Machine learning models in predictive factors for megaloblastic character of macrocytic anemia.

Background: Differentiating megaloblastic from non-megaloblastic macrocytic anemia usually relies on vitamin B12 and/or B9 assessment and bone marrow examination. In resource-limited regions, these diagnostic tools are not readily available. This study aims to develop a machine learning model to distinguish between these two forms of macrocytic anemia using commonly accessible clinical and laboratory parameters. Methods: A logistic regression model was developed using a dataset containing clinical features and routine laboratory data. Key predictive variables were identified through the SequentialFeatureSelector. The model was optimized using GridSearchCV and validated through stratified 5-fold cross-validation to ensure robust performance. Results: ESR, glossitis, and MCHC were the most influential variables in predicting non-megaloblastic anemia, with respective importance percentages of 23.11 %, 22.07 %, and 13.87 %. Other variables such as LDH (13.77 %), aregenerative status (13.42 %), and limb paresthesia (11.78 %) also contributed significantly to the model, while MCV had a comparatively lower importance (1.98 %). The model demonstrated a high discriminatory ability, with an area under the ROC curve (AUC) of 0.92 and an overall accuracy of 94 %. Learning curve analysis confirmed its stability and consistent performance across varying data sizes. Conclusions: By reducing dependence on specialized diagnostic tests and invasive procedures, this approach is particularly suited for resource-constrained healthcare settings.