The Impact of Treatment Recommendation By Leukemia Artificial Intelligence Program (LEAP) on Survival in Patients with Chronic Myeloid Leukemia in Chronic Phase (CML-CP)

• Published in: Blood 2019-11, Vol.134 (Supplement_1), p.1642-1642
• Main Authors: Sasaki, Koji, Kantarjian, Hagop M., Jabbour, Elias, Ravandi, Farhad, Konopleva, Marina Y, Borthakur, Gautam M., Wierda, William G., Daver, Naval G., Takahashi, Koichi, Naqvi, Kiran, DiNardo, Courtney D., Montalban Bravo, Guillermo, Kanagal-Shamanna, Rashmi, Pemmaraju, Naveen, Issa, Ghayas C., Jain, Preetesh, Skinner, Jeffrey, Rios, Mary B., Pierce, Sherry A., Soltysiak, Kelly, Sato, Junya, Garcia-Manero, Guillermo, Cortes, Jorge E.
• Format: Article
• Language: English
• ISSN: 0006-4971; 1528-0020
• DOI: 10.1182/blood-2019-130148

The survival of patients with chronic myeloid leukemia in chronic phase (CML-CP) is approaching that of general population after the approval of tyrosine-kinase inhibitors (TKI), particularly in younger patients who achieve remission. The optimal frontline TKI therapy in older patients in the context of comorbidity remains unknown. The aim of this study is to develop the LEukemia Artificial intelligence Program (LEAP) for treatment recommendations for patients with CML-CP. From July 30, 2000 to November 25, 2014, 630 consecutive patients with newly diagnosed CML-CP were enrolled in frontline TKI therapy (imatinib 400 mg/day, imatinib 800 mg/day, nilotinib, dasatinib, and ponatinib). We included 101 social, demographic, clinical, chromosomal, and molecular variables such as the distance from home address to our institution, primary language, the European Treatment and Outcome Study (EUTOS) risk, the EUTOS long-term survival (ELTS) risk, and the severity of comorbidities by Adult Comorbidity Evaluation 27 (ACE-27). We developed an extreme gradient boosting decision tree model through ensemble learning after hyperparameter tuning. Hyperparameter optimization was calculated with Stampede2, a supercomputer located at Texas Advanced Computing Center, which was ranked as the 15th fastest supercomputer in June 2018. The extreme gradient boosting decision tree model was developed for the prediction of overall survival using only the training/validation cohort. We evaluated the final performance with the independent test cohort. A difference in hazard ratios of less than 0.005 between the best treatment option and alternative TKI therapy was considered as the LEAP recommendation. The test cohort was divided into the LEAP recommendation and the LEAP non-recommendation cohort by the LEAP recommendation. To confirm the association and causation of the LEAP recommendation with survival, we performed backward multivariate Cox regression, and inverse probability of treatment weighing (IPTW) to balance baseline difference of covariates. We calculated SHapley Additive exPlanations1 values to interpret the black box of the LEAP recommendation for the evaluation of the significance of variable for prediction. The whole cohort was randomly divided into a training/validation (N=504) cohort and a test cohort (N=126) at a 4:1 ratio (Figure 1). The training/validation cohort was used for 3-fold cross validation to develop the LEAP CML-CP model. The number of decision trees was 841