Predicting long-term prognosis after percutaneous coronary intervention in patients with new onset ST-elevation myocardial infarction: development and external validation of a nomogram model

Abstract

Background: The triglyceride glucose (TyG) index is a well-established biomarker for insulin resistance (IR) that shows correlation with poor outcomes in patients with coronary artery disease. We aimed to integrate the TyG index with clinical data in a prediction nomogram for the long-term prognosis of new onset ST-elevation myocardial infarction (STEMI) following primary percutaneous coronary intervention (PCI) .

Methods: This retrospective study included new-onset STEMI patients admitted at two heart centers for emergency PCI from December 2015 to March 2018 in development and independent validation cohorts. Potential risk factors were screened applying least absolute shrinkage and selection operator (LASSO) regression. Multiple Cox regression was employed to identify independent risk factors for prediction nomogram construction. Nomogram performance was assessed based on receiver operating characteristic curve analysis, calibration curves, Harrell's C-index and decision curve analysis (DCA).

Results: In total, 404 patients were assigned to the development cohort and 169 to the independent validation cohort. The constructed nomogram included four clinical variables: age, diabetes mellitus, current smoking, and TyG index. The Harrell's C-index values for the nomogram were 0.772 (95% confidence interval [CI]: 0.721-0.823) in the development cohort and 0.736 (95%CI: 0.656-0.816) in the independent validation cohort. Significant correlation was found between the predicted and actual outcomes in both cohorts, indicating that the nomogram is well calibrated. DCA confirmed the clinical value of the development prediction nomogram.

Conclusions: Our validated prediction nomogram based on the TyG index and electronic health records data was shown to provide accurate and reliable discrimination of new-onset STEMI patients at high- and low-risk for major adverse cardiac events at 2, 3 and 5 years following emergency PCI.

Keywords: Major adverse cardiovascular events; Percutaneous coronary intervention; Prediction nomogram; ST-elevation myocardial infarction; Triglyceride glucose index.

Fig. 1

Inclusion of new onset ST-elevation myocardial infarction (STEMI) patients following primary percutaneous coronary intervention (PCI) (n = 706) and establishment of the development cohort (n = 404) and the external validation cohort (n = 169)

Fig. 2

LASSO model coefficient trendlines of 22 variables for long-term prognosis. The abscissa represents the optimal tuning parameter λ, and the ordinate represents the regression coefficient

Fig. 3

Tuning parameter (lambda, λ) selection cross-validation error curve. The abscissa represents the optimal tuning parameter λ, and the ordinate represents the binomial deviation (binomial deviance). Employing the optimized four nonzero coefficients derived by 10-fold cross-validation, the vertical line was drawn, and further multivariate Cox regression analysis was conducted

Fig. 4

Forest plot of hazard ratios (HRs) for the independent prognostic variables identified by multivariate Cox regression in the development cohort

Fig. 5

Prediction nomogram derived from multivariate Cox regression analysis. The vertical line is followed from the value of each clinical characteristic on the top row to assign points for each patient, and all six characteristics are added into a single matrix to obtain the point total (middle row). To calculate the probability of MACEs, a vertical line is drawn across the bottom two rows, applying the total points

Fig. 6

ROC curve analysis of the predictive accuracy of the nomogram for MACEs in the development (A) and independent validation cohorts (B)

Fig. 7

Calibration curves for MACE risk predictors in the development (A) and independent validation cohorts (B). Gray dotted line represents the ideal reference line where predicted probabilities would match the observed the risk possibility of MACEs. Thus, high-quality prediction of MACE risk is represented by the red line, and the black dashed line shows more precise predictions of long-term outcome

Fig. 8

Decision curve analysis (DCA) for predicting 3-year MACE risk in the development (A) and independent validation cohorts (C) and the 5-year MACE risk in the development (B) and independent validation cohorts (D). The graph depicts the expected net benefit for each patient with respect to the risk for MACEs as predicted by the nomogram. With increasing model curve length, the net benefit increases

Fig. 9

Cumulative MACE-free survival of patients in the development (A) and independent validation cohorts (B) stratified according to MACE risk (high risk vs. low risk) using the median nomogram score