Prognostic analysis of cutaneous Kaposi sarcoma based on a competing risk model

The data regarding the prognosis of cutaneous Kaposi sarcoma (KS) was limited. The current study aimed to explore the risk factors and develop a predictive model for the prognosis of cutaneous KS patients. Data were extracted from Surveillance, Epidemiology, and End Results database from 2000 to 2018 and randomly divided into training and validation cohort. The Kaplan–Meier analysis, cumulative incidence function based on the competing risk model and Fine–Gray multivariable regression model was used to identify the prognostic factors and then construct a 5-, 10-, and 15-year KS-specific death (KSSD) nomogram for patients. The concordance index (C-index), area under the curve (AUC) of operating characteristics and calibration plots were used to evaluate the performance of the model. The clinical utility of the model was measured by decision curve analysis (DCA). In 2257 cutaneous KS patients identified from database, the overall median survival time was about 13 years. Radiotherapy (p = 0.013) and surgery (p < 0.001) could lower the KSSD, while chemotherapy (p = 0.042) and surgery (p < 0.001) could increase the overall survival (OS) of patients with metastatic and localized lesions, respectively. Race, number of lesions, surgery, extent of disease, year of diagnosis and age were identified as risk factors associated with cutaneous KS-specific survival. Performance of the nomogram was validated by calibration and discrimination, with C‐index values of 0.709 and AUC for 5-, 10-, and 15-year-KSSD of 0.739, 0.728 and 0.725 respectively. DCA indicated that the nomogram had good net benefits in clinical scenarios. Using a competing-risk model, this study firstly identified the prognostic factors, and constructed a validated nomogram to provide individualized assessment and reliable prognostic prediction for cutaneous KS patients.


Construction of the nomogram
The main endpoint in the present study was KS-specific death (KSSD).KSSD and other causes of death (Non-KSSD) were two events in our competing-risk analysis.The competitive risk analysis model was applied to control the risk of other cause of death while comparing the difference in the risk of KSSD.The CIF (cumulative incidence function) was used to describe the probability of death and Gray's test to applied to compare CIF across categories 18 .We calculated the CIF for each event and plotted CIF curves.simple competing risk model was used to check each factor's power in predicting KSSD.Variables considered clinically relevant or significant differences in simple analysis (p < 0.2) would be introduced into the multiple competitive risk analysis model.variables with p values < 0.05 in the multivariable analysis were included in the final model.Based on the coefficients from the competing risks regression models, a nomogram based on Fine and Gray's model was built by the R packages "mstate" and "regplot" 19 .Meanwhile, a nomogram based on the multiple COX proportional hazard model was constructed to compare the two results.

Discrimination and calibration of the nomogram
The concordance (C)-index was applied to evaluate discrimination 20 and calibration was assessed using a calibration plot.1000 bootstraps were used for plotting the calibration curve and calculating the C-index.The C index ranges from 0.5 to 1, where 1 means perfect discrimination and 0.5 means no discrimination.The calibration plot showed the correlation between the predicted probability and the frequency of the observed outcome.The standard curve was a straight line with a slope of 1 passing through the origin of the coordinate axis.The closer the calibration curve was to the standard curve, the better the predictive ability of the nomogram 21 .Decision curve analysis (DCA) was used to evaluate the clinical utility and net benefits of competitive risk models 22 .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 23 .

Statistical analysis
Continuous variables conforming to the normal distribution were expressed as the mean ± standard deviation (SD), while continuous variables with a skewed distribution were presented as the median and interquartile range (IQR).Categorical variables were shown as frequencies and their proportions.Chi-square test or Fisher's exact test was used to analyze the statistical difference of the categorical covariates distribution between the training cohort and the validation cohort, and Mann-Whitney test for continuous and non-normally distributed covariates.A two-tailed p < 0.05 was considered to be statistically significant.All statistical analyses and visualization were performed by using R studio statistical software version 1.4.1717(https:// www.r-proje ct.org).

