In this study, a total of 64,357 patients were initially enrolled. After strict screening, about 9,653 patients met the criteria. The patients were separated into three groups according to the RT status (No RT, Preop RT and Postop RT groups). The specific patient information was as follows: 5421 patients did not receive RT, 1002 patients received preoperative RT, and 3230 patients received postoperative RT, accounting for 56.2%, 10.4%, and 33.5%, respectively (see Table 1 for details).
Survival And Lnr Categories
The definition of LNR was the positive nodes number divided by the total number of lymph nodes detected. The ROC curves were generated for the comparison of the discrimination for the LNR and positive lymph nodes number (no. of positive lymph nodes). As shown in Fig. 2a-c, the AUC values of LNR used to present the 1-, 3- and 5-year survival rates of patients were 0.699 (95% CI: 0.690 ~ 0.709), 0.705 (95% CI: 0.696 ~ 0.714), and 0.702 (95% CI: 0.693 ~ 0.711), respectively; the AUC values of the number of positive lymph node metastases used to determine the 1-, 3-, and 5-year survival rates of the patients was 0.670 (95% CI: 0.661 ~ 0.680), 0.683 (95% CI: 0.674 ~ 0.692), and 0.684 (95% CI: 0.675 ~ 0.0.694), respectively; and the survival predictions of the two were statistically significant (P < 0.001). Using LNR to predict the GC patient prognosis was more accurate, objective and effective. Therefore, we grouped the selection of the LNR through the cut-off point of the ROC curve.
As shown in Fig. 2d, the best cut-off point of the LNR was 0.1714, with a sensitivity of 63.1% and a specificity of 67.6% for predicting the prognosis of GC patients. The AUC was 0.696 (95% CI: 0.687 ~ 0.705, P < 0.001). We identified the patients with negative lymph nodes as a group (LNR1: LNR = 0%), the rest patients with positive lymph nodes were classified into two groups: LNR2: 0% < LNR ≤ 17.14%; and LNR3: LNR > 17.14%.
Kaplan-Meier survival analysis was performed according to LNR grouping (Fig. 3a). The results showed a significant group difference, in which LNR1 vs LNR2, P < 0.001; LNR2 vs LNR3, P < 0.001; and LNR3 vs LNR1, P < 0.001.
Survival analysis of the No RT, Preop RT and Postop RT groups
The survival curves of the No RT, Preop RT and Postop RT groups are presented in Fig. 3b. Significant differences were seen in the survival of all patients between the three groups: No RT vs Preop RT, P = 0.043; No RT vs Postop RT, P < 0.001; and Preop RT vs Postop RT, P < 0.001. The median survival time were 31months, 34 months, and 47 months in No RT, Preop RT and Postop RT groups, respectively. And 1-, 3-, and 5-year CSS rates were 71.9%, 46.8%, and 38.9% in No RT group, 84.3%, 48.9%, and 38.3% in Preop RT group, 84.9%, 55.0%, and 46.0% in Postop RT group, respectively.
Among the patients with negative regional lymph nodes, Postop RT group showed no significant survival benefit over No RT group (P = 0.057), but Postop RT and No RT groups showed significant survival benefits over Preop RT group, with P values of < 0.001 and 0.001, respectively (Fig. 3c).
For the patients with positive regional lymph nodes, both Preop RT (P < 0.001) and Postop RT (P < 0.001) groups showed significant survival benefits over No RT group, and Postop RT group showed obvious survival benefits over Preop RT group (P < 0.001) (Fig. 3d).
Based on AJCC stage, the subgroup analyzed Postop RT group, No RT group and Preop RT group. The results showed that Postop RT group had a significant survival benefits than No RT group (P = 0.005), and No RT group showed significant survival benefits over Preop RT group (P < 0.001) in AJCC I stage (Fig. 4a). While in AJCC II stage, Postop RT group showed significant survival benefits over both No RT and Preop RT groups (P < 0.001), and no survival difference was seen between Preop RT group and No RT group (P = 0.106) (Fig. 4b). Both Preop RT and Postop RT groups showed significant survival benefits over No RT group (P < 0.05), and no significant survival benefit was seen between Postop RT group and Preop RT group in both AJCC III stage and IV stage, with p values of 0.240 and 0.932, respectively (Fig. 4c, d).
