Clever Referrals – Building an Artificial Expert

Dafydd Loughran, ABCi Clinical Leadership Fellow

Aneurin Bevan University Health Board


Currently all GP referrals are ‘vetted’ in secondary care and a priority assigned. This process accounts for between 0.5% – 1% of all Welsh NHS Consultant time. These decisions are usually made from only a few lines of information from the referring GP.


We hypothesised that we could train an algorithm to predict the prioritisation decisions of our consultants, thereby allowing us to release this time, as well as providing instant information to referring GP’s and patients of the likely timescale to be seen in secondary care.


With the aid of several senior clinicians, ABCi’s mathematical modelling capability, and NWIS, decision-tree algorithms were developed, initially for secondary care breast services referrals, to be incorporated within the national electronic WCCG referral architecture.

GP assessment of case priority, either routine, urgent, or Urgent Suspected Cancer (USC), was equivalent to Consultant assessment in only 50% of cases, significantly surpassed by algorithm equivalence in 75-90% of cases.

Despite minimal algorithm under-prioritisation rates, patient safety was of paramount importance to the team. Any cases which the algorithm classed to be a ‘USC’ priority could be deemed as such, requiring no further human prioritisation, but all others – routine and urgent – would continue to be reviewed by a consultant as a safety net.


This approach delivers a potential time efficiency saving to the Welsh NHS of £256-641k per year once scaled nationally, releasing consultants to deliver direct patient care, whilst also providing GPs and patients with instant information regarding referral timeframes.


Part of cohort Bevan Exemplar Projects 2016-17