2.1 Patient selection

Women diagnosed with ESS between January 1988 and November 2015 were initially identified from the SEER database, in which all deposited cases came from the United States, using SEER*Stat 8.35. LG-ESS was defined as grade I/II, well and moderately differentiated tumors [6, 21]. The inclusion criteria were as follows: (i) the International Classification of Diseases (ICD) code O-3 morphology 8930/3 or 8931/3; (ii) SEER site recodes of ICD-O-3 included the corpus uteri/uterus not specified; (iii) active follow-up to ensure reliable patient status; (iv) ESS as the only or first primary tumor that was confirmed by histology. The exclusion criteria were as follows: (i) missing information on ethnicity, American geographic region, marital status, lymph node status, tumor stage (determined using the FIGO criteria-based tumor staging), surgery type, lymphadenectomy, adjuvant radiotherapy, or adjuvant chemotherapy; (ii) patient died within 1 month or was followed up less than 1 month since initial diagnosis.

2.2 Cohort definition and variable recode

The patients were divided into the training and validation cohorts with a ratio of 7:3 using the R function “createDataPartition” to ensure that outcome events were distributed randomly between the two cohorts. The training cohort was used to screen variables and construct the model. The validation cohort was used to validate the results obtained using the training cohort. Eleven variables from the SEER database were included: age (at diagnosis), ethnicity, geographic region, marital status (at diagnosis), tumor size, lymph node status, tumor stage, surgery type, lymphadenectomy, radiotherapy, and chemotherapy. Ethnicity, region, marital status, tumor size, and surgery type were recorded. Surgery types included hysterectomy with bilateral salpingectomy and ovariectomy oophorectomy (Hys + BSO); hysterectomy without bilateral salpingectomy and ovariectomy oophorectomy (Hys – BSO); hysterectomy, not specified (Hys NOS); local surgery, exenteration, or other surgery types (L/E/OTH). The tumor stage was recoded based on the 2009 FIGO staging criteria: localized stage corresponded to FIGO stage I, and regional stage corresponded to FIGO stages II and III, and distant stage corresponded to FIGO stage IV [21, 22]. Univariate and multivariate Cox regression analyses were performed for all 11 variables, and variables with P < 0.05 in both univariate and multivariate Cox regression were identified as independent risk factors.

2.3 Statistical analysis

The stepwise regression based on the Akaike information criterion minimum was used to select variables for inclusion in the nomogram [15]. The 5-/10-year overall survival (OS) probabilities were estimated using the nomogram. Concordance index (C-index) and area under the time-dependent receiver operating characteristic curve (time-dependent AUC) calculated by bootstrapping were used to evaluate discriminative ability. Calibration plots were used to evaluate calibrating ability. C-index and AUC values vary from 0.5 to 1.0, where 0.5 represents random chance and 1.0 indicates a perfect fit. Typically, C-index and AUC values greater than 0.7 suggest a reasonable estimation.

The net reclassification index (NRI), integrated discrimination improvement (IDI), and decision curve analysis (DCA) were used to evaluate the clinical benefits and utility of the nomogram compared with a FIGO criteria-based tumor staging alone. NRI and IDI are two alternatives to AUC to assess improvement in risk prediction and measure the usefulness of a new model [23, 24]. DCA is a method for evaluating the clinical benefit of alternative models [25, 26] and was applied to nomograms by quantifying net benefits at different threshold probabilities. The curves of treat-all-patients scheme (representing the highest clinical costs) and the treat-none scheme (representing no clinical benefit) were plotted as two references. Risk stratifications with the nomogram and the FIGO criteria-based tumor staging were compared using Kaplan-Meier method and the Cox model. The cut-off point for risk stratifications was selected using X-tile [27].

OS was the endpoint of interest in the present study. It was calculated from diagnosis to death of all causes or to date of last follow-up in November 2015. Data from patients alive at the last follow-up were censored. Inclusion of covariates in the nomogram followed Harrell's guideline (the number of events should exceed the number of covariates by at least 10 folds) [15]. Meanwhile, the variance inflation factor (VIF) was assessed among the covariates in the nomogram, and VIF > 4.0 was interpreted as indicating multicollinearity. Variables with VIF > 4.0 were not included in the final model analysis. Statistical differences of distribution in age between the training and validation cohorts were evaluated using the Wilcoxon-test, and other variables were analyzed by using the Chi-square test. All P values were two-tailed, and P < 0.05 was considered statistically significant. All statistical analyses were performed using the R programming language and environment (http://www.r-project.org/).

