Using intelligent automation to improve the triage and referral management pathway

Gastroenterology is one of the most pressured specialities with demand often outstripping clinical capacity. In the process of triaging GP referral letters into secondary care, every GP referral letter needs to be read and assessed by a specialist clinician to determine the appropriate care pathway. This process is time consuming for clinicians and creates delays for patients while being heavily guideline driven, making it a candidate for automation.

Situation

Scotland has an electronic referral and triage system whereby GPs submit a referral using a standard letter template which is triaged by a vetting clinician in secondary care. NHS Lothian receives 16,000 new gastroenterology referrals per year with urgent suspected cancer referrals on the rise. There are significant waiting list pressures in the speciality, with waiting times for routine appointments of up to 52 weeks.

Gastroenterology consultants at Western General Hospital in Edinburgh triage around 30 to 40 referrals per day. The triage process was complex, with over 120 outcome permutations identified. Validation of a sample of referrals detected significant variability in how clinical triage decisions were being made.

Aspiration

Automate the triage of referrals to the gastroenterology service thereby releasing clinician time to be spent with patients. Improve ease of communication between primary and secondary care, saving further time for clinicians, and use analytics to derive insights on referral trends.

Solution and impact

As part of a 3-year project partnered with Deloitte, NHS Lothian implemented a referral and intelligent triage (RITA) system to direct patients to treatment pathways, starting with the gastroenterology department.

The RITA model was built using natural language processing, a technique that enables the model to infer meaning from text, coupled with machine learning algorithms, which learn patterns from 12,000 historical referrals to make predictions on new referrals.

In January 2019, a pilot phase began to run referrals past the AI Triager in parallel to clinicians making their decisions. The AI decision was then compared to the decision made by the clinician and feedback provided to give a real view of how the solution would perform if it was fully automating decisions.

In January 2020 a live trial was run on 2 low risk cohorts of patients with the clinicians reviewing each automated referral after the AI had made its decision.

Impact

The impact has included:

  • faster triage
  • reduction of unwarranted variation in triaging
  • reduction of administrative burden
  • reduction of waiting list times by active management of the triage process
  • faster patient triage facilitated joint working across primary and secondary care

Functionality

The RITA model has 2 components.

1) Unattended AI Triager

  • a triage pipeline runs in the background without human interaction
  • receives referral letters through an application platform interface (API) call and makes predictions on urgency and action using trained AI models
  • makes prediction on urgency and action using trained AI models, for example ‘Urgent – Endoscopy Test’
  • if the referral meets the automation criteria, RITA will automatically complete the triage actions in the patient administration system and trigger the next step, such as sending the referral to the booking team

2) Attended Virtual Assistant

  • A virtual assistant (VA) that sits on the clinician’s desktop helps them with other triage tasks, such as writing back to the GP to provide advice or ask for tests.

Capabilities

  • Speed - the AI Triager takes only a few seconds to make predictions, while an experienced clinician could take 4 to 5 minutes
  • Availability - the AI Triager runs 24/7, hence the referrals will be processed immediately upon receiving them (for those that meet the automation criteria)
  • Accuracy - RITA is trained using advanced AI techniques on large volumes of historical data. It has been tested extensively, for example using the parallel run. The performance is in-line with the clinicians for the automated cohort

Scope

RITA has been developed and deployed in gastroenterology at NHS Lothian. Following this successful deployment Deloitte has been granted an NHSX AI Award to extend RITA to additional specialities and NHS Trusts across Scotland and England.

Key figures/quotes

The system has been able to automatically triage between 40% and 50% of urgent suspicion of cancer referrals since its implementation in January 2020.

The technology has been proved within a complex specialty which has over 120 different triage permutations.

The VA functionality has helped save time when communicating with primary care, with between 10% and 30% of referrals being processed using the VA.

Automating the referral triage process for urgent suspected cancer patients is estimated to reduce the overall referral to treatment (RTT) pathway by circa 1-3 days, or approximately 15% of a 2-week wait target.

Find out more

Download the Intelligent Automated Triage document (PDF, 689KB )

Key contact

Ian Arnott, clinical lead gastroenterology, Western General Hospital

ian.arnott@nhslothian.scot.nhs.uk

Philip Brocklehurst, Senior manager, Deloitte

pbrocklehurst@deloitte.co.uk