Development and validation of a nomogram model to predict postoperative delirium after resection of esophageal cancer

Abstract

The study aimed to establish and validate a nomogram model to predict postoperative delirium (POD) among esophageal cancer resection patients. Clinical data of 396 patients with esophageal cancer who underwent esophagectomy from November 2020 to June 2023 in the electronic medical records of cardiothoracic Surgery, Affiliated Hospital of Jiangnan University. Participants were randomly divided into training and testing sets in a 7:3 ratio. Predictors were screened by Least absolute shrinkage and selection operator (LASSO) regression analysis and a nomogram model was built. The discrimination and consistency of the model were evaluated using the area under the receiver operating characteristic curve (AUC), C-statistic, Brier score, Hosmer-Lemeshow goodness-of-fit test, calibration curve and decision curve analysis (DCA). The results were validated using 1000 bootstraps resampling internal validation and testing set. Among 32 potential predictors, the final prediction model included 6 variables: postoperative pain, postoperative infection, dexmedetomidine use, propofol use, duration of mechanical ventilation, and Prognostic Nutritional Index (PNI). The model showed a good discrimination with an AUC of 0.919 (95% CI: 0.885- 0.953) in the training set, and adjusted to 0.911 (95% CI: 0.878- 0.944) and 0.871 (95% CI: 0.802- 0.940) in the internal validation and the testing set, respectively. ROC curves, calibration curves, DCA curves, C-statistic, Brier score and Hosmer-Lemeshow goodness-of-fit test showed excellent model performance. This study successfully established and validated the first POD prediction model for patients with esophageal cancer resection. It could accurately predict the occurrence of POD and effectively identify the high-risk patients, which is of great significance for improving the risk stratification of the population and for implementing targeted prevention intervention measures.

Keywords: Delirium; Esophageal cancer; Machine learning; POD; Prediction model.

Fig. 1

Analysis of postoperative delirium in patients undergoing esophageal cancer resection. (A) Distribution of postoperative delirium types in patients undergoing esophageal cancer resection. The horizontal axis indicates the type of delirium, and the vertical axis indicates the number of patients. (B) Distribution of time of first appearance of postoperative delirium. The horizontal axis indicates the time (in days) when delirium first appeared, and the vertical axis indicates the number of people. (C) Distribution of duration of postoperative delirium. The horizontal axis indicates the duration of postoperative delirium (days) and the vertical axis indicates the number of people.

Fig. 2

Nomogram for postoperative delirium in patients undergoing esophageal cancer resection. Method of using the nomogram: Find the position of each variable on the corresponding axis, draw a line on the point axis to represent the number of points, add the number of points for all variables, and then draw a line from the total point axis to determine the probability of postoperative delirium.

Fig. 3

Calibration curves for risk prediction model for postoperative delirium in patients with esophageal cancer resection. (A) Training set; (B) Testing set. The grey dashed line represents the perfect prediction of the ideal model; the predictive ability of the model exactly matches the actual risk of POD. The blue solid line represents the actual performance of the column chart model in predicting POD, and the red solid line represents the actual performance of the column chart model in predicting POD after correction by the Bootstrap method, and the closer the two are to the diagonal grey dashed line represents the better prediction.

Fig. 4

Decision curves analysis of the nomogram for the risk of postoperative delirium in patients with esophageal cancer resection. (A) Training set; (B) Testing set. X-axis indicates threshold probability. The grey line indicates the net benefit of predicting all PODs using the gold standard. The black line indicates that no net benefit of any POD is detected. The blue line indicates the column-line graphical model.

Fig. 5

Clinical impact curves analysis of the nomogram for the risk of postoperative delirium in patients with esophageal cancer resection. (A) Training set. (B) Testing set. The ordinate represents the people at high risk, the first abscissa represents the threshold probability, and the second abscissa represents the loss-benefit ratio. When the horizontal coordinate is > 0.4, the closer the two lines are, the better the model effect is.

Fig. 6

Receiver operating curves (ROC) analysis of the nomogram predictive model. (A) Training set. (B) Testing set. The horizontal X-axis is 1-specificity, also known as false positive rate, and the closer the X-axis value is to zero, the higher the accuracy. The value corresponding to the Y-axis of the vertical coordinate is sensitivity, also known as the true positive rate (sensitivity), and the greater the value of the Y-axis, the better the accuracy. The Area Under the Curve (AUC) ranges from 0.5 to 1. The closer the value of AUC is to 1, the better the diagnostic effect of this variable in predicting the outcome.

Fig. 7

Study population and design flowchart.