Artificial intelligence‐driven streamlining of prostate cancer multidisciplinary team recommendations in a tertiary NHS centre in the UK

• Published in: BJU international
• Main Authors: Khattak, Ahmed, Ruta, Danny, Patkar, Vivek, Popert, Rick, Makanjuola, Jonathan, Martin, Martha, Ledwaba‐Chapman, Lesedi, Dodgson, Kate, Oldroyd, Robert, Noel, Jonathan, Cathcart, Paul, Hughes, Simon, Challacombe, Ben, Josephs, Deborah, Enting, Deborah, Pintus, Elias, McCarthy, Ruth, Kazmi, Majid
• Format: Journal Article
• Language: English
• Published: England 01.07.2025
• Subjects: MedTech; automated triage; prostate cancer diagnostics; prostate cancer; multidisciplinary meeting; artificial intelligence
• DOI: 10.1111/bju.16845
• ISSN: 1464-4096; 1464-410X

To evaluate the effectiveness of a rules-based artificial intelligence (AI) clinical decision support system (CDSS) called the PROState AI Cancer-Decision Support (PROSAIC-DS) in streamlining the prostate cancer multidisciplinary team (MDT) pathway by identifying patients meeting standard of care (SoC) guidelines for reduced discussion in MDT meetings. This study consisted of two phases. Phase one involved a retrospective concordance analysis of 287 patients referred to the prostate MDT at King's College Hospital over a 2-year period. In phase two, a prospective analysis included 416 patients from Guy's Hospital over another 2-year period. Clinical treatment recommendations were independently reviewed by a panel of urologists and oncologists to establish a 'ground truth.' Concordance between the medical recommendations and those generated by the PROSAIC-DS was assessed. In phase one, the overall concordance between the clinicians' recommendations and the PROSAIC-DS was 92% (95% confidence interval [CI] 88.1-94.7%), compared to just 53% (95% CI 47-59%) with historic MDT outputs (P < 0.01). In phase two, the PROSAIC-DS achieved an 85.6% concordance (95% CI 81.6-88.9%) with the MDT recommendations for 355 evaluable cases (P < 0.01). Notably, using a machine learning-derived decision tree enabled the identification of 93 patients for streamlined management, demonstrating a 97.8% concordance in this subgroup (P < 0.01). The implementation of the PROSAIC-DS into the prostate cancer MDT pathway allowed 33.8% of patients to bypass MDT discussions with high treatment concordance. This study showcases the potential for AI-based solutions to improve clinical workflow and patient management in oncology, thus addressing the workload challenges faced by MDTs.