التفاصيل البيبلوغرافية
| العنوان: |
DEVELOPMENT AND EXTERNAL VALIDATION OF NIMBLE, AN ARTIFICIAL INTELLIGENCE-BASED TOOL TO PREDICT PROGRESSION IN NON-MUSCLE INVASIVE BLADDER CANCER: A RETROSPECTIVE MULTI-INSTITUTIONAL COHORT STUDY. |
| المؤلفون: |
Kwong, Jethro C.C.1 (AUTHOR), Al-Daqqaq, Zizo1 (AUTHOR), Chelliahpillai, Yashan1 (AUTHOR), Lee, Soomin1 (AUTHOR), Kim, Kellie1 (AUTHOR), Chan, Amy2 (AUTHOR), Kuk, Cynthia2 (AUTHOR), Zlotta, Alexandre R.2 (AUTHOR), Ringa, Maximiliano3 (AUTHOR), Ali, Amna3 (AUTHOR), Feifer, Andrew3 (AUTHOR), Perlis, Nathan4 (AUTHOR), Lee, Jason Y.4 (AUTHOR), Hamilton, Robert J.4 (AUTHOR), Fleshner, Neil E.4 (AUTHOR), Finelli, Antonio4 (AUTHOR), Kulkarni, Girish S.4 (AUTHOR), Johnson, Alistair E.W.5 (AUTHOR) |
| المصدر: |
Urologic Oncology. Mar2024:Supplement, Vol. 42, pS53-S54. 2p. |
| مصطلحات موضوعية: |
*NON-muscle invasive bladder cancer, *BLADDER cancer, *ARTIFICIAL intelligence, *TUBERCULIN test, *COHORT analysis, *DECISION making |
| مصطلحات جغرافية: |
UGANDA, MISSISSAUGA (Ont.) |
| مستخلص: |
Several tools have been developed to predict the risk of progression in non-muscle invasive bladder cancer (NMIBC). However, they do not reflect current practice and perform poorly. We aimed to develop NIMBLE, an artificial intelligence (AI)-based tool, to better predict progression in contemporarily treated NMIBC patients. In addition, we sought to externally validate NIMBLE at two community-based hospitals to assess its generalizability in non-academic settings. A retrospective, multi-institutional cohort study was performed on all NMIBC patients treated at the University Health Network, Canada from Jan-2005 to Dec-2020 (n=1173); Credit Valley Hospital, Canada from Jan-2005 to Mar-2022 (n=754); and Mississauga Hospital, Canada from Jan-2005 to Mar-2022 (n=906). Patients were excluded if they had ≥T2 disease at initial diagnosis or < 1 year of follow-up. Primary outcome was time to progression, calculated from date of initial TURBT to date of first development of ≥T2, N+, or M+ disease. NIMBLE, based on a gradient-boosted survival forest, was trained on the University Health Network cohort, and externally validated on all other institutions. NIMBLE was compared against a LASSO Cox regression model using the same variables and the EAU prognostic risk groups. Model evaluation was based on c-index, calibration, and decision curve analysis. During a median follow-up of 55 months (IQR 25-96), 419 out of 2,833 patients (15%) developed progression. Median time to progression was 16 months (IQR 6-38). NIMBLE included the following predictors: age, sex, tumour history, stage, grade (WHO 2004/2016), concomitant carcinoma-in-situ, variant histology, lymphovascular invasion, number of tumours, tumour diameter, Bacillus Calmette-Guérin, and postoperative mitomycin C. In the training cohort, NIMBLE achieved a c-index of 0.78 (95% CI 0.74-0.83), compared to 0.75 (95% CI 0.68-0.80, p<0.001) and 0.75 (95% CI 0.72-0.77, p<0.001) for the Cox model and EAU risk groups, respectively. On external validation, NIMBLE achieved a c-index of 0.80 in both community cohorts compared to 0.75-0.77 for the other models (p<0.01). NIMBLE showed reasonable calibration for predicting 10-year progression for risks between 10-50%. At 1, 5, and 10 years, NIMBLE demonstrated the highest net benefit for clinically relevant thresholds between 15-30% (Figure 1). NIMBLE performed favourably compared to contemporary prediction tools in both academic and community settings. NIMBLE may be used to support patient counselling and better inform the need for treatment escalation in patients at high risk of progression. Ongoing work is being conducted to assess the generalizability of NIMBLE in larger NMIBC cohorts. [ABSTRACT FROM AUTHOR] |
| قاعدة البيانات: |
Academic Search Index |
