Development and external validation of a model to predict recurrence in patients with non-muscle invasive bladder cancer

Abstract

Background: Most patients initially diagnosed with non-muscle invasive bladder cancer (NMIBC) still have frequent recurrence after urethral bladder tumor electrodesiccation supplemented with intravesical instillation therapy, and their risk of recurrence is difficult to predict. Risk prediction models used to predict postoperative recurrence in patients with NMIBC have limitations, such as a limited number of included cases and a lack of validation. Therefore, there is an urgent need to develop new models to compensate for the shortcomings and potentially provide evidence for predicting postoperative recurrence in NMIBC patients.

Methods: Clinicopathologic characteristics and follow-up data were retrospectively collected from 556 patients with NMIBC who underwent transurethral resection of bladder tumors by electrocautery (TURBT) from January 2014 to December 2023 at the Affiliated Hospital of Zunyi Medical University and 167 patients with NMIBC who underwent the same procedure from January 2018 to April 2024 at the Third Affiliated Hospital of Zunyi Medical University. Independent risk factors affecting the recurrence of NMIBC were screened using the least absolute shrinkage and selection operator (Lasso) and Cox regression analysis. Cox risk regression models and randomized survival forest (RSF) models were developed. The optimal model was selected by comparing the area under the curve (AUC) of the working characteristics of the subjects in both and presented as a column-line graph.

Results: The study included data from 566 patients obtained from the affiliated hospital of Zunyi Medical University and 167 patients obtained from the third affiliated hospital of Zunyi Medical University. Tumor number, urine leukocytes, urine occult blood, platelets, and red blood cell distribution width were confirmed as independent risk factors predicting RFS by Lasso-Cox regression analysis. The Cox proportional risk regression model and RSF model were constructed based on Lasso, which showed good predictive efficacy in both training and validation sets, especially the traditional Cox proportional risk regression model. In addition, the discrimination, consistency, and clinical utility of the column-line graph were assessed using C-index, area under the curve (AUC), calibration curve, and decision curve analysis (DCA). Patients at high risk of recurrence can be identified early based on risk stratification.

Conclusion: Internal and external validation has demonstrated that the model is highly discriminative and stable and can be used to assess the risk of early recurrence in NMIBC patients and to guide clinical decision-making.

Keywords: Lasso-Cox regression; nomogram; non-muscle invasive bladder cancer; random forest; recurrence.

Figure 1

The flow diagram shows the sample selection.

Figure 2

Lasso regression analysis was used to select screening variables. (A) The variation characteristics of variable coefficients; (B) The optimal λ-value process is screened by a 10-fold cross-validation process.

Figure 3

RFS analysis in all 556 patients with ovarian cancer was based on the number of tumors (A), PLT (B), RDW (C), LEU (D) and BLD (E).

Figure 4

Nomogram ROC curves for 1-, 3-, and 5-year RFS for the training set and internal validation set. (A) training set; (B) internal validation set. ROC, receiver operating characteristic; AUC, area under the curve; RFS, recurrence-free survival.

Figure 5

Calibration curves of the training set for 1-year (A), 3-year (B), and 5-year (C) and the internal validation set for 1-year (D), 3-year (E), and 5-year (F). The blue curve is the prediction curve, and the dashed line is the reference curve.

Figure 6

Decision curve analysis of the training set for 1-year (A), 3-year (B), and 5-year (C) and the internal validation set for 1-year (D), 3-year (E), and 5-year (F).

Figure 7

The error rate of a random survival forest. (A) training set; (B) validation set.

Figure 8

The recurrence risk analysis of NMIBC is based on a random survival forest. (A) Out-of-bag variable importance ranking. (B) correlation heat map.

Figure 9

The confusion matrix is (A) the training set and (B) the validation set.

Figure 10

Random forest ROC curves of the 1-, 3-, and 5-year RFS for the training set and validation set (A) training set; (B) validation set.

Figure 11

Nomogram for predicting NMIBC patients at 1-, 3-, and 5-years. NMIBC: non-muscle-invasive bladder cancer.

Figure 12

The ROC curves of an external validation cohort were employed to forecast the 1-, 3-, and 5-year rates of RFS in patients with NMIBC.

Figure 13

Calibration curves and clinical decision curves at 1, 3, and 5 years for the external validation set. Calibration curves: 1-year (A), 3-year (B), and 5-year (C); clinical decision curves: 1-year (D), 3-year (E), and 5-year (F).