Baseline Clinical and Characteristics
In the present study, 346 eligible patients were analyzed in the training cohort, and 173 patients were included in the validation cohort. The median follow-up was 51.4 months (interquartile range (IQR):42.1 - 67.0) in the training cohort and 50.4 months (IQR: 41.9 - 66.0 months) in the validation cohort. In the training cohort, the 1-, 3-, and 5-year OS rates were as follows: 97.4%, 83.8%, and 48.3%. In the validation cohort, the 1-, 3-, and 5-year OS rates were the following: 94.2%, 84.4%, and 42.8%.
The optimal cut-off value for each continuous variable as follows: age (60 years), smoking index (20.0), BMI (26.33 kg/m2), WBC (4.3 × 109/L), Neutrophils (7.0 × 109/L), Lymphocyte (1.41 × 109/L), Monocyte (0.4 × 109/L), Platelet (293.0 × 109/L), HGB (130.0 g/L), NLR (3.91), dNLR (2.46), LMR (3.4), PLR (208.89), SII (1141.96), TP (77.2 g/L), ALB (42.4 g/L), GLOB (33.1 g/L), AGR (1.36), CRP (5.47 mg/L), CAR (0.16), APOA (1.28 g/L), APOB (1.03 g/L), ABR (0.96), LDH (167.5 U/L), HDL (1.16 U/L), Cys-C (0.94 mg/L), ALI (262.33), and PNI (47.35). Patients’ clinical characteristics and blood-biomarkers for the patients were listed in Table 1. There was no significant difference in the distribution of clinical characteristics and blood-biomarkers between training cohort and validation cohort.
Construction of the novel prognostic model
In order to find the prognostic variables in the training cohort, we used a LASSO-Cox regression analysis model. Figure 1A showed the change in trajectory of each prognostic variable was analyzed. Moreover, we plotted the partial likelihood deviance versus log (λ) in Figure 1B, where λ was the tuning parameter. The value of λ was 0.03987 was chosen by 10-fold cross-validation via the 1-SE criteria. So, we obtained 13 variables with nonzero coefficients at the value λ chosen by cross-validation. These prognostic variables included age, BMI, HGB, PLT, LMR, CRP, CAR, GLOB, AGR, LDH, Cys-C, ALI, and PNI. The coefficients of each prognostic variable were presented in Figure 1C. Then the prognostic model risk score of each patient was computed according to the summation of 13 variables multiplied coefficient from the LASSO regression generated: The prognostic model risk score = -0.680 + (0.569 × age) - (0.280 × BMI + (0.101 ×HGB) - (0.554 × PLT) + (0.197 ×LMR) - (0.199 ×CRP) + (0.186 ×CAR) + (1.248 ×GLOB) - (0.137 ×AGR) - (0.194 ×LDH) + (1.248 × Cys-C) - (0.137 ×ALI) - (0.194 ×PNI). Where each variable was valued as 0 or 1; a value of 0 was assigned when the variable was less than or equal to the corresponding cut-off value, and a value of 1 otherwise.
Comparison of predictive accuracy between the novel prognostic model, TNM staging, clinical treatment, and EBV DNA
As shown in Table 2, in training cohort, the C-index of the prognostic model was 0.786 (95% confidence interval (CI): 0.728-0.844), which was higher than the C-indices of the TNM staging, clinical treatment, and EBV DNA, with values of 0.740 (95% CI: 0.690-0.790), 0.554 (95% CI: 0.521-0.586), and 0.691 (95% CI: 0.623-0.758), respectively. The C-index for the prognostic model was statistically significantly higher than the C-index by the clinical treatment (P < 0.001), and EBV DNA (P = 0.013). In the validation cohort, the C-index of prognostic model was both higher than that of TNM staging and clinical treatment, but it was a little lower than that of EBV DNA. Subsequently, we compared the area under ROC curve (AUC) between the novel prognostic model, TNM staging, clinical treatment, and EBV DNA using tdROC. In general, the AUC of novel prognostic model was higher than others both in the training cohort (Figure 2A) and validation cohort (Figure 2B). Finally, the DCA displayed the prognostic model had a better overall net benefit than TNM staging, clinical treatment and EBV DNA across a wide range of reasonable threshold probabilities in training cohort (Figure 3A) and validation cohort (Figure 3B). These results indicated that the novel prognostic model displayed better accuracy in predicting OS compared with the TNM staging, clinical treatment and EBV DNA.
Building and validating a predictive nomogram
The prognostic model risk score, TNM staging, clinical treatment, and EBV DNA were integrated into nomograms to predict the 1-, 3-, and 5-year OS in the training cohort (Figure 4). Each variable was assigned a corresponding point value based on its contribution to the model. The point values for all the predictor variables are summed to arrive at the "Total Points" axis, and then a line is drawn vertically down from total points to predict the patient’s probability of OS at 1-, 3-, and 5-year. Finally, A calibration plot was used to visualize the performance of the nomogram. Nomogram predicted and actual observed outcome at 1-, 3-, and 5-year OS were plotted on the x-axis and y-axis, respectively. The 45° line represented the best prediction, the solid dark red line represented the performance of the nomograms. The calibration curve showed that the 1-, 3-, and 5-year OS predicted by the nomograms were consistent with actual observations (Figures 5), indicating that the nomograms did well performance. The nomograms and calibration curve in the validation cohort were showed in the supplementary Figure 1 and supplementary Figure 2, respectively.
Survival analyses of NPC patients according to prognostic model risk score
The optimal cut-off value of prognostic model risk score for predicting survival was determined to be -1.423 by R package “survminer” (Figure 6A). According to the cutoff value, we classified patients into two different subgroups, of which low-risk group with risk score ≤ -1.423, and high-risk group with risk score > -1.423. The distribution of the prognostic model risk score in training and validation cohort were showed in Figure 6B and Figure 6C, respectively.
In the training cohort, for the high-risk group, the median OS was 44.4 months (IQR: 24.7 – 66.1) The 1-, 3- and 5-year probabilities of OS were 95.4%, 63.2%, and 33.3%, respectively. For the low-risk group, the median OS was 61.2 months (IQR: 44.6 – 67.8). The 1-, 3- and 5-year probabilities of OS were 98.1%, 90.7% and 53.3%, respectively. In the validation cohort, low-risk group also had higher survival probabilities than high-risk group at 1-, 3-, and 5-year, respectively (Table 3). Kaplan–Meier curves were compared to assess the differences in survival between low-risk and high-risk groups. The results showed statistically significant better OS for low-risk group versus high-risk group both in training cohort and validation cohort (p <0.05; Figure 7).