Table 3

5-year MACE prediction model performance with or without blood test

MethodData usedBlood test requiredAUC (95% CI)
Proposed models with blood test
SCOREAge, gender, LDL cholesterol, HDL cholesterol, triglycerides, systolic blood pressure and smoking statusY0.682 (0.640 to 0.719)
SCORE 2Age, gender, total and HDL cholesterol, systolic blood pressure, diabetes statues and smoking statusY0.714 (0.674 to 0.753)
PCEAge, gender, total and HDL cholesterol, systolic blood pressure, diabetes statues and smoking statusY0.695 (0.656 to 0.73)
Framingham Risk ScoreAge, total and HDL cholesterol, systolic blood pressure, treatment for hypertension, smoking and diabetes statusY0.689 (0.648 to 0.730)
Logistic regression10 CVD risk factors*Y0.758 (0.729 to 0.784)
Neural networkY0.763 (0.734 to 0.790)
 XMACE+10 CVD risk factors*+end-to-end retinal image DL modelY0.769 (0.742 to 0.795)
Proposed models without blood test
Retinal imageEnd-to-end retinal image DL model following Poplin et al10N0.662 (0.632 to 0.694)
XMACEEstimated 10 CVD risk factors* from the imageN0.738 (0.710 to 0.766)
  • *10 cardiovascular disease (CVD) risk factors: Age, gender, BMI, systolic and diastolic blood pressure, HbA1c, HDL cholesterol, LDL cholesterol, triglycerides and smoking habits. The logistic regression and neural network used the same data source.

  • AUC, area under the receiver operating characteristic curve; BMI, body mass index; DL, deep learning; HDL, high-density lipoprotein; LDL, low-density lipoprotein; MACE, major adverse cardiovascular events; PCE, Pooled Cohort Equations; SCORE, Systematic COronary Risk Evaluation.