4.1 Study population
14,135 patients diagnosed with UTUC between 2004 and 2015 were collected from the SEER database. Finally, 1580 eligible T1-2N0M0 patients with UTUC as the only primary tumor were included for further analysis as shown in Figure 1. 306 failure events and 302 competition events were observed in the overall cohort. Median follow-up period for the overall cohort and after PSM was 44 months and 41 months.
Baseline characteristics of patients between the RN and NSS groups before and after PSM were shown in Table 1. For the entire cohort, 1187 (75.1%) patients were treated with RN, and 393 (24.9%) were treated with NSS. Patients undergoing NSS were associated with increasing age and renal pelvis tumors compared with those receiving RN (p<0.001). After PSM, a cohort of 766 patients was generated, 383 patients in each group. 10 of 393 patients from the NSS group didn’t have the matched individuals from the RN group. The baseline characteristics were well balanced between the two groups in the PSM cohort (Table 1; Figure 2). 167 failure events and 163 competition events were recorded in the post-PSM cohort. G4 was the most common histologic grade before and after PSM, accounting for above 40% in both the overall cohort and the post-PSM cohort.
4.2 Differences of prognosis between the RN and NSS groups
As shown in Figure 3, in the overall cohort, patients in the NSS group were associated with poorer OS compared with those in the RN group (5-year OS rate: 55.6% vs. 65.1%, p<0.001; Figure 3. A). After stratifying patients according to locations of lesions, for patients with ureteral tumors, there was no significant difference in OS between the RN and NSS groups (453 and 284 patients, respectively; 5-year OS rate: 59.6% vs. 56.9, p=0.232; Figure 3. B). For patients with renal pelvic tumors, RN was associated with better OS compared with NSS (734 and 109 patients, respectively; 5-year OS rate: 68.6% vs. 52.8%, p=0.001; Figure 3. C).
In the post-PSM cohort, NSS seemed to be associated with poorer OS compared with RN, whose differences between them were almost significant (5-year OS rate, 62.8% vs. 55.2%, p=0.085; Figure 3. D). This trend still existed in patients with renal pelvic tumors, for which RN was almost associated with better overall survival compared with NSS (5-year OS rate 67.5% vs. 52.8%: p=0.055; Figure 3. E). For ureteral tumors, OS between the two groups showed no significant difference (5-year OS rate, RN vs. NSS: 60.8% vs. 56.3%, p=0.404; Figure 3. F).
Competing-risks models revealed that the NSS group had poorer CSS compared with the RN group (subdistribution hazard ratio/ SHR=1.39, 95% confidence interval/ CI 1.03-1.89, p=0.033; Figure 4. A) in the post-PSM cohort. When the post-PSM cohort was stratified by tumor locations, RN was associated with better CSS in patients with ureteral tumors (SHR=1.44, 95% CI 1.01-2.06, p=0.044; Figure 4. B). There was no significant difference in survival between RN and NSS for renal pelvic tumors (SHR=1.24, 95%CI 0.689-2.24, p=0.471; Figure 4. C).
4.3 Independent prognostic factors of OS and CSS in the overall cohort
Age at diagnosis was divided into two groups as “>78.5 years old” and “<78.5 years old” using X-tile software for further building prognostic models. The results of univariate and multivariate Cox analysis for OS in the overall cohort were listed in Table 2. Univariate Cox analysis demonstrated that some characteristics of patients, which were age groups, tumor locations (renal pelvic tumors or ureteric tumors), AJCC T stage, histologic grade, marital status, surgery methods, were associated with OS in the overall cohort. Further multivariate Cox analysis including statistically significant variates in univariate Cox analysis by the forward-stepwise model selection revealed that age at diagnosis, AJCC T stage, and histologic grade were independent prognostic factors for OS in the overall cohort (Table 2). The strongest predictor for OS was the age at diagnosis (hazard ratio/ HR=2.80, 95% CI, 2.38-3.22, p<0.001). Advance histologic grade (HR=1.38, 95% CI, 1.15-1.66, p=0.001) and T2 stage (HR=1.44, 95% CI, 1.22-1.70, p<0.001) had similar predictive strengths about poorer OS. Patients accepting NSS (HR=1.24, 95% CI, 1.04-1.48, p=0.0018) held poorer OS compared with patients accepting RN.
Competing-risks models based on the Fine-Gray models were used to assess CSS in the entire cohort. The results of univariate analysis for CSS in the overall cohort were shown in Table 3. Age groups, tumor location, AJCC T stage, histologic grade, and surgery methods were associated with CSS. Gender, race, tumor laterality, and marital status seemed not to be related to CSS. Statistically significant variables in the univariate analysis in the competing-risks models were included in multivariate analysis through forward-stepwise selection methods, which validated that older age at diagnosis, advanced AJCC T stage, advanced histologic grade, and NSS were independent predictors of CSS (Table 4). Age at diagnosis, AJCC T stage, and histologic grade hold a similar predictive power for competing risk of CSS. Compared with the younger group (age at diagnosis < 78.5 years old), the older group (age at diagnosis > 78.5 years old; SHR=1.82, 95% CI, 1.44-2.30, p<0.001) carried higher competing risks. Advanced AJCC T stages and advanced histologic grades were also strongly predictive for CSS. Patients with T2 tumors (SHR=1.88, 95% CI, 1.50-2.37, p<0.001) or patients with advanced histologic G3 or G4 tumors (SHR=1.83, 95% CI, 1.38-2.42, p<0.001) experienced increased competing events compared with patients with T1 tumors or patients with histologic G1 or G2 tumors. Although the predictive strength was not as strong as the other three factors, patients accepting NSS rather than RN also hold higher CSS risks (SHR=1.47, 95% CI, 1.15-1.88, p=0.002).
4.4 Developing nomograms for OS and CSS
As mentioned above, the final prognostic models for OS and CSS both included four factors, which were age at diagnosis, surgery methods, histologic grade, and AJCC T stage of tumors. The prognostic nomograms for OS (Figure 5) and CSS (Figure 6) using the overall cohort that integrated all these four significant independent factors were shown in Figure 5 and Figure 6. The nomogram for OS was based on multivariate Cox analysis to predict the 3-year and 5-year OS, while the nomogram for CSS was based on the competing risk model mentioned above to predict the 3-year and 5-year CSS. After adding up scores corresponding to each value and check against the bottom probability axis, the possibility of all-cause mortality or cancer-specific mortality was then got. The score of each value was shown in Table 5.
Then the overall cohort was used for internal validation of prognostic models. The nomogram for OS displayed a C-index of 0.672, as well as the nomogram for CSS, displayed a C-index of 0.643. Using 1000 resampled bootstrap data sets for cross-validation, the calibration plots for the two nomograms in the overall cohort were shown in Figure 7 and Figure 8. The Brier scores of corresponding models were also shown in the figures.