Clinical and pathological characteristics & Survival analysis
A total of 22,181 patients who met the inclusion criteria were screened in our study. Analysis of patients included in our study showed that EOC incidence in the elderly (≥65 years old) exhibited an increasing trend, with the highest incidence in the 65-79 years age group, accounting for 81.3% of all elderly ovarian cancer patients. Only 45% of patients had health insurance coverage, and only 42% of elderly patients were positive for CA125. Most elderly ovarian cancer patients presented with advanced-stage disease (stage III-IV accounting for 71.7%) and high tumor grade, with stage III-IV tumor accounting for 72.6%. Serous tumors accounted for a large proportion (71.9%). However, only 40% of patients underwent complete cytoreduction at the initial surgery, and complete lymph node dissection was only conducted in 20% of patients who underwent lymph node dissection. Residual disease was found in 4% of patients with a residual tumor less than 1 cm after surgery. 69.3% of patients received chemotherapy (Table 1).
Prognostic factors of elderly (≥65years old) patients with OC
X-tile software (v3.6.1) was used to identify the cut-off point and stratify each variable: age 65-71, 72-79, 80-101years; the number of lymph nodes examined 1-2, 3-7, ≥8; the number of positive lymph nodes 1, ≥2 and tumor size ≤6.4cm, 6.6-9.4cm, ≥9.5cm. (Figure 2A)
Factors influencing prognosis in elderly ovarian cancer patients were identified in the training cohort using a Cox regression model. During univariate analysis, 21 parameters, including stage, lymph node-positive rate, site of primary surgery, age, residual lesion size, CA125 level, grade, tumor size, pathological type, and chemotherapy, exhibited statistically significant differences in OS (ps< 0.05). Further screening was performed by LASSO analysis (P>0.01 were screened), yielding 18 factors related to OS (Figure 2B). Further multivariate Cox analysis (P<0.01) showed that 12 parameters, including stage, pathological type, age, size of residual lesions, and chemotherapy, were independent prognostic risk factors for elderly patients with ovarian cancer (Table 2). Similarly, 21 items such as age, tumor size, lymph node-positive rate, grade, stage, pathological type, CA125, residual lesion size, and chemotherapy were screened out by univariate analysis to identify risk factors associated with CSS. After LASSO analysis, we found that age, residual lesion size, tumor size, grade, stage, pathological type, chemotherapy and 17 other items were closely related to CSS. Multivariate COX analysis further removed confounding factors and found that 12 parameters, such as stage, pathological type, age, size of residual disease, and whether or not chemotherapy, were closely associated with CSS (Figure 2C).
Nomogram Construction and Validation
Construction nomogram
Independent prognostic factors identified by multivariate Cox analysis were used to build a nomogram model to predict 3-, 5-, and 10-year OS and CSS in elderly patients with ovarian cancer. The scores of each independent prognostic factor in the nomogram were summed up to obtain a total score projected to OS and CSS at 3, 5, and 10 years. A high total score indicated a poorer prognosis in elderly patients with ovarian cancer. The OS nomogram model showed that among the 12 prognostic-related factors screened, the tumor stage had the greatest impact on prognosis. Independent risk factors that affected the prognosis were pathological type, age, whether receiving chemotherapy, size of residual lesions, and the number of positive lymph nodes. Moreover, we found that the number of lymph node biopsies, tumor grade, tumor laterality, CA125 level, tumor size, and the primary surgical site had the least effect on OS prognosis. The CSS nomogram model showed that among the 11 prognostic factors screened out, tumor stage had the greatest impact on prognosis, followed by pathological type, age, residual disease, chemotherapy, tumor grade, positive rate of lymph nodes, number of lymph node biopsies. The location of primary surgery, tumor size, tumor laterality, and the positive rate of tumor marker CA125 had the least impact on prognosis (Figure 3).
Validation of nomogram
Internal validation of the nomogram. The nomogram showed good predictive value for OS (C-index 0.705 and 0.699) and CSS (C-index 0.703 and 0.707) in the training and validation groups(Table 3), significantly higher than the AJCC staging, suggesting that our nomogram yielded significantly better predictive performance than the AJCC system.
The AUCs of the training and validation cohorts are shown in Figures 4A, 5A. The AUCs of the training cohort for OS prediction at 0.5, 1, 3, 5, and 10 years were significantly higher than the AJCC staging system (0.789, 0.765, 0.749, 0.766, and 0.781 vs. 0.633, 0.640, 0.676, 0.714, and 0.739, respectively). In the validation cohort, the AUCs for OS prediction at 0.5, 1, 3, 5, and 10 years were significantly higher than the AJCC staging system (0.776, 0.765, 0.740, 0.757, and 0.776, vs. 0.646, 0.640, 0.671, 0.703, and 0.737, respectively). Besides, the AUCs for predicting CSS in the training cohort at 0.5,1, 3, 5, and 10 years were significantly higher than the AJCC staging system (0.784, 0.760, 0.741, 0.764, and 0.803 vs. 0.625, 0.632, 0.667, 0.708, and 0.766). Finally, the AUCs to predict CSS in the validation cohort at 0.5, 1, 3, 5, 10 years were significantly higher than the AJCC staging system (0.783, 0.751, 0.743, 0.768, and 0.801 vs. 0.646, 0.633, 0.662, 0.701, and 0.761). We further analyzed and compared the time-dependent AUC of OS and CSS at 1 to 10 years between the nomogram and AJCC staging system in the training and validation groups. The results showed that our nomogram yielded significantly better prediction power than the AJCC staging system (Figure 4 C and D, Figure 5 C and D). Indeed, in contrast to the AJCC staging system, which only involves the primary site of the tumor, lymph node metastasis and distant metastasis, our OS nomogram prediction model consisted of 12 independent prognostic risk factors, while the CSS nomogram prediction model incorporated 11 independent prognostic factors. Even though our nomogram prediction contained multiple independent and potential confounding factors than the AJCC analysis system, the predictive power was still better than the AJCC staging system. Calibration plots at 1, 3, 5, 10 showed good agreement between OS/CSS nomogram predictions and actual observations in training and external validation cohorts (Figure 6). DCA was used to compare the benefits of our established OS and CSS nomograms and AJCC staging systems. Compared with the AJCC staging system, the DCA curves of the nomogram showed a greater net gain in the training and external validation cohorts (Figure 7).
Risk stratification in elderly (≥65 years old )ovarian cancer patients with ovarian cancer
A risk score for each variable was generated from the nomogram, and an overall score was calculated for all patients. X-tile was used to determine the cut-off values. The entire cohort was divided into low-risk and high-risk subgroups based on the median risk score. KM was used to conduct survival analysis between the groups (Figure 8), and significant differences in training cohort and validation cohort of OS (P < 0.001) and CSS (P < 0.001) were observed between the low- and high-risk groups, indicating an excellent nomogram risk stratification performance.