Demographics and clinicopathological characteristics of patients
A total of 2257 patients with cutaneous KS were identified from the SEER database from 2000 to 2018, of which 1808 (80%) were assigned to the training cohort, and 449 (20%) cases were assigned to the validation cohort.Figure 1 presented the detailed screening process in the SEER database.The median age at diagnosis was 40 years (IQR: 33-49), and the median follow-up of the whole study cohort was 47 months (IQR: 8-120).A larger proportion of cutaneous KS patients were aged under 60 years (1960, 86.8%), male (2146, 95.1%), single (1736, 76.9%), Non-Spanish-Hispanic-Latino (1682, 74.5%), and non-multifocality (1866, 82.7%).Among all patients, 283 (12.5%) patients performed surgery, 158 (7.0%) and 815 (36.1%) patients undergone radiotherapy and chemotherapy respectively.The distribution of race was as follows: 622 (27.6%) were black, 1458 (64.6%) were white, 177 (7.8%) were other race.Most patients were diagnosed with a localized disease (887, 39.3%), followed by regional (667, 29.6%), distant (125, 5.5%), and other (578, 25.6%).The distant metastatic organs of cutaneous KS include 19 (0.8%) cases of lung, 7 (0.3%) cases of liver, 11 (0.5%) cases of distant lymph nodes, and 7 (0.3%) cases of bone.The baseline demographics and clinicopathological characteristics of all eligible patients were presented in Table 1.The patient characteristics of the training (n = 1808) and validation cohorts (n = 449) were concluded in Table 2. Since all cases were randomly assigned to the training and the validation cohort, there was no statistical difference between the variables of the two cohorts (all the P values > 0.05).The regional distribution and year of diagnosis characteristics of all acquired cases were visualized in Fig. 2A, B, respectively.www.nature.com/scientificreports/California has the most confirmed cases.There seemed to be a downward trend in the annual number of cases over the past two decades, which may reflect a decline in the incidence of cutaneous KS.

Kaplan-Meier analysis of OS
A total of 981 (43.5%) patients died in this study, and 723 (32.0%) of them had a KSSD, while 258 (11.4%) did not.The results of OS analysis for each variable were presented in Fig. 3, which showed that men (p = 0.003), age < 60 (p < 0.001), White race (p < 0.001), year of diagnosis after 2008 (p = 0.017), surgery performed (p = 0.002) were associated with better OS for the patients.In terms of extent of disease, patients with a localized lesions had the best prognosis, followed by regional lesions and distant metastatic lesions (p < 0.001).Figure 3A indicated that nearly half of cutaneous KS patients died within 13 years.Further subgroup survival analysis (Fig. 4) suggested that for regional or metastatic lesions, age or gender did not show statistical differences (All p value > 0.05).For localized lesions, men (p < 0.0001) or patients younger than 60 years old (p < 0.0001) have a better OS.For regional lesions, multifocality may be associated with worse OS (p = 0.021).The benefits of surgery and chemotherapy were www.nature.com/scientificreports/manifested in localized (p = 0.0052) and distant metastatic lesions (p = 0.042), respectively.For regional lesions, although chemotherapy seemed to be associated with better OS, no statistical significance was shown in this study (p = 0.066).No statistical difference was found between radiotherapy and OS (Fig. 4E, all p value > 0.05).

Nomogram variable screening by competing risk analysis
The CIF curves of KSSD and Non-KSSD by different clinicopathological characteristics were shown in Fig. 5.

Construction of competing risks regression nomogram model
All statistically significant clinical parameters after multiple analysis, and clear clinical prognosis-related factors including age, year of diagnosis, race, extent of disease, number of lesions and surgery were used for the construction of nomogram.Therefore, a nomogram established by competing risks regression models were constructed to calculate the 5-year, 10-year and 15-year cumulative KSSD probabilities in patients with cutaneous KS (Fig. 6A).The C-index of this nomogram for the training cohort was 0.709, demonstrating good accuracy for KSSD prediction.At the same time, another nomogram with a C-index of 0.625 (Fig. 6B) based on the multiple Cox regression model was developed to compare the KSSD difference with the above model.For each patient, first located the values of different variables in the corresponding variable row, and then added the scores of all the variables to get the total score.Drew a vertical line pointing to the bottom scale according to the total score to get the corresponding KSSD possibility.For example, a for a given patient (Patient ID = 35767086), the nomogram based on the competing risks regression models indicated that this cutaneous KS patient may had a 5-year KSSD of 29.2%, a 10-year KSSD of 34.9%, and a 15-year KSSD of 38.4%.However, the nomogram based on the Cox regression model suggested that the corresponding values were 36.3%, 40.9%, and 43.8%, which were all higher than the values based on the competing risk model.