Comparison of Pathological Features between Preop RT group and Postop RT and No RT combined groups
Table 2 shows different pathological features in lymph node and primary tumor statuses between patients in the Preop RT group and those in the Postop RT and No RT combined groups. Lymph node pathology included lymph node metastases number, detected lymph nodes number and LNR, and the Kruskal-Wallis test was used to compare whether the two groups were different. The number of positive lymph nodes (mean rank, 3344.23 vs 4998.74 [P < 0.001]), lymph nodes removed (mean rank. 4383.52 vs 4878.37 [P < 0.001]) and LNR (mean rank. 3423.52 vs 4989.56 [P < 0.001]) in the Preop RT group were significantly less when compared with the Postop RT and No RT combined groups. For the T stage of the primary tumor status, the chi-square test was used to compare the differences between the Preop RT group and the Postop RT and No RT combined groups. The distributions according to AJCC T classification were as follows: 4.5% vs 5.6% for T1 (P = 0.142), 58.4% vs 59.4% for T2 (P = 0.520), 32.3% vs 26.4% for T3 (P < 0.001), and 4.8% vs 8.6% for T4 (P < 0.001). The distribution of T1 and T2 did not have a significant group difference. However, a significant group difference was seen in the distribution of T3 and T4. T3 was more widely distributed in the Preop RT group, and T4 was more widely distributed in the Postop RT and No RT combined group. This result indicates that preoperative RT did not affect the primary tumor, but the microscopic lymph node disease.
Table 2
Comparison of Pathological Features between Preop RT group and Postop RT and No RT combined groups
Characteristic | Preop RT | Postop RT and No RT | P-value |
No. of positive LNs: mean rank | 3344.23 | 4998.74 | < 0.001 a |
No. of LNs removed: mean rank | 4383.52 | 4878.37 | < 0.001 a |
LNR: mean rank | 3423.52 | 4989.56 | < 0.001 a |
Tumor classification, % | | | |
T1 | 45 (4.5%) | 485 (5.6%) | 0.142 b |
T2 | 585 (58.4%) | 5142 (59.4%) | 0.520 b |
T3 | 324 (32.3%) | 2281 (26.4%) | < 0.001 b |
T4 | 48 (4.8%) | 743 (8.6%) | < 0.001 b |
Preop, preoperative; RT, radiotherapy; Postop, postoperative; LNR, lymph node metastasis |
a. Kruskal-Wallis test. |
b. Chi-square test. |
The Analyzation Of Cox Regression Model
In the primary GC cohort, the median survival time was 36 months, and the 1-, 3-, and 5-year CSS rates were 77.6%, 49.8%, and 41.3%, respectively. The hazard ratios for CSS are listed in Table 3 obtained by Cox regression model according to all variables. All variables except for sex (P = 0.280) were revealed as prognostic factors in Cox regression models (i.e., age, race, tumor differentiation, tumor histology type, primary tumor site, T stage, LNR, RT status, and chemotherapy information).
