Abstract 11614: Prognostic Model for Stable Coronary Artery Disease Based on 101,956 Patients From Linked Electronic Health Records

Abstract

Context: There are no validated prognostic models for stable coronary artery disease in clinical use.

Objectives: We developed, validated and evaluated the clinical utility of prognostic models based on the type of stable coronary artery disease, clinical history and clinically assessed biomarkers.

Methods: We linked electronic health records between primary care, hospital admissions, the national myocardial infarction registry and cause-specific mortality. We studied 101,956 patients diagnosed with stable coronary disease between 2000 - 2010. We compared four models (m1 to m4). Model m1 included age, sex, coronary artery disease type and revascularization status, m2 added clinical history (of recurrent MI, heart failure, peripheral arterial disease, diabetes and smoking), m3 added biomarkers (diastolic blood pressure, heart rate, low-density lipoprotein cholesterol, glomerular filtration rate, white cell count and haemoglobin) and m4 added C-reactive protein. Main Outcome: Non-fatal MI or coronary death (n=13,335) over 4.5 years mean follow-up.

Results: The C-index increased from 0.711 in m1 to 0.759 in m4. The most clinically useful biomarkers were haemoglobin, glomerular filtration rate and heart rate. Addition of C-reactive protein may have further improved clinical utility measures but confidence intervals were wide because only few people had C-reactive protein measured. For routine clinical assessment we propose model m3.

Conclusion: A risk score combining readily available clinical information can reliably predict long-term coronary outcomes across the spectrum of stable coronary disease.