Predicting Hemoglobin Levels for ESA Dosing With Machine Learning Models

Patients receiving maintenance hemodialysis who develop anemia are commonly treated with erythropoiesis-stimulating agents (ESAs). According to Siwei Zhao and colleagues, prediction of patient-specific hemoglobin response to ESA therapy is a challenge.

During a poster session at the American Society of Nephrology Kidney Week 2023, the researchers reported on the use of machine learning techniques to predict hemoglobin response in patients on hemodialysis treated with ESAs. The poster was titled Machine Learning Approach to Predict Hemoglobin Levels for Erythropoietin Dosing in Hemodialysis Patients.

The model included patients undergoing hemodialysis at Sanderling Renal Services clinics in Nashville, Tennessee. Eligible patients were receiving intravenous (IV) ESA (Mircera®) and IV iron and had five or more hemoglobin measurements. The researchers collected data on ESA dose, iron dose, demographics, and clinical variables. The data were preprocessed to create a patient-specific biweekly dataset.

The study included a range of machine learning models, including long short-term memory networks, random forest, XGBoost, and support vector machine. Each model was implemented individually for each patient with K-fold cross-validation. Mean absolute percentage error (MAPE), (absolute percentage error between the predicted future hemoglobin and the observed future hemoglobin) was used to define the performance of models in predicting future hemoglobin levels.

The study cohort included 427 patients receiving hemodialysis. Of those, mean age was 65 years and 44% were women. Patients received an average of 93 mg IV iron per week, and mean Mircera dose of 103 mcg per month. Mean hemoglobin level was 10.1 mg/dL, and mean transferrin saturation was 32%.

Over the study cohort, the average MAPE for the model was 5.9% versus hemoglobin variation of 13.9%.

“Our results showed the promising performance of machine learning tools in predicting future hemoglobin levels within 6% of observed levels  in hemodialysis patients. Future studies should focus on refining these models with the goal of personalized ESA dosing to maximize on-target hemoglobin levels,” the researchers said.

Source: Zhao S, Yang JW, Zhang J, et al. Machine learning approach to predict hemoglobin levels for erythropoietin dosing in hemodialysis patients. TH-PO043. Abstract of a poster presented at the American Society of Nephrology Kidney Week 2023; November 2, 2023; Philadelphia, Pennsylvania.