Development and validation of predictive models for prognostic assessment in patients with acute non-ST-segment elevation myocardial infarction: a multicentre real-world study from China.

BACKGROUND

Few models specifically predict the prognoses of non-ST-segment elevation myocardial infarction (NSTEMI) patients. Additionally, these models have inadequate predictive performance for combined ischemic events other than all-cause death.

METHODS

This analysis included 4845 NSTEMI patients who completed a follow-up within 180 days after discharge. Data from three hospitals were randomly selected for the model's training and tuning, and data from two hospitals were used for external validation. Five machine learning algorithms were used to develop the predictive models. The receiver operating characteristic curve was plotted to reflect the model's sensitivity and accuracy. The area under the curve (AUC) was calculated to evaluate the model's classification performance. The model's prediction performance was compared with the global registry of acute coronary events (GRACE) risk score.

RESULTS

Incorporating clinical variables such as coronary artery lesions, reperfusion strategies, laboratory tests, and echocardiography parameters, predictive models were developed and validated to assess the incidence risk of adverse clinical outcomes in NSTEMI patients. The logistic regression model performed well in predicting the incidence risk of all-cause death (sensitivity 82.7%,specificity 75.7%, AUC 0.869) and moderately in predicting the incidence risk of major adverse cardiac and cerebrovascular events (sensitivity 60.5%, specificity 67.7%, AUC 0.700).Compared with the GRACE risk score, this model demonstrated significant improvements in both discrimination and incremental prognostic value (all p < 0.05).

CONCLUSION

The machine learning-based prognosis model significantly improves the performance of the current risk assessment model for predicting adverse clinical outcomes in NSTEMI patients.