Advancing personalized medicine in primary bladder adenocarcinoma (PAB): A machine learning study.

Abstract

e16617 Background: PAB is a rare and aggressive tumor with a poor prognosis. Comprising 0.5-2% of bladder cancers. This poses a significant diagnostic challenge and requires early treatment. We are the first to utilize machine learning (ML) to identify prognostic factors with the aim of improving prognosis and survival. Methods: Data were obtained from the SEER database (2004-2021). Patients who met any of the following criteria were excluded: diagnosis not confirmed by histology; previous history of cancer or other concurrent malignancies; or unknown data. To identify prognostic variables, we conducted Cox regression analysis and constructed prognostic models using ML algorithms to predict the 5-year survival. Patient records were randomly divided into training (70%) and validation (30%) sets. A validation method incorporating the area under the curve (AUC) of the receiver operating characteristic curve was used to validate the accuracy and reliability of the ML models. We also investigated the role of multiple therapeutic options using Kaplan-Meier survival analysis. Results: A total of 2,505 patients were included. Most patients (57.4%) were male and 76% were white. The most frequent tumor sites were the urachus (23.7%) and dome of the bladder (20.8%). The median patient age was 61 years, and the median tumor size was 4 cm. Tumor staging was 18.2% at T2, 21.9% at T3, 49.7% had no nodal involvement, and 55.6% showed no metastasis. The treatments administered included surgery (94.4%), chemotherapy (32.4%), and radiation (11.5%). The highest 5-year OS rates were observed in the ureteric orifice and urachus (62.5% and 56.7%, respectively), whereas the bladder neck had the lowest (15.7%). Patients without metastasis had a significantly higher OS rate (45.5%) than those with metastasis (5.3%). Radiation was associated with lower survival rates (OS, 18.1%; CSS, 25.4%) than those who did not receive radiation (OS, 44.3%; CSS, 51.5%). Multivariate Cox regression analysis identified metastasis, radiation, and older age as poor prognostic factors, whereas surgery and male gender were good prognostic factors. The Random Forest and MLP classifiers were the most accurate models. The ML models identified the primary site as the most significant prognostic factor, followed by TNM stage and sex. The performance metrics for all the ML algorithms are summarized in Table. Conclusions: This study highlights the potential of ML to advance personalized medicine. Further investigation is required to clarify the roles of chemotherapy and radiation, as radiation is linked to worse outcomes and chemotherapy showed no significant benefit. ML Model Accuracy Precision Recall F1 score AUC LR 72.5% 72.5% 97.6% 84% 0.684 KNN 76.7% 80% 90.5% 84.9% 0.841 RFC 82.2% 84.7% 92.% 88.2% 0.895 GBC 73.1% 73.7% 97.8% 84.1% 0.783 MLP 77.8% 78.4% 95.7% 86.2% 0.817