Background Following the introduction of electronic patient clinical records, ambulance service managers wished to combine clinical and operational data to devise a method of risk stratifying 999 calls by the MPDS disposition code assigned at call triage. Initial aims were to establish the risk threshold if an ambulance was no longer routinely dispatched.

Methods Data selected were representative of high or low clinical risk, and reliably recorded. The following ‘risk factors’ were chosen:

• Call outcome

• Emergency conditions

• Clinical interventions

• Medications administered

With expert local opinion, a scoring algorithm was created using weighted factor scores to create an aggregate risk score for each MPDS code. It was also designed to distribute codes along a ‘risk range’, allowing for thresholds setting suitable to the specific purpose of individual projects. These factors and their scores were captured alongside contextual information and to date contains over 1.4 million records over 3 years.

In collaboration with academic colleagues, we also developed an AI model to refine the algorithm used to reflect acuity. With one year of data the tool did not demonstrate the sensitivity or specificity to reliably contribute to prediction, however this exercise may be repeated now there is a greater volume of data.

Applications: This Tool has been successfully used for a variety of purposes:

• Developing the Enhanced Hear and Treat policy

• Assessing risk of code downgrades in the pandemic response

• Identifying codes suitable for automatic specialist clinician allocations

• Supplementing analysis of harm caused by long response delays

• Identifying codes for protection within End of Shift protocols

• Providing intelligence to aid national decisions on code categorisation

Next steps: The Tool continues to assist in decision-making locally. Future ambitions include:

• Validation of the scoring algorithms

• Process automation to ensure more timely data is available

• Collaboration to improve the variety and volume of data