Towards precision hematology: Machine learning-driven comparative feature importance analysis of hematologic parameters in newly diagnosed chronic myeloid leukemia vs healthy controls

Background: Chronic Myeloid Leukemia (CML) is a myeloproliferative neoplasm characterized by uncontrolled granulocytic proliferation and is often initially suspected based on peripheral blood smear findings. Utilizing machine learning to analyze routinely available Complete Blood Count (CBC) parameters and derived inflammatory indices may facilitate early identification of CML at the time of initial diagnosis. Objectives: The objective of this study is to evaluate the diagnostic value of routinely available Complete Blood Count (CBC) parameters and derived inflammatory indices for the early identification of chronic myeloid leukemia at the time of initial diagnosis, using machine learning–based models. Materials and methods: This study was conducted on 295 newly diagnosed cases of CML and 340 normal control samples. A total of 49 factors were subjected to study, including the following: variables included in the CBC of patients, inflammatory indices, and demographic data. Logistic regression analysis was performed to identify relevant variables, resulting in the selection of 22 features. Subsequently, multiple machine learning algorithms, including Random Forest (RF), Recursive Feature Elimination (RFE), Simulated Annealing (SA), Decision Tree (DT), K-Nearest Neighbor (KNN), and Xg-Boost (XGB), were applied to evaluate the diagnostic performance pf the selected features. Results: The findings of this study indicate that the factors most pertinent in initial diagnosis in comparison with normal control include the WBC count, the relative percentage of cells including neutrophils, monocytes, and lymphocytes, and a series of indicators related to RBC such as RBC count and RDW-CV, as well as the index of inflammatory NLR, and BLR and PDW. Conclusion: This study demonstrates that machine learning models based solely on routinely available CBC parameters and derived inflammatory indices can support the early identification of CML at the time of initial diagnosis. In addition to leukocyte-related variables, RBC-related and inflammatory indices provided complementary diagnostic information, highlighting their potential value in early-stage CML detection. The application of machine learning techniques could prioritize the development of more user-friendly dashboards to facilitate the diagnosis of CML at initial diagnosis.