Background
Improvement in prediction models for progression to muscle invasive bladder cancer (MIBC) are needed to identify those patients who might benefit from immediate cystectomy in non-invasive disease.
Objective
To identify prognostic factors in patients with pT1 bladder cancer who are more likely to fail intravesical therapy and progress to MIBC and build a clinical decision-making tool for the selection of these patients.
Design, setting and participants
We retrospectively evaluated the pathological report of 1304 patients treated with TURB for pT1 bladder cancer. Median follow-up for the entire cohort was 56.6 months (IQR: 20.4 – 96).
Outcome measurements and statistical analysis
Univariate and multivariable competing risk regression models were used to assess the association of LVI and VH with progression to MIBC. We generated a nomogram for calculating the patient-specific probabilities of progression after 2 and 5 years. The predictive accuracy of the model was assessed by its discrimination and calibration. A split-sample internal validation procedure with 1000 resample bootstrapping was performed to establish that the model worked sufficiently among patients other than those whose data generated the model. Finally, a decision curve analysis (DCA) was applied to explore the clinical value of our newly derived model by increasing the net benefit over a realistic range of threshold probabilities.
Results and limitations
A total of 192 patients were diagnosed having VH (14.7%) and 158 (12.1%) LVI in the TURB specimen. Both factors were significantly associated with disease progression (VH: HR 8 [95%CI 5.9-10.9], p <0.001; LVI: HR 4.2 [95%CI 3.03-5.9], p <0.001). On multivariable analysis LVI and VH remained independently associated with disease progression (HR: 3.8, 95%CI: 2.4 – 6, p <0.001 and HR: 5.6, 95%CI: 3.7 – 8.6, p <0.001, respectively). Based on these findings, we integrated VH to the nomogram building model for the prediction of PFS as 2 and 5 years, which otherwise included concomitant Cis, ≥3cm and multifocal tumor. The standard model without VH and LVI had a C-index of 0.59. The addition of VH to this model significantly increase its discrimination by 16% (C-index 0.75; p < 0.001). On DCA our model led to superior outcomes for any decision associated with a threshold probability of progression between 5% and 60%. Limitations are inherent to the retrospective design of the study.
Conclusions
LVI and VH should be part of every prognostic tool, as they provided independent association with disease progression. They could be used for improving risk stratification and identifying those patients with pT1 disease who might benefit from an immediate cystectomy.