Predicting 90-Day Mortality After Radical Cystectomy for Bladder Cancer

A newly developed model for prediction of 90-day mortality after radical cystectomy for bladder cancer has outperformed currently available nomograms in a large contemporary cohort.

Results of the analysis were presented at the American Urological Association 2023 Annual Meeting.

Multiple models exist that use preoperative parameters for the prediction of 90-day mortality after radical cystectomy, including PROMETRICS2011, Morgan, and Taylor.

Ekaterina Laukhtina, MD, and colleagues used a machine learning-based approach to select valuable predictors of 90-day mortality after radical cystectomy after externally validating the performance of existing nomograms. Data from 3816 patients with bladder cancer who underwent radical cystectomy were retrospectively collected from 25 tertiary referral centers. Patients were randomly divided into development and validation cohorts. A machine learning-based variable selection approach was used to create a new multivariable prediction model.

The discriminatory ability of the model was quantified by the area under the curve (AUC) of receiver operating curves. After validation of the model, a nomogram was created, and decision curve analysis was used to evaluate its clinical net-benefit.

External validation of the PROMETRICS2011, Morgan, and Taylor nomograms showed a prediction accuracy of 90-day mortality of 70.3%, 67.2%, and 61.3%, respectively.

In the new model, machine learning selected ASA score, clinically metastatic disease, albumin, and  estimated glomerular filtration rate as the variables with high discriminatory power. The new model had an AUC of 80% in the development cohort compared with 65% in the validation cohort. In a sensitivity analysis that included only nonmetastatic patients, the AUC addressing 90-day mortality after radical cystectomy was 72%.

Researchers concluded that this new model can be used in a preoperative setting for patient counselling and risk stratification, although external validation is still needed before clinical use.