To the best of our knowledge, the current study represents a novel research to successfully build a prognostic nomogram model and risk-stratification system incorporating multiple hematologic inflammatory indices for patients with ESCC receiving definitive IMRT based on a large cohort. In this study, cTNM, GTVp, CT, NLR, and PLR were first identified as independent prognostic indicators by multivariate Cox regression analysis in the training cohort. A nomogram was developed based on these five independent factors to predict OS, and verified in the validation cohort as well. The nomogram model was found to be better than the AJCC staging in predicting OS and clinical survival benefits. Lastly, depending on the nomogram total scores, the entire cohort was divided into low-, intermediate-, and high-risk subgroups. This risk stratification was found to be superior than AJCC clinical staging in predicting OS. Collectively, these results suggest that the nomogram model is a useful for risk stratification and for predicting survival of patients with ESCC treated with radical IMRT.
In this study, several hematological inflammatory indices including NLR and PLR were found to independently impact the prognosis besides other common factors, such as cTNM, GTVp, and CT. Of note, several studies have demonstrated a close correlation of NLR and PLR with prognosis of patients with several other kinds of malignancies including gastric cancer, non-small cell lung cancer, and colorectal carcinoma [19-21]. Moreover, some studies have also demonstrated a significant association of pre-treatment elevated PLR and NLR with poorer prognosis and deeper tumor invasive depth in patients with ESCC [22, 23]. This phenomenon may be attributable to the potential involvement of systemic inflammatory response in the migration, invasion, and metastasis of malignant cells in various tumors, including ESCC [24]. Therefore, inclusion of NLR and PLR in the nomogram model may help improve the prognostic assessment of patients with ESCC receiving definitive IMRT.
We evaluated the predictive performance of the nomogram model using a variety of indices, including accuracy, discrimination ability, and clinical validity. Calibration, which is generally used to evaluate the accuracy of nomogram, refers to the agreement between the observed and estimated probabilities of the occurrence of an event or outcome [25]. In this study, the calibration curves between the OS predicted by nomogram and the actual OS showed a good agreement in both the training and validation cohorts. In addition, the C-index and the ROC curve are most commonly used to assess the discrimination of the nomogram model [26]. In the training cohort, the C-index and AUC values for 5-year OS were 0.627 and 0.706, respectively, showing good differentiation and predictive ability for OS. Moreover, the C-index and AUC values for 5-year OS were 0.629 and 0.719 in the validation cohort, respectively, confirming the reproducibility and stability of the nomogram model. Hence, the constructed nomogram model presented a good predictive performance.
To further compare the predictive ability of the nomogram model with that of the AJCC staging, 4 evaluation indicators including AUC, C-index, NRI, and IDI were employed. AUC value and C-index were used as the basic reference indicators to estimate the improvement of predictive performance of the nomogram model as compared to the AJCC staging [26]. In our study, the AUC values and C-index of the nomogram in the training and validation cohorts were superior to those of the AJCC staging. In recent years, NRI and IDI have been strongly recommended for evaluating and comparing the distinctive ability between the two prediction models [14]. NRI was originally applied for quantitative evaluation of the improvement in classification performance of the new model over the original model, while the IDI was employed to assess the changes in risk differentials [26]. In our study, the NRI and IDI of the AJCC staging were significantly inferior to those of the nomogram, suggesting better predictive ability of the nomogram model for OS.
DCA was developed to determine whether use of predictive models to inform clinical decision-making does more harm than good, and to further evaluate the clinical applicability of predictive models [27]. In our study, the nomogram offered a higher net benefit than the AJCC staging at any given threshold, indicating a better clinical application value of the constructed nomogram. It is noteworthy that the clinical net benefit of GTVp was almost the same as that of the clinical AJCC staging. GTVp has been shown to be an independent predictor of survival in patients with ESCC treated with definitive radiotherapy [16, 28]. Larger GTVp always means greater tumor load, and a greater proportion of tumor radioresistant hypoxic cells and clonogenic cells, which causes poor survival [28]. This explains the markedly superior net benefit of the nomogram incorporating hematological inflammatory indices, GTVp, cTNM, and CT compared to the AJCC staging.
Finally, risk stratification system was formed depending on the total nomogram scores using the X-tile software. In the study population, 28.6%, 48.2%, and 23.2% of patients were categorized into low-, moderate-, and high-risk subgroups, whereas the proportion of patients in AJCC stages II, III, IVa, and IVb were 14.6%, 21.7%, 44.1%, and 19.6%, respectively. This indicated a more balanced patient distribution among the three risk subgroups as compared to that among the clinical AJCC stage. Moreover, this risk stratification showed a significantly higher C-index than the clinical AJCC stage, suggesting that the nomogram had a better discrimination ability for risk stratification. Patients in the high-risk group require more intensive therapies: (1) adjuvant chemotherapy; (2) targeted drugs [29]; or (3) immunotherapy. In particular, immunotherapy has developed rapidly in recent years and has been actively explored and applied to patients with ESCC [30, 31]. Studies have demonstrated that the combination of immunotherapy and radiotherapy can synergistically promote anti-tumor activity in vitro, thus effectively controlling local lesions and distant micrometastases [32, 33]. Therefore, radiotherapy plus immunotherapy may improve the treatment efficacy for high-risk ESCC patients. In addition, for patients in the low-risk group, it may be appropriate to reduce the radiation dose or chemotherapy cycles to decrease not only the side effects of radiotherapy combined with adjuvant chemotherapy, but also the treatment cost.
Nevertheless, several limitations of our study should be considered. First, we did not analyze all inflammatory parameters, as some inflammatory mediators such as procalcitonin, C-reactive protein, interleukin-1, interleukin-6, and tumor necrosis factor were not routinely examined at our institution. In addition, this was a retrospective, single-center study and our results may have been influenced by confounders. Further prospective, multicenter studies are required to verify the precision of the nomogram model. Finally, we did not evaluate the dynamic changes in hematological indicators before and after treatment. Such an analysis may help improve the prognostic capability of the nomogram and improve the risk stratification of patients.