Improving event prediction using general practitioner clinical judgement in a digital risk stratification model: a...

Area of Science:

• Health Informatics
• Clinical Risk Management
• Primary Care Medicine

Background:

• Existing electronic health record (EHR) tools for predicting urgent care and mortality risk often lack external validation and can lead to inefficient patient escalation.
• Clinical judgment is crucial for accurate risk prediction, especially in ruling out serious conditions.

Purpose of the Study:

• To evaluate the performance of a digitally driven risk stratification model combined with general practitioner (GP) global clinical judgement (GCJ).
• To identify patients at risk of escalating urgent care and mortality events using a hybrid approach.

Main Methods:

• A clinically risk-stratified cohort study was conducted across 6 GP practices in a UK city.
• An initial digital stratification identified an 'Escalated' group based on 7 risk factors.
• The 'Escalated' group was further stratified by GPs into 'Concern' and 'No concern' categories using GCJ.

Main Results:

• Out of 31,392 patients, 3968 were in the 'Escalated' group, with 518 (1.7%) categorized as 'Concern' by GPs.
• The 30-day event rate (unscheduled care or death) was significantly higher in the 'Concern' group (168.0/1000) compared to the whole population (19.0/1000).
• GP assessment significantly improved negative predictive value, with an odds ratio of 0.25 (p < 0.001) for de-escalation from 'Concern' to 'No concern'.

Conclusions:

• The digital risk stratification model demonstrated good performance on its own.
• Integrating GP clinical judgement significantly enhanced the accuracy of risk prediction.
• The addition of GP GCJ notably improved the negative predictive value of the risk stratification model.