carefully managed alongside appropriate quality assurance measures and changes to existing patient management pathways.