Survival Prediction Model for Esophageal Squamous Cell Carcinoma Patients with Lymph Node Metastasis who had Surgery: Based on SEER Database

Data Source

The data in the present study were from the SEER database (https://seer.cancer.gov), which collects cancer incidence data from US population-based cancer registries. The SEER registries collect data on patient demographics, pathological stage, treatments, follow-up time as well as vital status, etc. The specific dataset with radiotherapy and chemotherapy - “SEER 18 Regs Custom Data (with additional treatment fields), Nov 2018 Sub,” was entitled to use in the present study. As a public dataset with identified data, there was no need to access local ethics committee approval for applying the data.

Patient Selection

Patients diagnosed as EC were firstly selected in the SEER database, by using the SEER*Stat 8.35. Then, the data were sifted by the followed included criteria: [1] pathological diagnosis of ESCC according to positive histology: the Code of International Classification of Diseases, 3rd edition (ICD-O-3) defined as 8070-8075; [2] American Joint Committee on Cancer (AJCC) (7th Edition) N stage was N1-N4; [3] received surgery; [4] complete data of vital status and survival months.

The qualified data were randomly stratified into a training cohort and a validation cohort in the ratio of 7:3 by the R function “createDataPartition” to ensure that data were distributed without a bias between the two cohorts.

Variable Selection

The complete data were selected in the present study including demographic data which were age at diagnosis; race; gender; marital status, stage groups which were primary site; grade; summary; T stage; M stage, clinical treatments which were surgery status; radiotherapy status; chemotherapy status, and other clinical factors, as regional nodes examined, regional nodes positive, and number of malignant tumors. Stage groups were determined by the American Joint Committee on Cancer (AJCC) staging system. Data on duration of follow-up, vital status, and survival months were also extracted from the database to assess overall survival (OS) which represented the primary endpoints of the study. For the analysis of OS, the definition of vital status was as follow: till the end of follow-up time, death was considered as complete data (defined as “0”) and alive as censored data (defined as “1”) and the duration of survival time was in months.

Statistical Analysis

The class variables were described as frequencies and percentages. The student’s t-test was performed to compare continuous variables, and a chi-square test was used for categorical variables. Univariate and multivariate Cox regression analysis was performed to identify the factors related to prognosis. The significant variables of the univariate analysis were screened out to construct the multivariate Cox regression model. Integrated the variables from Cox multivariate regression analyses into the nomogram to predict 3- and 5-year OS probabilities.

Then the established model was validated by several indicators. The discrimination capability of the model was evaluated by the concordance index (C-index) and the receiver operating characteristic (ROC) curves with area under the curve (AUC) which ranged from 0.5 to 1.0. The nomogram would provide a reasonable estimation if the C-index and AUC were over 0.7. The net reclassification index (NRI) quantifies the ability of a new model to reclassify subjects as compared to an old model, while the integrated discrimination improvement index (IDI) can be used to assess the whole predicted improvement of the model. NRI and IDI were both employed to compare the discriminative usage between the present nomogram and the 7th AJCC staging system (The period of the dataset applied in the present study was till 2016 when the 8th AJCC staging system was not released. The present data was staged by the 7th system, so it was the suitable system to be compared to. Besides that, 7th and 8th AJCC staging existing similar power in predicting the disease prognosis for ESCC) [10]. The probability of accurate prediction for the new model is higher than the old model for prediction if NRI>0 or IDI>0. The clinical validity of the nomogram was evaluated by Decision-curve analysis (DCA) by quantifying the net benefit compared to the 7th AJCC staging system. Calibration plots were applied to assess the consistency between the observed probabilities and the predicted outcomes. All statistical data were analyzed by R version 4.1.1 (www.r-project.org), with a

2-sided significance level set at p <.05.

Patient Characteristics

A total of 468 ESCC patients with LNM who had surgery from 2004 to 2015 were identified (Figure 1). The mean age of the targeted patients was 62.25±8.89 years. White (73.50%), male (58.97%), and married status (65.17%) were majority of patients. The most common Grade was Grade II (50.43%), the summary stage was regional (80.13%), the T stage was the T2 stage (59.19%), and the M stage the was M0 stage (89.74%). For clinical treatments, over half of the patients had radical resection (73.93%); 81.20% of patients received radiotherapy after surgery, while 93.80% had chemotherapy. The mean duration of follow-up was 29.72±6.10 months. The demographic and clinical characteristics of these patients are summarized in Table 1.

Surg Prime Site S1=Local tumor destruction; S2=Local tumor excision; S3=Less than total resection; S4=Total resection; S5=Radical resection Regional nodes examined: 95-98= Regional lymph node was removed, but the number of nodes is unknown/not stated Regional nodes positive: 95-97= Positive nodes are documented, but number is unspecified; 98= No nodes were examined. Table1: Demographic and clinical characteristics of ESCC patients with LNM who had surgery.

