Assessing the influence of smoking on inflammatory markers in bacillus Calmette Guérin response among bladder cancer patients: a novel machine-learning approach
Matteo Ferro, Octavian Sabin Tătaru, Giuseppe Fallara, Cristian Fiori, Matteo Manfredi, Francesco Claps, Rodolfo Hurle, Nicolò Maria Buffi, Giovanni Lughezzani, Massimo Lazzeri, Achille Aveta, Savio Domenico Pandolfo, Biagio Barone, Felice Crocetto, Pasquale Ditonno, Giuseppe Lucarelli, Francesco Lasorsa, Giuseppe Carrieri, Gian Maria Busetto, Ugo Giovanni Falagario, Francesco Del Giudice, Martina Maggi, Francesco Cantiello, Marco Borghesi, Carlo Terrone, Pierluigi Bove, Alessandro Antonelli, Alessandro Veccia, Andrea Mari, Stefano Luzzago, Raul Gherasim, Ciprian Todea-Moga, Andrea Minervini, Gennaro Musi, Francesco Alessandro Mistretta, Roberto Bianchi, Marco TOZZI, Francesco Soria, Paolo Gontero, Michele Marchioni, Letizia M. JANELLO, Daniela Terracciano, Giorgio Ivan Russo, Luigi Schips, Sisto Perdonà, Riccardo Autorino, Michele Catellani, Maria Chiara Sighinolfi, E. Montanari, Savino M. Di Stasi, Francesco Porpiglia, Bernardo Rocco, Ottavio De Cobelli, Roberto Contieri
Abstract
BACKGROUND: Approximately 70% of bladder cancer is diagnosed as non-muscle invasive (NMIBC) and inflammation is known to impact the oncological outcomes. Adjuvant intravesical BCG in intermediate/high risk can lower recurrence and progression. The efficacy of intravesical BCG can be impacted by smoking effects on systemic inflammation. METHODS: Our retrospective, multicenter study with data from 1.313 NMIBC patients aimed to assess the impact of smoking and the systemic inflammatory status on BCG response in T1G3 bladder cancer, using a machine-learning CART based algorithm. RESULTS: In a median of 50-month follow-up (IQR 41-75), 344 patients experienced progression to muscle invasive or metastatic disease and 65 died due to bladder cancer. A CART algorithm has been employed to stratify patients in three prognostic clusters using smoking status, LMR (lymphocytes to monocytes ratio), NLR (neutrophil-to-lymphocyte ratio) and PLR (platelet-to-lymphocyte ratio) as variables. Cox regression models revealed a 1.5-fold (HR 1.66, 95%, CI 1.20-2.29, P=0.002) and three-fold (HR 2.99, 95% CI 2.08-4.30, P<0.001) risk of progression, in intermediate and high risk NMIBC respectively, compared to the low-risk group. The model's concordance index was 0.66. CONCLUSIONS: Our study provides an insight into the influence of smoking on inflammatory markers and BCG response in NMIBC patients. Our machine-learning approach provides clinicians a valuable tool for risk stratification, treatment, and decision-making. Future research in larger prospective cohorts is required for validating these findings.