Validation and calibration of the nomogram
The

Discussion
Cutaneous KS was a rare cancer, epidemiological research into it was therefore scarce.So far, there was no universally accepted staging classification for the cutaneous KS 24 ,which illustrated the necessity to explore the Among the states included in the registration, California had the most recorded cases, followed by Georgia and New Jersey (Fig. 2A), which may due to a higher proportion of MSM (Men who have sex with men) among HIV population in these states 25 .Meanwhile, the annual number of cutaneous KS cases recorded in 18 registered areas in the United States seemed to show a downward trend, which may roughly reflect changes in the overall incidence from 2000 to 2018 (Fig. 2B).It was consistent with the previous reports in the literature 26 .Additionally, other countries also showed a decrease in the incidence of KS such as Italy 27 and Switzerland 28 , etc. Scholars attributed this decrease to the introduction of antiretroviral therapy (ART), which had led to fewer HIV-infected patients developing KS 29,30 .
Preliminary analysis showed that nearly 50% of cutaneous KS patients died about 13 years after diagnosis (Fig. 3A).However, a study on 204 classic KS from Italian population-based cancer registries showed a median survival of 9.4 years 31 .From the Fig. 3A, the result was similar to the 67% 5-year net survival reported in the literature, which lower than 79% in Europe 32 .Male (vs.female) and patients with a diagnosis age less than 60 (vs.> 60 years) had better OS (Fig. 3B, C), and further subgroup analysis suggested that this difference was mainly manifested in cases with localized lesions (Fig. 4A, B).Age had been reported to be an important risk factor 33 , while male as a risk factor was mainly confirmed in AIDS-related KS in Africa 24 .Additionally, year of diagnosis later than 2008, White race, and localized lesions were associated with better OS (Fig. 3D, E, G).This may be www.nature.com/scientificreports/due to patients diagnosed later 2008 receiving more systematic ART or more advanced treatment options 29 .Meanwhile, Royse et al. also reported that African Americans compared to white race were associated with lower 5-year overall survival (61.1% vs. 64.9%)and KS-specific survival (63.3% vs.75.5%) 34 .For patients with regional  lesions, multiple lesions may mean worse OS (Fig. 4C).Spanish-Hispanic-Latino and marriage were not factors affecting OS (Fig. 3F, L).In terms of treatment, surgery and chemotherapy could improve the OS of patients with localized lesions and patients with metastatic lesions, respectively (Fig. 4D, F).It was considered that surgical excision should not be applied to extensive lesions, and repeated surgical excisions may cause severe functional impairment 24 .Systemic therapy was only to achieve disease control and symptom relief while maintaining quality of life.Accordingly, the recommended first-line agents were pegylated liposomal doxorubicin (PLD) and paclitaxel (PCT) 24 .The latest research also confirmed that paclitaxel plus ART could be used for the treatment of advanced AIDS-related KS with limited resources 13 .However, this analysis shown that whether radiotherapy did not affect the patient's OS (Fig. 4E).Other local therapies also included cryosurgery and laser 35 , isolated limb perfusion 4 and local or intralesional chemical or immune-modifying agents 36 and so on.In this study, competition events accounted for about 26.3% of the total deaths, accounting for a very high proportion, which indicated that traditional Cox proportional hazards model may not be suitable for analyzing 20 .At the same time, total competition risk model analysis also proved that there was a potential competitive relationship between Non-KSSD and KSSD (Fig. 5A).Further analysis clarified that gender and Spanish-Hispanic-Latino did not affect the KSSD of patients (Fig. 5B, F).Meanwhile, the KSSD was higher for patients with the following characteristics: the diagnosis age was less than 60 years (Fig. 5C), the diagnosis year was earlier than 2008 (Fig. 5D), Black race (Fig. 5E), metastatic or regional lesions (Fig. 5G), with non-multifocal lesions (Fig. 5H), and single (Fig. 5L).For single patients (never married), it may mean a higher likelihood of homosexual sexual behavior, which may lead to the rapid progression of AIDS-related KS.And for most types of cancer, the survival rate decreases with age.But literature had confirmed that the prognosis of young KS patients with AIDS-related diseases was worse than that of elderly patients during the pre-HAART period 37 , which was consistent with our research results.As for treatment, surgery and radiotherapy improved the KS-specific survival of patients, but chemotherapy did not (Fig. 5I-K).And radiotherapy had been considered to be one of the most effective treatments for localized KS 38,39 .
Considering the difference in the prognosis among patients with cutaneous KS, we attempted to incorporate independent prognostic factors into the multiple competitive risk model to construct a nomogram that could predict the 5-, 10-, and 15-year KSSD (Fig. 6A).Meanwhile, compared with the model based on Cox proportional hazard regression (Fig. 6B), it had a higher C index (0.709 vs.0.625).As shown in Fig. 6, when predicting the prognosis of the same patient (Patient ID = 35767086), it showed a lower 5-year, 10-year, 15-year KSSD than the nomogram based on the Cox proportional hazard regression.This was because the model based on the Cox proportional hazard regression did not eliminate the bias caused by the competing events, thus overestimating the cumulative incidence 20,40 .Also, the evaluation of nomogram proved to have relatively high discrimination and calibration both in the training and validation cohorts (Fig. 7A-F), and showed good clinical practicability within an appropriate threshold probability range (Fig. 7G-I), which could assist doctors to make a personalized clinical decision for different cutaneous KS patients.
This study had some limitations that should not be ignored.Firstly, as a retrospective study, there were inherent selection biases and uncontrollable confounding factors.Secondly, this research was based on the SEER database of the United States and lacked external verification, so the results may not be completely applicable to other countries and populations.Thirdly, the SEER database lacked a description of the schemes of radiotherapy and chemotherapy, and the classification of KS subtypes, limited effective variables and rough analysis may limit the effects of this study.Fourthly, such as ART, low CD4 count had been shown to significantly affect the prognosis of KS 41,42 .However, the contribution of these variables was not evaluated in our study.This omission may have influenced the results, thereby limiting the best management strategy.Therefore, further studies were needed to clarify the effects of these factors on the prognosis and provide guidance for the treatment of cutaneous KS.

