Identification of High-Risk Patients for Postoperative Myocardial Injury After CME Using Machine Learning: A 10-Year Multicenter Retrospective Study

Figures & data

Figure 1 Flow diagram of patients included in the study.

Table 1 Univariate and Multivariate Analyses of Variables Related to Postoperative Myocardial Injury

Variables		Univariate Analysis			Multivariate Analysis
		OR	95%CI	P-value	OR	95%CI	P-value
Sex	Female	Reference
	Male	1.123	[0.748, 1.687]	0.575
Age	<65	Reference			Reference
	≥65	10.832	[6.811, 17.228]	<0.001	4.655	[1.711, 13.402]	0.003
BMI	<25 kg/m^2	Reference			Reference
	≥25 kg/m^2	4.151	[2.744, 6.280]	<0.001	4.268	[1.698, 11.38]	0.003
ASA	<3	Reference
	≥3	1.046	[0.700, 1.563]	0.828
ALB	≥30g/L	Reference			Reference
	<30g/L	3.3	[2.146, 5.075]	<0.001	2.384	[0.901, 6.503]	0.082
NRS2002 score	<3	Reference
	≥3	0.771	[0.483, 1.231]	0.276
Drinking history	No	Reference			Reference
	Yes	4.163	[2.740, 6.326]	<0.001	3.144	[1.288, 8.031]	0.013
Smoking history	No	Reference			Reference
	Yes	9.308	[6.018, 14.398]	<0.001	10.163	[2.945, 41.443]	0.001
Surgical history	No	Reference
	Yes	1.257	[0.837, 1.887]	0.271
Anemia	No	Reference
	Yes	1.191	[0.767, 1.848]	0.436
Hyperlipidemia	No	Reference			Reference
	Yes	2.787	[1.855, 4.188]	<0.001	2.872	[1.161, 7.473]	0.025
Hypertension	No	Reference			Reference
	Yes	6.082	[3.932, 9.407]	<0.001	2.03	[0.765, 5.508]	0.157
Diabetes	No	Reference			Reference
	Yes	6.327	[4.135, 9.682]	<0.001	18.837	[7.069, 57.242]	<0.001
Preoperative tachycardia	No	Reference			Reference
	Yes	4.373	[2.887, 6.625]	<0.001	9.24	[3.536, 27.134]	<0.001
COPD	No	Reference			Reference
	Yes	6.198	[4.031, 9.529]	<0.001	4.861	[1.334, 18.447]	0.017
CKD	No	Reference
	Yes	1.33	[0.800, 2.209]	0.271
Adjuvant Radiotherapy	No	Reference
	Yes	1.491	[0.923, 2.409]	0.103
Adjuvant Chemotherapy	No	Reference
	Yes	1.008	[0.648, 1.568]	0.971
Surgical procedure	Laparoscopic surgery	Reference
	Open surgery	0.688	[0.460, 1.031]	0.07
Emergency surgery	No	Reference
	Yes	1.067	[0.694, 1.642]	0.767
Lymph node dissection	<12	Reference
	≥12	0.805	[0.514, 1.261]	0.344
Surgery time	<270 min	Reference			Reference
	≥270 min	6.901	[4.492, 10.602]	<0.001	4.202	[1.683, 11.11]	0.003
Intraoperative bleeding	<100 ml	Reference			Reference
	≥100 ml	4.626	[3.053, 7.009]	<0.001	4.288	[1.639, 11.914]	0.004
SpO^2	≥90%	Reference			Reference
	<90%	4.081	[2.695, 6.180]	<0.001	5.438	[2.15, 14.995]	0.001
Blood transfusion	No	Reference			Reference
	Yes	3.343	[2.073, 5.391]	<0.001	5.491	[1.758, 18.277]	0.004
Intraoperative tachycardia	No	Reference			Reference
	Yes	4.507	[2.960, 6.864]	<0.001	4.838	[1.914, 13.176]	0.001
T-stage	T1~T2	Reference			Reference
	T3~T4	4.343	[2.843, 6.636]	<0.001	3.706	[1.44, 9.97]	0.007
N-stage	N0	Reference			Reference
	N1~N2	4.082	[2.688, 6.198]	<0.001	3.097	[1.126, 8.992]	0.031
Tumor number	<2	Reference
	≥2	1.017	[0.611, 1.695]	0.947
Tumor size	<5 cm	Reference			Reference
	≥5 cm	3.062	[1.926, 4.868]	<0.001	3.658	[1.177, 11.87]	0.027
CEA level	<5 ng/ml	Reference
	≥5 ng/ml	1.11	[0.677, 1.821]	0.679
CA19-9 level	<37 U/mL	Reference
	≥37 U/mL	0.77	[0.464, 1.276]	0.31
PCT level	<0.05 ng/ml	Reference
	≥0.05 ng/ml	1.169	[0.771, 1.772]	0.461
CRP level	<10 mg/l	Reference			Reference
	≥10 mg/l	2.236	[1.485, 3.366]	<0.001	2.891	[1.115, 7.863]	0.032
SAA level	<10 mg/l	Reference
	≥10 mg/l	0.851	[0.570, 1.272]	0.432
NLR	<3	Reference			Reference
	≥3	3.219	[2.090, 4.959]	<0.001	4.646	[1.793, 12.782]	0.002