3.1 Characteristics of patients and disease

A total of 1172 patients were identified as having LG-ESS and randomly divided into a training cohort and a validation cohort by a ratio of 7:3. The median follow-up was 127 [interquartile range (IQR): 62-189] months in the whole population, 127 (IQR: 62-190) months in the training cohort, and 128 (IQR: 60-186) months in the validation cohort. The demographic and clinical characteristics of these LG-ESS patients are summarized in Table 1. In the whole population, training, and validation cohorts, the median ages of patients with LG-ESS were 48 [interquartile range (IQR): 42–53], 48 (IQR: 43-54), and 47 (IQR: 42-53) years, respectively. Patients of white ethnicity (75.5%), those from the Pacific coast/Asian region (51.4%), and married women (63.3%) constituted the majority of the whole LG-ESS population. Moreover, the whole population had a relatively low rate of lymph node metastasis (12.6%). However, a third of the population received lymphadenectomy, and 9.0% of them underwent positive lymphadenectomy. More than 70% of patients had local tumor invasion. Hys + BSO (76%) was the main type of surgery underwent by LG-ESS patients. The training and validation cohorts were comparable in terms of demographic and clinical characteristics (P > 0.05).

3.2 Nomogram variable screening

According to the stepwise regression results, the model containing age, marital status, tumor size, tumor stage, chemotherapy, radiotherapy, and lymphadenectomy had minimal AIC value in the training cohort. The VIF values were all < 4, indicating that no collinearity existed between screened variables. In the univariate regression analysis, eight variables (age, marital status, tumor size, tumor stage, surgery type, chemotherapy, radiotherapy, and lymphadenectomy) were significantly associated with OS. In the multivariate Cox regression analysis, age, marital status, tumor size, tumor stage, chemotherapy, and lymphadenectomy were identified as independent prognostic factors for LG-ESS (Table 2).

3.3 Nomogram construction and validation

We constructed a nomogram for LG-ESS according to the variables screened. The top four factors ranked by the standard deviation (SD) along nomogram scales for the nomogram model were age (SD: 14.71), tumor size (SD: 12.91), chemotherapy (SD: 10.61), and tumor stage (SD: 8.51). Figure 1 shows an example of using the nomogram to predict survival probability of a given patient. The total score was determined based on the individual scores calculated using the nomogram; most patients in the present study had total risk points ranged from 260 to 400.

The C-index value was 0.814 [95% confidence interval (CI) = 0.769-0.860] in the training cohort and 0.837 (95% CI = 0.781-0.893) in the validation cohort. The time-dependent AUC was > 0.7 for the prediction of OS within 10 years in both the training and validation cohorts (Figure 2), indicating favorable discrimination by the nomogram. The calibration curves of the nomogram showed high consistencies between the predicted and observed survival probability in both the training and validation cohorts (Figure 2). In summary, the nomogram for LG-ESS had considerable discriminative and calibrating abilities.

3.4 Clinical value of the nomogram compared with the FIGO criteria-based tumor staging

The changes in C-index, NRI, and IDI were used to compare the accuracy between the nomogram and the FIGO criteria-based tumor staging alone. While using the nomogram in the training cohort, the C-index was 0.163 (95% CI = 0.107-0.209, P < 0.001), the NRI for the 5- and 10-year OS were 0.271 (95% CI = 0.078-0.44) and 0.433 (95% CI = 0.276-0.595), and the IDI values for 5- and 10-year OS were 0.146 (95% CI = 0.08-0.247, P < 0.001) and 0.185 (95% CI = 0.121-0.282, P < 0.001) (Table 3). These results were validated in the validation cohort (Table 3), indicating that the nomogram predicted prognosis with greater accuracy than the FIGO criteria-based tumor staging.

The clinical benefits of the nomogram were compared with those of the FIGO criteria-based tumor staging. DCA curves showed that the nomogram could better predict the 5- and 10-year OS, as it added more net benefits compared with the FIGO criteria-based tumor staging for almost all threshold probabilities in both the training and validation cohorts, and with both the treat-all-patients scheme and the treat-none scheme (Figure 3).

3.5 Risk stratification based on the nomogram

We finally made a risk stratification based on total points calculated using the nomogram. Patients with LG-ESS were divided into three risk groups[27]: low risk (total points < 325), middle risk (325 ≤ total points < 360), and high risk (total points ≥ 360). The Kaplan-Meier OS curves showed great discrimination among the three risk groups, whereas the FIGO criteria-based tumor staging had limited ability to recognize high-risk patients in both the training and validation cohorts (Figure 4).

ETHICAL APPROVAL AND CONSENT TO PARTICIPATE

Not applicable

CONSENT FOR PUBLICATION

Not applicable.

COMPETING INTERESTS

The authors declare no competing interests.