The present study included 38 patients with sarcomatoid carcinoma of the bladder and 76 patients with non-sarcomatoid bladder cancer who were treated at the Department of Urology, The Affiliated Hospital of Qingdao University (Qingdao, China), between August 2010 and May 2021. The inclusion criteria were as follows: i) pathologically diagnosed surgical specimens; and ii) pT staging ≥T1. Patients with carcinosarcoma and sarcoma, which are two pathological types that differ from sarcomatoid carcinoma, were excluded. All patients diagnosed with other cancer types at the same time were excluded. The patients underwent radical cystectomy or transurethral resection, performed by experienced urologists, and volunteered to participate in this study. The Affiliated Hospital of Qingdao University Ethics Committee approved the study (approval number, AHQU-MAL 20210110). All methods were performed in accordance with the relevant guidelines and regulations, such as the Declaration of Helsinki.

Patient details, such as age at diagnosis, sex, body mass index, smoking status, alcohol consumption, diagnosis of diabetes mellitus, presence of hypertension, type of surgery, and documentation of adjuvant regimens of chemotherapeutic agents and immune checkpoint inhibitors, were obtained from the medical records or telephone follow-up. Imaging data were reviewed to determine whether distant metastases were present at the time of surgery. The pathology reports included immunohistochemical information, proportion of sarcomatoid components, pathological T stage, regional lymph node invasion, tumour necrosis, lymphovascular invasion and vascular tumour thrombus. Pathological staging was on the basis of the eighth edition of the TNM staging criteria of the Union for International Cancer Control (20). Lymphovascular invasion implied the invasion of endothelium-lined spaces of microvessels and lymphatics (21). The subgroup analysis was performed, and the patients were divided into two groups, namely, muscle-invasive BC and non-muscle-invasive BC, according to whether the tumour invaded the muscle.

Patients were followed up every 3 months for 2 years and annually thereafter. Recurrence or metastasis was detected using computed tomography and/or magnetic resonance imaging. The primary outcomes were cancer-specific survival (CSS) and overall survival (OS), defined as death from BC and all causes, respectively, or the end of follow-up. Another outcome was recurrence-free survival (RFS), defined as disease recurrence, newly detected metastasis or death from all causes.

To balance the confounders of sarcomatoid and non-sarcomatoid BC, propensity score matching was applied. The logistic regression model was performed according to age, sex and pathological T stage. The matching algorithm was 1:2 nearest neighbour matching with a 0.1 calliper width.

The Wilcoxon rank-sum test, χ² test and Fisher's exact test were used to compare the continuous and categorical variables of the patients. The Kaplan-Meier method was used to obtain survival rates and the log-rank test was used to compare survival curves. The association between the sarcomatoid components and other variables and the survival outcomes were calculated by Cox proportional hazard regression models and summarised using hazard ratios (HRs) and 95% confidence intervals (CIs). Collinearity diagnosis was used to test for multi-collinearity. The optimal cut-off value of the proportion of sarcomatoid components for prognostic value was determined through the receiver operating characteristic curve. Statistical analysis was performed using R 3.5.0 software (R Foundation) and SPSS version 26.0 (IBM Corp.). A two-sided P-value of <0.05 was considered to indicate a statistically significant difference.

A total of 38 patients with sarcomatoid carcinomas of the bladder were identified from the hospital database. Another 76 patients with non-sarcomatoid BC were selected as the control group after matching by propensity score. The median follow-up time was 46 months. The patients with sarcomatoid and non-sarcomatoid carcinoma of the bladder were mostly male (76 vs. 78%, respectively; P=0.875). The median age at diagnosis was 72 years (range, 65–79 years) and 71 years (range, 64–76 years), respectively (P=0.658). The majority of patients with stage T1 BC underwent transurethral resection of the bladder (88 vs. 84%, respectively; P>0.999). By contrast, most patients with muscle-invasive BC underwent radical cystectomy (82 vs. 96%, respectively; P=0.090). Muscle-invasive BC accounted for the majority of sarcomatoid and non-sarcomatoid carcinomas (58 vs. 58%, respectively; P=0.921) (data not shown). The proportion of patients with regional lymph node invasion (11 vs. 11%, respectively; P>0.999) and lymphovascular invasion (13 vs. 26%, respectively; P=0.109) was low. A sarcomatoid component proportion of >50% was observed in 45% of the patients with sarcomatoid carcinoma. A single patient with sarcomatoid carcinoma used tyrosine kinase inhibitor anlotinib after the discovery of lung metastases (data not shown). Patient characteristics after matching were not significantly different and are listed in Table I.

