A Novel Nomogram to predict survival of patients with choroidal melanoma

The purpose of this study is to develop and validate a prediction model of overall survival (OS) in choroidal melanoma patients. Methods A total of 3415 patients with choroidal melanoma from 2004 to 2015 were collected from the Surveillance, Epidemiology, and End Results database. A novel nomogram was built based on independent risk factors evaluated by Cox proportional hazards regression model. And internal validations were performed to assess the performance of our nomogram by concordance index (C-index), calibration plot and decision curve analysis (DCA).


Introduction
Uveal melanoma (UM) is not only the most common intraocular malignant tumor in adults but also the second most common melanoma [1,2]. According to different primary sites, UM can be divided into the iris, ciliary body and choroidal tumors, among which choroidal melanoma accounts for more than 90% [3].
Unlike cutaneous melanomas, nearly half of choroidal melanoma develops hematogenous metastasis, whose median survival is only 3.6 months [4]. And although the latest immunotherapy has been successful in treating patients with cutaneous melanoma, its effect on UM is very limited [5,6]. Due to a high rate of distant metastasis, limited treatment and poor prognosis, it is urgently needed to further understand choroidal melanoma.
So far, multiple clinicopathological variables such as pathologic and genetic factors, are used to predict prognosis and guide clinical management. The increasing tumor size, largest basal tumor diameter, tumor thickness, ciliary body involvement and extraocular spread are all considered as important clinical characteristics [3,7]. When the thickness of the choroidal melanoma is less than 3 mm, the 10-year mortality rate is about 12%. And when it is greater than 8 mm, the 10-year mortality rate can be as high as 49%. Hence, the American Joint Committee on Cancer (AJCC) combined these clinical features to evaluate tumor staging for predicting poor prognosis. Also, chromosome abnormalities monosomy 3, loss of 6q and additional copies of 8q are all closely related to the poor outcome of patients with choroidal melanoma [8,9]. However, a comprehensive prediction model including various prognostic factors for the survival of choroidal melanoma patients is still lacking.
With the promotion of precision medicine, the prognostic models of cancer have been widely used to assist in decision-making and manage patient prognosis. One method is the nomogram. According to the regression coe cient of each variable in the multivariate regression model, each subtype of variables is scored by the nomogram. And then a total score can be calculated. Finally, the possibility of an individual occurrence of outcome events can be predicted by nomogram through analyzing the relationship between the total score and the outcome event. In this study, we will conduct and validate a nomogram to predict the overall survival (OS) of choroidal melanoma patients.

Patient selection
In this study, all cases were obtained by the SEER*Stat software (version 8.3.6) from the Surveillance, Epidemiology, and End Results (SEER) database, which covers publicly cancer data of 28% of the United States population [10]. The inclusion criteria were: 1. primary site record C69.

Construction and veri cation of nomogram
We enrolled 3415 eligible patients from the SEER database from 2004 to 2015, and employed univariable and multivariable cox regression analyses to selected variables and build regression models. The hazard ratio (HR) was calculated to show the contribution degree of each variable. And the nomogram was established for predicting outcomes of choroidal melanoma patients based on the multivariate cox regression model. For the most e cient use of datasets, we used bootstrap resampling for internal validation. The concordance index (C-index) was used to test the discrimination of models and the calibration plot was conducted to evaluate the calibration of models. Furthermore, the clinical net bene t of prediction models was measured by the decision curve analysis (DCA).

Identi cation of independent prognostic factors of choroidal melanoma
The cox proportional hazards models were applied to select risk and protective factors of choroidal melanoma. Univariate cox regression analyses indicated seven prognostic factors (age at diagnosis, race, sex, T-category, histological type, marital status and total number of primary tumors) of OS.
Moreover, as shown in Fig. 1, multivariable cox regression analysis indicated age at diagnosis (HR = 1.535, 95% con dence interval

Veri cation of nomogram
For the most use of the dataset, we employed bootstrap resampling to validate the nomogram. When compared with simple T-category, the c-index showed our nomogram presented good discrimination with 1000 bootstraps from 36 months to 60 months (Fig. 3A). The calibration plots also indicated that the nomogram had good calibration (Fig. 3B). Moreover, the nomogram has shown better clinical net bene t in both 3 and 5 years of OS prediction than the T-category by DCA (Fig. 3C).

