Российский кардиологический журнал (Jul 2023)

Prediction of in-hospital mortality in patients with ST-segment elevation acute myocardial infarction after percutaneous coronary intervention

  • B. I. Geltser,
  • K. I. Shahgeldyan,
  • I. G. Domzhalov,
  • N. S. Kuksin,
  • E. A. Kokarev,
  • V. N. Kotelnikov,
  • V. Yu. Rublev

DOI
https://doi.org/10.15829/1560-4071-2023-5414
Journal volume & issue
Vol. 28, no. 6

Abstract

Read online

Aim. Development of models for predicting in-hospital mortality (IHM) in patients with ST-segment elevation myocardial infarction (STEMI) after percutaneous coronary intervention (PCI) based on multivariate logistic regression (MLR).Material and methods. This retrospective cohort study of 4735 electronic health records of patients (3249 men and 1486 women) with STEMI aged 26 to 93 years with a median of 63 years who underwent PCI was performed. Two groups of persons were identified, the first of which consisted of 321 (6,8%) patients who died in the hospital, while the second — 4413 (93,2%) patients with a favorable PCI outcome. To develop predictive models, univariate logistic regression (ULR) and MLR were used. Model accuracy was assessed using 3 following metrics: area under the ROC curve (AUC), sensitivity, and specificity. The end point was represented by the IHM score in STEMI patients after PCI.Results. Statistical analysis made it possible to identify factors that are linearly associated with IHM. ULR was used to determine their weight coefficients characterizing the predictive potential. IHM predictive algorithms based on GRACE scale predictors, represented both by ULR model and by 5 factors in continuous MLR model, had acceptable predictive accuracy (AUC — 0,83 and 0,86, respectively). The MLR model had the best quality metrics, the structure of which, in addition to 5 GRACE factors, included left ventricular ejection fraction (LVEF) parameters and white blood cell (WBC) count (AUC — 0,93, sensitivity — 0,87, specificity — 0,86) . The greatest contribution to endpoint was associated with the Killip class and LVEF, and the smallest contribution was associated with WBC and the age of patients.Conclusion. The predictive accuracy of the developed MLR models was higher than that of the GRACE score. The model with the structure represented by 5 fac­tors GRACE, LV EF and WBC had the highest quality metrics.

Keywords