Predicting In-Hospital Mortality in Myocardial Infarction: A Nomogram-Based Retrospective Analysis of the MIMIC-IV Database

Abstract

Background: Despite significant advancements in early reperfusion therapy and pharmacological treatment, which have reduced mortality rates after myocardial infarction in recent decades, the in-hospital mortality rate remains high due to factors such as rapid disease progression, comorbid conditions, and potential complications. We aimed to develop and validate a predictive model for in-hospital mortality in myocardial infarction patients.

Methods: LASSO regression analysis, univariate analysis, and multivariate logistic analysis were used to construct the nomogram in the training set, followed by model comparison, internal validation, and sensitivity analysis.

Results: The analysis comprised 4688 patients in total. The population of patients was randomly assigned to the training set (n = 3512) and validation set (n = 1176). According to the results of LASSO regression analysis and other results, our nomogram contained a total of 10 independent variables related to patient death, including age, respiratory rate, blood glucose, lactate, PTT, BUN, cerebrovascular disease, chronic lung disease, mild liver disease, and metastatic solid cancer. Moreover, the web calculator and nomogram performed exceptionally well at predicting in-hospital death in myocardial infarction patients. The AUC for the training and validation sets' respective prediction models was 0.869 (95% CI: 0.849-0.889) and 0.846 (95% CI: 0.807-0.875) (p<0.01). Compared to the Sequential Organ Failure Assessment (SOFA), the nomogram showed greater discrimination in the training and validation sets, and the calibration plots demonstrated an adequate fit for the nomogram in predicting the risk of in-hospital mortality in both groups. The decision curve analysis (DCA) of the nomogram demonstrated a higher net benefit in the training and validation sets and in terms of clinical usefulness than the SOFA.

Conclusion: We developed a useful nomogram model and developed a nomogram-based web calculator to predict in-hospital mortality in myocardial infarction patients, which will support doctors in patient counseling and logical diagnosis and therapy.

Keywords: in-hospital mortality; myocardial infarction; nomogram; predictive models; web calculator.

Figure 1

(A) Flowchart of patient selection (N=4688); (B) Model development flowchart.

Figure 2

LASSO regression: selection of significant parameters in variables in the training set. (A)Ten-fold cross-validation for tuning parameter selection in the LASSO model; (B) LASSO coefficient profiles. Predictor variables selected by LASSO. The LASSO was used for regression of high dimensional predictors. The advantage of the LASSO regression method that we adopted is that by penalizing regression on all variable coefficients, the relatively minor independent coefficients are reduced to zero and are thus disregarded in the modeling. Using 10-fold cross-validation, the vertical line was drawn at the value chosen, where the optimal results was 18 features with non-zero coefficients.

Figure 3

The nomogram for predicting in-hospital mortality in patients with myocardial infarction. The nomogram included 10 variables, including Age, Respiratory rate, Glucose, PTT, Lactate, Cerebrovascular disease, Chronic pulmonary disease, Mild liver disease, and Metastatic solid tumor. In order to calculate the score when utilizing the nomogram, a vertical line should be drawn from each variable to the “Points” line. The values should then be totaled to determine the final score. Finally, to determine the in-hospital mortality of the myocardial infarction patients, a vertical line is drawn downward from the “Total Points.”.

Figure 4

The ROC curve for predictive model. (A)Training set. (B)Validation set. Receiver Operating Characteristic (ROC) curve and area under the ROC curve (AUC). (A) The ROC in the training set; (B) The ROC in the internal validation set.

Figure 5

The ROC curve for predictive model. (A)Training set. (B)Validation set. The ROC curve combined model, the prediction model (Model 1) and SOFA (Model 2) in the Training set (A) and Validation set (B). The combined model is incorporated by all the independent risk variables. The prediction nomogram includes Age, Respiratory rate, Glucose, PTT, Lactate, Cerebrovascular disease, Chronic pulmonary disease, Mild liver disease, and Metastatic solid tumor.

Figure 6

Calibration plots. (A) Training set. (B) Validation set. Bootstrap-based calibration curves in the training set (A) and validation set (B) Calibration plots. Show the consistency of the predicted potentiality and actual values of the training set and validation set. A excellent conformity between observation and prediction is seen in both sets, even though the apparent curve and bias-corrected curve both somewhat diverged from the reference line.

Figure 7

Decision Curve Analysis. (A) Training set. (B) Validation set. The DCA curve of the prediction model (Model 1) and SOFA (Model 2) in the training set (A) and validation set (B). Show the net benefit, represented by a backslash with a negative slope, in the training set and validation set.

Figure 8

Online version of the nomogram. http://www.empowerstats.net/pmodel/?m=20358_inhospitalmortality.