Bladder Cancer (NMIBC) in a population-based cohort from Stockholm County with long-term follow-up; A comparative analysis of prediction models for recurrence and progression, including external validation of the updated 2021 E.A.U. model

Abstract

Non muscle invasive bladder cancer (NMIBC) has recurrence and progression rates of approximately 55-75% and 5-45% respectively. After diagnosis, risk stratification guides management decisions regarding surveillance, intravesical therapy or surgery. This prospective cohort of patients from Stockholm County is ideal for external validation of the current risk stratification models used in clinical practice.The cohort consisted of 395 patients diagnosed with bladder cancer across all the hospitals in Stockholm County between the years 1995-96, with up to 25 years follow up. All patients with pathologic Ta or T1 disease were included. Patients with muscle invasive disease (MIBC) referred for radical treatment at diagnosis were excluded. External validation of EORTC, CUETO and updated EAU Sylvester et al. (2021) models was done and multivariate Cox regression analysis was performed to generate hazard ratios for covariables of interest using both WHO '73 and WHO '04/16 pathological grade classifications.Overall Harrel's C-indices (CIs) for EORTC and CUETO models for recurrence were 0.66 and 0.63 respectively. The CIs for the EORTC, CUETO and EAU Sylvester et al. (2021) WHO '73 and '04/16 models for progression were higher at 0.82, 0.84, 0.83 and 0.83 respectively. All models tended to underestimate both recurrence and progression rates at 1 and 5 yrs. A simplified model devised to include only multifocality, tumor stage, size and grade performed with similar accuracy to all models for both recurrence and progression.Current risk stratification models are clinically useful but only moderately accurate across different patient populations, and the results of this study suggest a model using fewer variables is of similar accuracy to all models tested. In the future, research into the use of genomic classifiers will hopefully contribute to more accurate, modern risk stratification models.