Table 3
Multivariate analyses for CSS using the Cox regression model
Variable | All patients | Lymph node-positive patients |
HR (95% CI) | P-value | HR (95%CI) | P-value |
Age | | | | |
< 60 years | 1 | | 1 | |
≥ 60 years | 1.275 (1.200- 1.355) | < 0.001 | 1.254 (1.174–1.339) | < 0.001 |
Sex | | | | |
Male | 1 | | 1 | |
Female | 0.969 (0.915–1.026) | 0.280 | 0.965 (0.906–1.028) | 0.268 |
Race | | | | |
White | 1 | | 1 | |
Black | 1.058 (0.974–1.148) | 0.182 | 1.091 (0.998–1.193) | 0.055 |
Others | 0.795 (0.739–0.856) | < 0.001 | 0.821 (0.758–0.889) | < 0.001 |
Grade | | | | |
Well | 1 | | 1 | |
Moderate | 1.061 (0.876–1.285) | 0.545 | 1.126 (0.892–1.420) | 0.318 |
Poor | 1.307 (1.083–1.577) | 0.005 | 1.449 (1.154–1.820) | 0.001 |
Undifferentiated | 1.476 (1.149–1.897) | 0.002 | 1.663 (1.245–2.223) | 0.001 |
Histology | | | | |
Adenocarcinoma | 1 | | 1 | |
Mucinous adenocarcinoma | 0.979 (0.839–1.141) | 0.782 | 1.042 (0.882–1.231) | 0.628 |
Signet ring cell carcinoma | 1.175 (1.099–1.257) | < 0.001 | 1.205 (1.121–1.295) | < 0.001 |
Primary site | | | | |
Upper third | 1 | | 1 | |
Mid and low third | 0.770 (0.719–0.824) | < 0.001 | 0.826 (0.767–0.890) | < 0.001 |
Overlapping stomach | 0.861 (0.771–0.963) | 0.009 | 0.963 (0.855–1.084) | 0.531 |
Stomach, NOS | 0.852 (0.762–0.952) | 0.005 | 0.927 (0.821–1.047) | 0.223 |
T stage | | | | |
T1 | 1 | | 1 | |
T2 | 1.885 (1.623–2.190) | < 0.001 | 2.255 (1.941–2.619) | < 0.001 |
T3 | 2.733 (2.344–3.185) | < 0.001 | 3.354 (2.877–3.909) | < 0.001 |
T4 | 3.525 (2.982–4.166) | < 0.001 | 4.098 (3.458–4.858) | < 0.001 |
Lymph node ratio | | | | |
0% | 1 | | - | |
0%<LNR ≤ 17.14% | 1.687 (1.542–1.847) | < 0.001 | - | - |
LNR > 17.14% | 3.446 (3.200- 3.712) | < 0.001 | - | - |
Radiation record | | | | |
No RT | 1 | | 1 | |
Preop RT | 0.917 (0.825–1.009) | 0.092 | 0.844 (0.751–0.948) | 0.004 |
Postop RT | 0.710 (0.662–0.763) | < 0.001 | 0.742 (0.689- 0.800) | < 0.001 |
Chemotherapy | | | | |
Yes | 1 | | 1 | |
No | 1.378 (1.286–1.477) | < 0.001 | 1.487 (1.380–1.603) | < 0.001 |
Preop, preoperative; RT, radiotherapy; Postop, postoperative; LNR, lymph node ratio; HR, hazard ratio |
Among the subgroups, patients ≥ 60 years old, patients with poorly-differentiated and undifferentiated tumors, patients with signet ring cell carcinoma, patients with a high T stage and LNR underwent a higher risk of death than the references. Patients with sites of mid- and low-third and overlapping stomachs, patients with postoperative RT, and patients undergoing chemotherapy had a lower risk of death than the references. Preoperative RT group did not reduce the risk of death than no RT group in the multivariate analysis (HR: 0.917, 95%CI [0.825 ~ 1.009]).
Additionally, multivariate COX regression analysis based on lymph node status suggested preop RT (HR: 0.844, 95%CI [0.751 ~ 0.948]) and postop RT (HR: 0.742, 95%CI [0.689 ~ 0.800]) were associated with a lower risk of death than No RT in lymph node-positive patients, which is consistent with Kaplan-Meier survival analysis. The other factors related results are presented in Table 3.
Establishment And Validation Of The Nomograms
The selected prognostic factors were used to construct the nomogram for 1-, 3- and 5-year CSS by using multivariate Cox regression model (Fig. 5). The nomogram included parameters such as age, race, tumor differentiation, tumor histology type, primary tumor site, T stage, LNR, RT status, chemotherapy information.
The performance of the model (i.e., discrimination, calibration and clinical usefulness) was evaluated through C-index, calibration plots, and DCAs. The C-index value of the nomogram for CSS was significantly greater than those of nomogram based AJCC system (0.725 [95%CI: 0.717 ~ 0.734] vs. 0.643 [95%CI: 0.635 ~ 0.651]). Moreover, high-quality calibration plots demonstrated excellent consistency between the actual and nomogram-predicted survival probabilities (Fig. 6a-c). The DCAs curves revealed relatively good performance for the model according to clinical application. The threshold probabilities of the new model had excellent net benefits for predicting the 1-, 3- and 5-year CSS compared with the AJCC system (Fig. 6d-f). Therefore, we believe that the prediction model has highly accurate prediction ability as a whole and can be used in clinical work.