Click to enlarge

Identifying Variables for the Nomogram

16 variables were analyzed by the Cox univariate regression analysis (Table 2). Ten of them were integrated into the Cox multivariate analysis and 7 factors served as independent prognostic factors associated with OS (p < 0.05) $$
\begin{array}{l l l l l l l} \text {=} & \text {U nivariate analysis} & \text {p - value} & \text {Multivariate analysis} & \text {p - value} \\ & \mathrm {H R} & 95 \% \mathrm {C I} & \mathrm {H R} & 95 \% \mathrm {C I} \\ \mathrm {A g e} & & & & & \\ 2 5 - 4 4 \mathrm {years} & 1 & - & - & 1 & - \\ 4 5 - 6 4 \mathrm {years} & 0. 6 6 9 & 0. 2 1 0 - 2. 1 3 2 & 0. 4 9 6 &

16 variables were analyzed by the Cox univariate regression analysis (Table 2). Ten of them were integrated into the Cox multivariate analysis and 7 factors served as independent prognostic factors associated with OS (p < 0.05) $$ \begin{array}{l l l l l l l} \text {=} & \text {U nivariate analysis} & \text {p - value} & \text {Multivariate analysis} & \text {p - value} \\ & \mathrm {H R} & 95 \% \mathrm {C I} & \mathrm {H R} & 95 \% \mathrm {C I} \\ \mathrm {A g e} & & & & & \\ 2 5 - 4 4 \mathrm {years} & 1 & - & - & 1 & - \\ 4 5 - 6 4 \mathrm {years} & 0. 6 6 9 & 0. 2 1 0 - 2. 1 3 2 & 0. 4 9 6 & 1. 0 9 8 & 0. 3 1 7 - 3. 8 0 1 \\ 6 5 - 8 4 \mathrm {years} & 0. 9 6 8 & 0. 3 0 2 - 3. 1 0 3 & 0. 9 5 7 & 1. 7 7 3 & 0. 5 0 2 - 6. 2 5 6 \\ \geq 8 5 \mathrm {years} & 1. 4 9 9 & 1. 1 1 5 - 2. 0 6 4 & 0. 0 4 5 & 2. 0 1 8 & 1. 6 0 3 - 2. 7 0 2 \\ \mathrm {R a c e} & & & & & \\ \end{array} $$

American Indian/Asian/Pacific Islander

1 - - 1 - - Black 0.924 0.490-1.742 0.808 0.589 0.287-1.208 0.148 White 0.646 0.392-0.906 0.046 0.455 0.256-0.806 0.007 Gender Female 1 - - 1 - - (Table 2). These factors were surgery prime site, T status, M status, radiation sequence with surgery, age, race, and gender. The Cox multivariate analysis indicated that ≥85 years, male, local tumor destruction, and radiation after surgery were related to the deterioration of OS. For white, the T2 stage, and M0 stage acted as the protective factor for OS.

Surg Prime Site: S1=Local tumor destruction; S2=Local tumor excision; S3=Less than total resection; S4=Total resection; S5=Radical resection Regional nodes examined: 95-98= Regional lymph node was removed, but the number of nodes is unknown/not stated Regional nodes positive: 95-97= Positive nodes are documented, but number is unspecified; 98= No nodes were examine. Table 2: Univariate and multivariate Cox analyses for screening factors related to overall survival of ESCC patients with LNM who had surgery.

Establishment of the Nomogram

The nomogram for predicting 3 and 5-year OS was established by incorporating the seven factors from Cox multivariate analysis. Among these factors, the surgery prime site was the most contributed factor for forecasting OS probability, then the T stage, M stage, and radiation sequence with surgery (Figure 2).

Click to enlarge

Validation of the Nomogram

C-index and ROC-related AUC were employed to evaluate the discrimination ability of the nomogram. The C‐index was 0.709 (95% CI = 0.661‐0.760) in the training cohort and 0.775 (95% CI = 0.708‐0.860) in the validation cohort. Both figures were significantly higher than those of the 7th AJCC stage (p < 0.001), which achieved a score of 0.607 (95% CI = 0.544- 0.670) and 0.683 (95% CI = 0.597-0.769), respectively (Table 3). The AUCs > 0.7 for the prediction of 3- and 5- year OS in both the training (0.712 and 0.716) and validation cohorts (0.725 vs. 0.857), indicating favorable discrimination usage by the nomogram compared to that of the 7th AJCC staging system (Figure 3).

Click to enlarge

Figure 3: ROC Curves of the nomogram and 7th AJCC staging system to predict 3-year and 5-year OS for ESCC patients with LNM who had surgery in the training cohort (A-B) and the validation cohort (C-D). The red curve was for the nomogram and the blue one was for the AJCC staging system. The larger the AUC, the better discrimination of the model would be.

NRI and IDI were applied to evaluate the accuracy of the nomogram, the NRI for the 3‐ and 5‐year OS probabilities were 0.307 (95% CI: 0.054-0.560) and 0.291 (95% CI: 0.078- 0.526) in the training group, and the IDI for 3‐ and 5‐year OS probabilities were 0.079 (95% CI: 0.017-0.222) and 0.129 (95% CI: 0.046-0.242) (Table 3). These results were validated in the validation cohort (Table 3). These results indicated that the nomogram predicted 3‐ and 5‐year prognosis with superior accuracy than the 7th AJCC staging system.

DCA was performed to compare the clinical usage of the nomogram to the 7th AJCC staging system. In both the training cohort and the validation cohort, compared to the

7th AJCC stage system the nomogram added more net clinical benefits, as shown in Supplementary Figure 4.

Click to enlarge

The red dashed lines represent the DCA of the nomogram, and the mint green dashed lines represent the AJCC staging system. The green solid line assumes that all patients were alive, while the purple dashed line assumes that all patients were deceased.

Click to enlarge

Calibration plots of the nomogram indicated that the uniformity was favorable between the predicted and observed survival probability for 3-year and 5-year in the training and validation cohorts (Figure 5).