Conclusion
This study was the first to analyze the prognosis of cutaneous KS based on a competing-risks model.This model revealed that radiotherapy and surgery could lower the KS-specific mortality, while chemotherapy and surgery could increase the OS of patients with metastatic and localized lesions, respectively.This validated nomogram provided individualized assessment and reliable prognostic prediction for cutaneous KS patients.

Figure 1 .
Figure 1.Flow diagram presenting the screening process in the SEER database.

Figure 2 .Figure 3 .
Figure 2. The regional distribution and year of diagnosis characteristics of all cutaneous KS cases from SEER database consisting of 18 population-based cancer registries; KS: Kaposi sarcoma.

Figure 6 .
Figure 6.The nomogram for predicting 5-, 10-and 15-year KSSD of cutaneous KS patients.KS-specific mortality was determined by adding up the scores of all variables and drawing a vertical line between the total point scale and the probability of death scale.(A) The nomogram based on the competing risk model; (B) The nomogram based on the Cox proportional hazard model.KSSD Kaposi sarcoma-specific death.

Figure 7 .
Figure 7. Discrimination and calibration validation of the nomogram model.(A-C) The 5-, 10-and 15-year calibration curves for the training cohort.The X-axes indicated the mean predicted probability of KSSD according to the prediction model, and the Y-axes represented the observed cumulative incidence of Kaposi sarcoma-specific death.The gray diagonal line indicated that the predicted value and the observed value were equal.(D-F) The area under the curve (AUC) of operating characteristics curve (ROC) for 5-, 10-, and 15-year-KSSD respectively.(G-I) Decision curve analysis of the nomogram.Horizontal green lines assumed no cases would experience the KSSD; Red lines assumed all cases would experience the KSSD.Blue lines represented the clinical net benefits across a range of threshold probabilities, within which applying the nomogram to predict the KSSD gained more benefit than the hypothetical treat-all or treat-none scenarios, KSSD Kaposi sarcomaspecific death.

Table 1 .
Demographics and clinicopathological characteristics of patients with cutaneous Kaposi sarcoma.KSSD Kaposi sarcoma-specific death, IQR interquartile range, NOS not otherwise specified, mm millimetre.

Table 3 .
Univariate and Multivariate competing risk model for KSSD in patients with Kaposi sarcoma.KS Kaposi sarcoma, KSSD Kaposi sarcoma-specific death, HR hazard ratio, CI confidence interval.Significant values are in bold.