048 Computer beats doctor? Estimating the probability of acute coronary syndrome for individual patients. Issue 12 (21st November 2019)

• Record Type: Journal Article
• Authors: Oliver, Govind; Reynard, Charles; Morris, Niall; Body, Richard
• Is Part Of: Emergency medicine journal. Volume 36:Issue 12(2019)
• Page Start: 807
• Page End: 807
• Publication Date: 2019-11-21
• Subjects: Emergency medicine -- Periodicals; 616.02505
• DOI: 10.1136/emermed-2019-RCEM.48
• Languages: English
• ISSNs: 1472-0205

Abstract

Abstract : Chest pain is one of the most common reasons for patients attending the Emergency Department (ED). Accurately assessing for Acute Coronary Syndromes (ACS) remains a challenge. There is strong evidence supporting use of the Troponin-only Manchester Acute Coronary Syndrome (T-MACS) risk prediction model. How clinicians perform compared to these models is unknown. We aimed to externally validate the diagnostic accuracy of clinicians' estimated probability of ACS (gestalt) compared to the T-MACS calculated probability of ACS. The Bedside Evaluation of Sensitive Troponin prospective multi-centre diagnostic accuracy study included adults presenting to the ED with potential ACS. Alongside clinical, ECG and blood sample data, the emergency clinician recorded their estimated probability of ACS (%) following review. The probability of ACS was also calculated using T-MACS. The primary outcome was Major Adverse Cardiac Events (MACE) within 30-days. For this planned secondary analysis, patients from sites using the high-sensitivity cardiac troponin T (Roche Diagnostics Elecsys) were eligible. Of 782 included, 116 (14.8%) had MACE. The C-statistic for clinician gestalt and T-MACS were 0.76 (95% CI 0.71–0.81) and 0.93 (0.90–0.95) respectively (p<0.0001). Compared to T-MACS, clinicians overestimated the probability of ACS (positive bias 18.0%) and were less likely to stratify patients to extremes of probability. For 'rule out' of ACS, clinicians identified 72 (9.3%) patients as 'very low risk' (<2%) compared to 385 (49.2%) with T-MACS. For 'rule in' of ACS, clinicians identified 16 (2.1%) patients as 'high risk' of ACS (≥95%) in comparison with 50 (6.4%) with T-MACS. Assessment of model calibration comparing observed against predicted outcomes gave an R square of 0.78 and 0.97 for clinicians and T-MACS respectively. Clinician gestalt has inferior diagnostic accuracy to T-MACS. T-MACS requires a clinician's skill for appropriate application. Our conclusion is therefore not that computers are better, but that clinician performance can be augmented using T-MACS.