Development and Validation of a Nomogram to Predict Overall Survival in Stage I-III Colorectal Cancer Patients after Radical Resection with Normal Preoperative Serum Carcinoembryonic Antigen

Abstract

We aimed to develop a clinical predictive model for predicting the overall survival (OS) in stage I-III CRC patients after radical resection with normal preoperative CEA. This study included 1082 consecutive patients. They were further divided into a training set (70%) and a validation set (30%). The selection of variables for the model was informed by the Akaike information criterion. After that, the clinical predictive model was constructed, evaluated, and validated. The net reclassification index (NRI) and integrated discrimination improvement (IDI) were employed to compare the models. Age, histologic type, pT stage, pN stage, carbohydrate antigen 242 (CA242), and carbohydrate antigen 125 (CA125) were selected to establish a clinical prediction model for OS. The concordance index (C-index) (0.748 for the training set and 0.702 for the validation set) indicated that the nomogram had good discrimination ability. The decision curve analysis highlighted that the model has superior efficiency in clinical decision-making. NRI and IDI showed that the established nomogram markedly outperformed the TNM stage. The new clinical prediction model was notably superior to the AJCC 8th TNM stage, and it can be used to accurately assess the OS of stage I-III CRC patients undergoing radical resection with normal preoperative CEA.

Keywords: CA125; CA242; colorectal cancer; nomogram; overall survival.

Figure 1

Study flowchart.

Figure 2

A constructed nomogram for survival prediction of colorectal patients undergoing radical surgery with normal preoperative carcinoembryonic antigen (CEA). For example, a CRC patient after radical resection with normal preoperative CEA was 60 years old at the time of surgery, had a histological type of adenocarcinoma grade III, with pT and pN stages of T3 and N2, respectively, and was negative and positive for CA242 and CA125, respectively. To use the nomogram, each variable’s position on its axis is identified, and lines are drawn from these positions to the points axis. The intersection points determine the number of points attributed to each variable. By summing up the points (230) obtained from all variables, the nomogram provides an estimation of the patient’s likelihood of 3−year and 5−year overall survival (OS). The 3−year and 5−year overall survival rates for this patient were approximately 30% and 12%, respectively.

Figure 3

The Receiver Operating Characteristic (ROC) curves for the survival prediction of colorectal cancer patients undergoing radical surgery with normal preoperative carcinoembryonic antigen (CEA) in the training (A) and the validation (B) sets.

Figure 4

The calibration curves of the nomogram for the survival prediction of colorectal cancer patients undergoing radical surgery with normal preoperative carcinoembryonic antigen (CEA) in the training (A) and validation (B) sets. The x-axis represents the model’s predicted probability or score, usually in the range of 0 to 1. This is the model’s estimate of the probability of an event occurring. The y-axis represents the actual observed rate of event occurrence (or survival), also in the range 0 to 1. This is the proportion of events that occur in real data. The two curves in red and green represent the calibration curves of the model. The calibration curve shows the relationship between the predicted probabilities of the model and the actual observations. The dotted line represents the perfect calibration line of the theory, which is the 45-degree diagonal line. If the calibration curve coincides with this line, it means that the model’s predictions are perfectly accurate. The error lines (yellow and blue) shown on the calibration curve are used to represent the uncertainty of the calibration curve. These error lines are usually representations of confidence intervals.

Figure 5

Comparison of decision curve analysis (DCA) results between the nomogram and the AJCC 8th TNM stage for the survival prediction of colorectal cancer patients undergoing radical surgery with normal preoperative carcinoembryonic antigen (CEA). 3−year survival clinical net benefit in the training (A) and validation (C) sets, 5−year survival clinical net benefits in training (B) and validation (D) sets.

Figure 6

Kaplan–Meier survival curves of colorectal cancer patients undergoing radical surgery with normal preoperative carcinoembryonic antigen (CEA) in the training (A) and validation (B) sets (the cutoff was stratified by the median of the total score of the training set).