Derivation, internal validation, and recalibration of a cardiovascular risk score for Latin America and the Caribbean (Globorisk-LAC): A pooled analysis of cohort studies

Abstract

SummaryBackgroundRisk stratification is a cornerstone of cardiovascular disease (CVD) prevention and a main strategy proposed to achieve global goals of reducing premature CVD deaths. There are no cardiovascular risk scores based on data from Latin America and the Caribbean (LAC) and it is unknown how well risk scores based on European and North American cohorts represent true risk among LAC populations.MethodsWe developed a CVD (including coronary heart disease and stroke) risk score for fatal/non-fatal events using pooled data from 9 prospective cohorts with 21,378 participants and 1,202 events. We developed laboratory-based (systolic blood pressure, total cholesterol, diabetes, and smoking), and office-based (body mass index replaced total cholesterol and diabetes) models. We used Cox proportional hazards and held back a subset of participants to internally validate our models by estimating Harrell's C-statistic and calibration slopes.FindingsThe C-statistic for the laboratory-based model was 72% (70–74%), the calibration slope was 0.994 (0.934–1.055) among men and 0.852 (0.761–0.942) among women; for the office-based model the C-statistic was 71% (69–72%) and the calibration slope was 1.028 (0.980–1.076) among men and 0.811 (0.663–0.958) among women. In the pooled sample, using a 20% risk threshold, the laboratory-based model had sensitivity of 21.9% and specificity of 94.2%. Lowering the threshold to 10% increased sensitivity to 52.3% and reduced specificity to 78.7%.InterpretationThe cardiovascular risk score herein developed had adequate discrimination and calibration. The Globorisk-LAC would be more appropriate for LAC than the current global or regional risk scores. This work provides a tool to strengthen risk-based cardiovascular prevention in LAC.FundingWellcome Trust (214185/Z/18/Z)