Enhancing prehospital triage for patients with suspected cardiac chest pain using prediction models

Abstract

Background: Chest pain, a primary symptom of acute coronary syndromes (ACS), frequently prompts ambulance calls and emergency department (ED) visits. Despite this, a significant number of patients transported for chest pain ultimately receive diagnoses of self-limiting, non-cardiac conditions. This leads to systematic over-triage, elevating ambulance resource utilization and contributing to crowding in EDs. Aim: To understand and improve the accuracy of current processes for telephone triage and prehospital assessment of patients with acute chest pain. For telephone triage, specific objectives included to systematically review the literature; gain consensus on the life-threatening conditions (LTCs) that should be identified by call handlers; and to derive and validate a prediction model. I also aimed to refine and validate prediction models to enhance prehospital risk stratification by paramedics. Methods: I completed (a) a systematic review searching three databases with narrative synthesis; (b) a Delphi study to define the LTCs that should guide future telephone triage; and (c) a retrospective cohort study using North West Ambulance Services (NWAS) data linked with Manchester University NHS Foundation Trust (MFT) to derive and validate a prediction model. Further, I completed secondary analyses of a multi-centre prospective diagnostic test accuracy study to create meta-models of established decision aids using stacked regression, and I validated the Manchester only Acute Coronary Syndrome ECG (MACS-ECG) prediction model in prehospital setting. Result: The systematic review identified three relevant papers supporting the feasibility of using prediction models to reduce over-triage by telephone but highlighting a paucity of data. The Delphi study identified 26 LTCs that should receive a priority 1 or 2 ambulance response. Due to a cyber-attack, results of the derivation and validation of a new prediction model for telephone triage cannot be presented. The existing MACS-ECG risk model achieved a sensitivity of 2.3% (95% CI 0.3-8.1%) and specificity of 99.5% (95% CI 98.6-99.9%) for myocardial infarction with poor calibration (gradient 0.0628, intercept 0.0852). Meta-modelling did not enhance the accuracy of established prediction models. The paramedic ECG interpretation showed a 30.8% (95% CI: 22.1-40.6%) sensitivity and 89.7% (95% CI: 87.2-91.9) specificity. Conclusions: This work has (a) identified an important evidence gap to guide telephone triage for chest pain, (b) defined outcome measures for such future research, and (c) identified that currently available prediction models do not have any advantage over human interpretation of an ECG by a paramedic. Future research should focus on completing the work to develop a new prediction model a telephone triage and supporting training paramedics for ECG interpretation.