Presence of an Artificial Intelligence-powered Predictive Biomarker Is Associated with a Poor Response to Intravesical Bacillus Calmette-Guerin but Not to Intravesical Sequential Gemcitabine/Docetaxel in Patients with High-grade Non-muscle-invasive Bladder Cancer

Abstract

Intravesical bacillus Calmette-Guerin (BCG) is considered first-line adjuvant therapy for high-risk or high-grade non-muscle-invasive bladder cancer (NMIBC). Recently, sequential intravesical gemcitabine and docetaxel (Gem/Doce) has emerged as a promising alternative to intravesical BCG. Biomarkers to select the optimal treatment regimen could facilitate clinical decision-making. The Computational Histologic Artificial Intelligence (CHAI) platform was previously used to develop an artificial intelligence-augmented histologic assay (CHAI biomarker) that identified patients with NMIBC at an increased risk of recurrence and progression events following BCG treatment. In this study, we assessed use of the CHAI biomarker among patients with treatment-naive high-grade NMIBC who received intravesical BCG or Gem/Doce. Among patients with the presence of the CHAI biomarker, those treated with BCG had a 24-mo high-grade recurrence-free survival (HG-RFS) rate of 56% (95% confidence interval [CI] 43-73%) and those treated with Gem/Doce had an HG-RFS rate of 90% (95% CI 79-100%; hazard ratio [HR] 5.4, 95% CI 1.6-18.3, p = 0.007). Among patients with an absence of the CHAI biomarker, those treated with BCG or Gem/Doce had no significant difference in HG-RFS (HR 1.3, 95% CI 0.6-2.6, p = 0.5). The interaction term between the CHAI biomarker and the treatment type was significant (p = 0.029), indicating an association between the biomarker and the clinical outcome that is dependent on the treatment received. This study suggests that the CHAI biomarker predicts which specific high-grade NMIBC patients are less likely to benefit from BCG and may benefit from alternative treatments including, potentially, Gem/Doce.

Keywords: Artificial intelligence; Bacillus Calmette-Guerin; Biomarker; Docetaxel; Gemcitabine; Machine learning; Non–muscle-invasive bladder cancer.

Fig. 1 –

High-grade recurrence free survival stratified by treatment and CHAI biomarker status. Kaplan-Meier estimates for HG-RFS are reported for patients stratified by treatment and biomarker status. Among BCG-treated patients, patients with the CHAI biomarker present performed worse than those with the CHAI biomarker absent (HR 2.0, 95% CI 1.1–3.6, p = 0.023). Among Gem/Doce-treated patients, there was no significant difference in HG-RFS based on biomarker status (HR 0.47, 95% CI 0.13–1.7, p = 0.2). Presence of the CHAI biomarker was associated with worse HG-RFS in BCG-treated patients but not in patients treated with Gem/Doce (HR 5.4, 95% CI 1.6–18.3, p = 0.007). Significant differences in HG-RFS were not seen in patients with the CHAI biomarker absent who were treated with either BCG or Gem/Doce (HR 1.3, 95% CI 0.6–2.6, p = 0.5). BCG = bacillus Calmette-Guerin; CHAI = Computational Histologic Artificial Intelligence; CI = confidence interval; Gem/Doce = gemcitabine and docetaxel; HG-RFS = high-grade recurrence-free survival; HR = hazard ratio.