A total of 11 patients with sarcomatoid carcinoma and 9 with non-sarcomatoid carcinoma died of cancer during the follow-up. Significant differences were found between the sarcomatoid and non-sarcomatoid carcinoma groups in terms of 1-year CSS rate (80.2 vs. 100%, respectively), 3-year CSS rate (73.4 vs. 96.4%, respectively) and 5-year CSS rate (69.4 vs. 86.9%, respectively) (log-rank P=0.008) (Fig. 1A). After stratification by depth of invasion, non-muscle-invasive BC with sarcomatoid carcinoma components did not confer a worse CSS rate (5-year CSS rate: 100 vs. 91.8%, respectively; log-rank P>0.05) (Fig. 1C). Among the patients with muscle-invasive BC, sarcomatoid carcinomas had a significantly worse median CSS (38 vs. >120 months, respectively), with a 5-year CSS rate of 39.1% compared with 81.8% for non-sarcomatoid carcinomas (log-rank P<0.001) (Fig. 1E). A total of 15 patients with sarcomatoid BC and 12 with non-sarcomatoid BC experienced recurrence. The common metastatic sites of sarcomatoid carcinoma of the bladder were the bones (n=4), lungs (n=3), liver (n=2) and pelvic cavity (n=2). The 5-year RFS rates of the sarcomatoid and non-sarcomatoid carcinoma groups were 64.1 and 83.6%, respectively (log-rank P=0.001) (Fig. 1B). The 5-year RFS rates with and without sarcomatoid differentiation were 93.8 and 96.2%, respectively, in the non-muscle-invasive subgroup (Fig. 1D). The patients with sarcomatoid carcinoma had a worse median RFS time (11 vs. >120 months) in the muscle-invasive subgroup (Fig. 1F).

On multivariate analysis, larger tumour diameter, pT stage ≥T2, regional lymph node invasion and sarcomatoid carcinoma were independent prognostic factors for cancer-specific death. pT stage ≥T2, regional lymph node invasion and sarcomatoid carcinoma increased the risk of disease recurrence in the patients with BC. Independent predictors of OS were BMI, tumor size, pT stage and sarcomatoid carcinoma. Multivariate analysis showed that sarcomatoid carcinoma of the bladder was significantly associated with worse CSS (HR, 6.19; 95% CI, 2.31-16.6; P<0.001), RFS (HR, 4.44; 95% CI, 2.03-9.71; P<0.001) and OS (HR, 3.04; 95% CI, 1.39-6.66; P=0.005) (Table II).

Among the patients with sarcomatoid carcinoma, 11 had pure sarcomatoid carcinomas and 27 had mixed component carcinomas (sarcomatoid and non-sarcomatoid carcinoma). No significant difference in pathological stage (P=0.5), CSS and RFS was observed between the two groups, but both groups had higher CSS and RFS rates than the non-sarcomatoid group (Fig. S1A and B). According to the receiver operating characteristic curve, 45% sarcomatoid components was identified as the optimal cut-off score (Fig. S1C). The area under the curve was 0.68. Finally, 50% was used for the comparison of sarcomatoid components to achieve higher specificity. No significant differences in characteristics were found between patients with a sarcomatoid component >50% and patients with a sarcomatoid component ≤50% (Table SI), with the exception that tumours were larger in patients with a sarcomatoid component >50% (4 vs. 3 cm; P=0.009). The 5-year CSS and RFS rates for patients with a sarcomatoid component >50% were lower than those for patients with a low proportion (≤50%) of sarcomatoid component and patients with non-sarcomatoid BC (CSS: 53.5 vs. 80.6 vs. 89.2%, respectively; log-rank P<0.001; Fig. 2A; RFS: 47.4 vs. 78.5 vs. 83.6%, respectively; log-rank P<0.001; Fig. 2B). In the subgroup of muscle-invasive BC, patients with a sarcomatoid component >50% similarly had worse 5-year CSS and RFS rates (CSS: 22.9 vs. 37.9 vs. 81.8%, respectively; log-rank P<0.001; Fig. 2E; RFS: 23.3 vs. ns (indicating <5 years of follow-up) vs. 71.6%, respectively; log-rank P<0.001; Fig. 2F). In the non-muscle-invasive BC subgroup, in the >50% and ≤50% sarcomatoid component, and non-sarcomatoid groups, there was no significant difference in CSS (100.0 vs. 100.0 vs. 96.0%, respectively; log-rank P=0.3; Fig. 2C) but there was a significant difference in RFS (85.7 vs. 100.0 vs. 96.2%, respectively; log-rank P=0.003; Fig. 2D). Multivariate analysis showed that a sarcomatoid component >50%, pT stage ≥T2 and tumour size >3.5 cm predicted cancer-specific death. Independent predictors of recurrence were a higher proportion of sarcomatoid component, higher T stage, regional lymph node invasion and larger tumour size. Tumour T stage, tumour size, and Eastern Cooperative Oncology Group score were predictors of OS (Table III). No multi-collinearity was observed between variables in all regression analyses.