A novel nomogram to predict the overall survival in esthesinoeroblastoma

For the training cohort, histological feature code 9522 was used to identify all patients diagnosed with ENB from January 1, 1976, to December 30, 2016, in the SEER database. We signed and adhered to the data use agreement for SEER radiation therapy and chemotherapy information to obtain the chemotherapy and radiation therapy data from SEER database. Specific site codes C30.0, C31.0, C31.1, C31.2, C31.3, C31.8, C31.9 were used to identify the specific location of the tumor in the nasal cavity or paranasal. Although modified Kadish staging was not available in the SEER database, we used SEER extent of disease, collaborative stage extension, historic stage, and primary site to deduce modified Kadish staging. Jethanamest et al. and Tajudeen et al. used this method of modified Kadish stage derivation for SEER studies pertaining to ENB [11, 18]. Extent of disease and collaborative staging extent codes for anatomic involvement of primary tumors were grouped and correlated with the appropriate modified Kadish stage as follows: confined to the nasal cavity (stage A), extension to the paranasal sinuses (stage B), extension beyond the nasal cavity and sinuses, including the cribriform plate and base of skull (stage C), and lymph node and distant metastases (stage D). We invited two experienced clinicians to derive the modified Kadish stage for the SEER cohort. When there are disagreements, they determined through consultation. We defined tumor differentiation grades I and II as low-grade tumors, and defined grades III and IV as high-grade tumors. The retrospective study followed the Helsinki Declaration (1964) and its later amendments or comparable ethical standards. No consent was required for the deidentified data, and no additional ethical approval processes were required for access to the database. Patient information was acquired by SEER*Stat software (version 8.3.6).

The inclusion criteria for the training cohort was as follows: 1. ENB with positive histological confirmation and not from an autopsy or death certificate; 2. active follow-up patients; and 3. known survival months after diagnosis and cause of death. The exclusion criteria for both the training cohort and the validation cohort were as follows: 1. unknown demographic information (age, race, sex); 2. unknown clinicopathological information (tumor grade and modified Kadish stage); 3. ENB was not the primary tumor if there were 2 or more; and 4. the follow-up time was less than 1 month. A total of 639 patients were excluded due to unknow demographic and clinicopathological information, or not first tumor, or follow up time less than 1 month after treatment. The flow diagram of training cohort data selection is shown in supplementary Figure 1. After applying the screen criteria, 225 patients were included in the final SEER cohort.

For the validation cohort, ENB patient data were collected from Sun Yat-sen University Cancer Center and The First Affiliated Hospital of Sun Yat-sen University in the same period. This study was approved by the institutional review boards of the First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China ([2020]111). All inclusion criteria were identical to those used in the SEER cohort. A total of 96 patients with pathologic-proven ENBs were screened in this period, but 2 patients were excluded due to not first tumor and 7 patients were excluded because unknow demographic and clinicopathological information. After applying the screen criteria, 87 patients were included in the validation cohort. The Pearson’s chi-squared test was used to assess significance of clinicopathological and demographic characteristics.

The modified Kadish classification was used to classify primary tumor extension: stage A is confined to nasal cavity, stage B extends into paranasal sinuses, stage C extends beyond the paranasal sinuses and stage D presents cervical lymph node involvement (supplementary Table 1). For training cohort, tumor grade based on record of the SEER data base, while for the validation cohort, the tumor grade criterion was the Hyams Grading system (supplementary Table 2) and reviewed by two trained pathologists.

OS was defined as the time from ENB diagnosis to the time of death or last follow-up. The OS length was calculated from the time of death for any cause or censoring. In this study, we used Kaplan-Meier analysis to calculate 3- and 5-year survival for covariates and used the log-rank test to determine statistical significance.

Based on the results of multivariate analysis, a nomogram model was formulated. All variables with significant differences at p < 0.05 in univariate analysis were included in the multivariate analysis. We used the Cox proportional hazards model for multivariate analyses. We used model coefficients to determine hazard ratios. Prognostic factors with a p-value < 0.05 in multivariate analysis were included in the nomogram. The optimal cut-off values of age group and the nomogram score for risk group stratification were calculated by X-tile 3.6.1 software (Yale University, New Haven, CT, USA).