Discussion
UM is a kind of rare malignant tumor, with around 1500 new cases each year in American. The 5-year relative survival of ocular melanoma is 78.4%. However, once patients develop distant metastases occurred, the reported 5-year survival rate would be around 16% [11]. Hence, the accurate prediction of patient outcome is critical to the selection of the optimal therapeutic strategy and improving the survival of patients. Here, we built and validated nomogram predicting the OS of choroidal melanoma in the light of the SEER database.
Mariani et al [12] developed a nomogram for liver metastasis in patients with UM and found the interval between the diagnosed and liver metastases < 6 months, more than 10 metastases of liver, a maximum liver metastasis area over 1200 mm 2 and lactate dehydrogenase (LDH) value > 1.5 were risk prognostic predictors for patients. However, this study only retrospectively enrolled 224 patients from a single center.
Valpione et al [13] also conducted a prognostic nomogram for metastatic UM including two cohorts and 254 patients and reported area of liver involvement, the increased value of serum LDH and a World Health Organization performance status between 2 and 3 were prognostic risk factors. Although more and more scholars have begun to pay attention to the prediction of tumor survival rate, the prediction model of the survival rate of choroidal melanoma is still lacking.
Over the last decades, the incidence rate of UM has been increasing, especially among white people [14,15]. However, for this rare disease, few studies have analyzed the prognostic differences between different races. Cormeier et al [16] reported that non-Caucasians are more likely to be in advanced stages of cutaneous melanoma than whites. Interestingly, a study shows that there is no difference in mortality among different races in UM [17]. And in our study, we found white is an independent prognostic risk factor in both univariate and multivariate regression. The result may be in uenced by the small sample size of non-Caucasians patients, and need to be veri ed by a larger sample size cohort study.
Multiple primary cancers are relatively uncommon [18], and whether patients with UM are more likely to have multiple primary cancers remains controversial [19,20]. In our study, we discovered that choroidal melanoma patients with multiple primary cancers are a protective factor for the rst time. Marzena et al [21] once reported choroidal melanoma patients with a family history of cancer, higher education, living in cities, previous surgery history except for UM, and female pregnancy two or less than two were more likely to be detected with multiple primary cancers. And these factors may be useful for early detection or treatment and prolong survival.
However, there were still some limitations to this study. The genetic (chromosome abnormalities monosomy 3, loss of 6q and additional copies of 8q) and genomic abnormalities are also predictors of patient outcome [22][23][24]. Besides, many circulating biochemical markers, yet to be studied, may play important roles in predicting the prognosis of choroidal melanoma as well [25,26]. Unfortunately, due to the lack of these data in the SEER database, these factors cannot be included in this retrospective study. We also need to use an external cohort to verify the transportability and generalizability of the nomograms [27], and this will be our next research direction. Besides, due to the latest data having not been updated, our study only adopted the 7th AJCC TNM staging. Furthermore, our nomogram will be optimized on the basis of the 8th AJCC TNM staging.

Conclusion
In conclusion, we found that old white unmarried male patients with T4 category choroidal melanoma had shorter OS. Furthermore, we developed and validated two nomograms based on these variables to predict 3 and 5-year OS of patients with choroidal melanoma. And our prognostic models may assist clinical medical decision-making and managing patient prognosis. Ethical approval was not required for this study, because the data used in this study were obtained from the SEER database in a publicly available manner. manuscript. YY, XTH and ZMY reviewed and revised the manuscript and produced the nal version. All authors read and approved the nal manuscript.   Validation of the nomogram