Abbreviations: OR, odds ratio; CI, confidence interval; BMI, body mass index; ASA, The American Society of Anesthesiologists; ALB, albumin; CA19-9, carbohydrate antigen 19-9; PCT, procalcitonin; CRP, C-reactive protein; SAA, serum amyloid A; NRS2002, nutrition risk screening 2002; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; SPO², percutaneous arterial oxygen saturation; NLR, neutrophil to lymphocyte ratio.

Figure 2 The variable influence factor ranking plots of the four models. (A) Variable importance ranking diagram of the XGBoost model. (B) Variable importance ranking diagram of the RF model. (C) Variable importance ranking diagram of the MLP model. (D) Variable importance ranking diagram of the KNN model.

Table 2 Evaluation of the Four Models

		AUC (95%CI)	Accuracy (95%CI)	Sensitivity (95%CI)	Specificity (95%CI)
KNN	Training set	0.981 (0.972–0.990)	0.951(0.944–0.959)	1.000(1.000–1.000)	0.900(0.895–0.905)
	Validation set	0.863 (0.740–0.979)	0.927(0.912–0.943)	0.763(0.722–0.803)	0.931(0.878–0.984)
XGBoost	Training set	0.997 (0.993–1.000)	0.974(0.970–0.978)	0.977(0.964–0.990)	0.976(0.971–0.981)
	Validation set	0.956 (0.920–0.991)	0.910(0.884–0.936)	0.973(0.941–1.005)	0.883(0.829–0.937)
RF	Training set	0.816 (0.761–0.871)	0.826(0.781–0.871)	0.837(0.778–0.896)	0.685(0.596–0.774)
	Validation set	0.814 (0.697–0.931)	0.837(0.788–0.886)	0.772(0.679–0.865)	0.788(0.712–0.865)
MLP	Training set	0.964 (0.945–0.984)	0.902(0.865–0.939)	0.907(0.878–0.936)	0.902(0.859–0.946)
	Validation set	0.940 (0.882–0.997)	0.895(0.872–0.918)	0.942(0.864–1.019)	0.862(0.815–0.909)

Abbreviations: AUC, area under the curve; RF, random forest; XGBoost, extreme gradient boosting; MLP, multilayer perceptron; KNN, k-nearest neighbor algorithm; CI, confidence interval.

Figure 3 Evaluation of the four models for predicting myocardial injury. (A) ROC curves for the training set of the four models. (B) ROC curves for the validation set of the four models. (C) Calibration plots of the four models. In a calibration plot, the 45° dotted line represents perfect calibration, where the observed and predicted values are perfectly matched. The distance between the observed and predicted values is shown by the curves, and a closer distance indicates greater accuracy of the model. (D) DCA curves of the four models. In a decision curve analysis, the x-axis represents the threshold probability of the outcome, while the y-axis represents the net benefit of making a decision based on the model’s prediction. The “All” curve represents the net benefit of treating all patients, regardless of the model’s prediction, while the “None” curve represents the net benefit of not treating any patients.

Figure 4 Internal validation of the XGBoost model. (A) ROC curve of the XGBoost model for the training set. (B) ROC curve of the XGBoost model for the validation set. (C) ROC curve of the XGBoost model for the test set. (D) External validation of the XGBoost model.

Figure 5 SHAP summary plot. Risk factors are arranged along the y-axis based on their importance, which is given by the mean of their absolute Shapley values. The higher the risk factor is positioned in the plot, the more important it is for the model.

Figure 6 SHAP force plot. SHAP analysis is a method used to explain the output of a machine learning model by attributing the contribution of each feature to the final prediction. The SHAP values represent the contribution of each feature to the difference between the actual prediction and the baseline prediction. In the horizontal line plot, each feature is represented by a bar with a corresponding SHAP value. The position of the bar along the horizontal line indicates the magnitude and direction of the impact of the feature on the prediction. The features are sorted by their absolute SHAP value, with the most important features on the left and the least important on the right. The color of the bar represents the direction of the feature’s impact. Blue bars indicate features that have a negative effect on the disease prediction, meaning that an increase in the feature value decreases the likelihood of disease. Red bars indicate features that have a positive effect on disease prediction, meaning that an increase in the feature value increases the likelihood of disease. (A) Predictive Analysis of Patient I. (B) Predictive Analysis of Patient II.

Data Sharing Statement

The original data presented in the study are included in Supplementary Material, and further inquiries can be directed to the corresponding author ([email protected]).