The nomogram’s predictive power was evaluated by the c-index for both of training cohort and validation cohort. We used the c-index to evaluate the predictive power of the nomogram for both cohorts. The c-index was used to quantify the difference between the prediction and the actual situation [19]. Values ranged from 0.5 (no discrimination) to 1.0 (complete discrimination). A larger C-index predicts a more accurate prediction of the prognosis. The agreement between predicted survival and the observed survival after bias correction was quantified by calibration curves of the nomogram for the 3-year and 5-year OS. Statistical analysis was conducted by R software version 3.5.2 (R Foundation for Statistical Computing, Vienna, Austria; www.​R-project.​org). All calculated p values were two-sided, and p < 0.05 was considered statistically significant.

The training cohort comprised 225 patients with ENB who were recruited between 1976 and 2016. The nomogram was based on the training cohort, and the median OS time was 48 months (1–155 months). In the training cohort, 122 (54.2%) patients were diagnosed at the age of 54 or younger. A total of 133 (59.1%) patients were males, and 92 (40.9%) were females. Additionally, 129 patients had a low-risk tumor grade, accounting for 57.3% of the total, and 96 (or 42.7%) had a high-risk tumor grade. A total of 60.9% of the patients were diagnosed with stage C disease. Most of the patients in training cohort had received surgery (90.2%) and radiotherapy treatment (68.0%), while most of the patients had no chemotherapy treatment or had no information about chemotherapy (64.9%). In terms of treatment options, due to the limitation of the SEER database, the sequence of chemotherapy with surgery and radiotherapy was unknown.

For the validation cohort, we studied 87 consecutive patients in the same period, and the median OS time was 29 months (1–208 months). Fifty-eight (66.7%) patients were males and 29 (33.3%) patients were females in the validation cohort. The most common age at diagnosis of these patients with ENB was ≤54 (70.1%). With regard to tumor stage, modified Kadish C stage (42.5%) was most frequent, followed by B stage (29.9%), D stage (26.4%) and A stage (1.1%). The majority of patients received radiotherapy (66.7%). Only 41.4% of patients in the Chinese cohort had received surgical treatment (Table 1).

Patient characteristics and tumor characteristics were assessed by Kaplan-Meier survival analysis. The modified Kadish A or B group had the highest 5-year OS rate (89.4%). The C group and D group represented 72.1 and 50.7%, respectively (Fig. 1a). In terms of the tumor differentiation grade characteristics, patients in the high-grade group had a comparatively lower OS rate, reaching 63.9%, while in the low-grade group, the 5-year OS rate was 81.8% (Fig. 1b). The 5-year OS rates for training cohort patients younger than 55 years old, 55–69 years old and older than 70 were 83.1, 71.2 and 44.2%, respectively (Fig. 1c). The 5-year OS of the low-risk group was 93.0%, followed by the medium-risk group (63.4%) and the high-risk group (28.3%) (Fig. 1d).

Univariate analyses have demonstrated that modified Kadish stage, gender, tumor differentiation grade, age at diagnosis, chemotherapy, and surgery are associated with OS. We included all of the above prognostic factors with p < 0.05 in the multivariate analysis, and multivariate analysis showed that modified Kadish stage, age at diagnosis, and tumor differentiation grade were independent risk factors for patients with ENB. The detailed results of the multivariate analysis are presented in Table 2.

In the Cox model, modified Kadish stage, tumor differentiation grade, and age at diagnosis were independent prognostic factors revealed by multivariate analyses. Modified Kadish stage, tumor differentiation grade, and age at diagnosis were used to develop the nomogram for estimating 3- and 5-year OS (Fig. 2). To use a nomogram, lines are drawn to score the prognostic variables on the top point scale for an individual patient. The number of points received for each variable value and the score for each prognostic variable on the point scale are added together. The sum of scores is on the total point axis, and one line is drawn to the survival axis to convert to a 3- or 5-years probability.

Patients were subdivided into a low-risk group (0 ≤ score ≤ 57.5), an intermediate-risk group (scoring 57.5 < score < 157.5) and a high-risk group (157.5 ≤ score ≤ 300).

In this study, we performed both internal and external validation of the nomogram. The plotted calibration curves corresponded to the ideal plot (the 45°line), which revealed a favorable agreement on the nomogram estimation and actual observation regarding the probability of 3-year and 5-year survival (Fig. 3a, b, c, d). In the training cohort, the model showed a high accuracy with a c-index of 0.737 (95% CI, 0.709 to 0.765) which was higher than the modified Kadish staging system, at 0.614 (95%CI, 0.579 to 0.649). In the validation cohort, the nomogram prediction was 0.791 (95% CI, 0.767 to 0.815) was also higher than for the modified Kadish staging system prediction (0.674, 95% CI, 0.643–0.705). These results suggest that the nomogram was reasonably accurate, repeatable and had a better accuracy in predicting OS than the modified Kadish staging system.