89 Development and external validation of a multivariable model of pre-test likelihood of coronary artery disease based on a contemporary uk population, with comparison to existing risk models. (5th June 2017)

• Record Type: Journal Article
• Authors: Foley, James RJ; C Everett, Colin; Cairns, David; Bijsterveld, Petra; Ripley, David P; Plein, Sven; Sharples, Linda; Brown, Julia; Greenwood, John P
• Is Part Of: Heart. Volume 103(2017)Supplement 5
• Page Start: A64
• Page End: A66
• Publication Date: 2017-06-05
• Subjects: Coronary artery disease -- Pre-test likelihood -- Angiography; Heart -- Diseases -- Treatment -- Periodicals; Cardiology -- Periodicals; 616.12
• DOI: 10.1136/heartjnl-2017-311726.88
• Languages: English
• ISSNs: 1355-6037

Abstract

Abstract : Introduction: Pre-test likelihood models recommended in current UK and US guidelines have been shown to overestimate the probability of coronary artery disease(CAD). We aimed to develop a UK population-based multivariable risk model from CE-MARC(a contemporary study of stable CAD where all patients underwent X-ray coronary angiography) and validate it prospectively in the CE-MARC 2 trial population. Methods: CE-MARC (development population) enrolled patients between 2006–9 with suspected angina. Multivariable logistic regression modelled presence of significant stenosis (QCA≥70% in epicardial arteries or ≥50% in Left Main Stem) as a function of baseline demographic and clinical characteristics. The validation population were from the CE-MARC 2 trial (2012–16) that underwent angiography, plus additional low and high-risk patients from Leeds General Infirmary (2014–2016) to ensure adequate numbers across the full risk spectrum. Discrimination and calibration were assessed and compared to existing CAD consortium models(ESC 2013 guidelines) and the Duke Score(NICE CG95 2010 and US 2012 guidelines). Results: There were 675 patients in the development population, and 369 patients in the validation population(Table 1). A multivariable model was developed that included age, sex, angina type, smoking, diabetes, dyslipidaemia, hypertension, ECG Q-waves and ST segment abnormalities.The validation population had a similar CV risk profile. The new CE-MARC model discriminated well (c-statistic: 0.78 (95%CI 0.73–0.82)) and was well-calibrated (Table 2, Figure 1). In comparison, the Duke clinical model was very poorly calibrated (−1.016; −1.265 to −0.766; p<0.001) indicating substantial overestimation of pre-test likelihood compared to the average in the two populations (Figure 2, Table 2). The models used in the ESC 2013 guidelines under-estimated risk (0.74, 95% CI 0.51–0.97; p<0.001), but performed well once adjusted for different baseline risk levels (Table 2). Conclusion: The CE-MARC risk model, developed from a large contemporary UK population undergoing angiography, performed very well in the independent validation sample, without needing any adjustment for different risk prevalence or for miscalibration. In contrast, the earlier Duke risk score substantially over-predicted CAD risk, and remained poorly-calibrated even when this was corrected. The CAD consortium model (ESC 2013 guidelines), slightly under-estimated average CAD risk, but performed well once this was accounted for lower margin presents histogram of number of patients with each predicted risk score