Answers are grounded in this paper and cite their sources.
Anemia is a common complication of chronic kidney disease (CKD) and often requires treatment with erythropoietin (EPO). However, adjusting the EPO dosage remains a complex and subjective task. In this work, we proposed the use of machine learning (ML) methods to predict and adjust EPO doses for CKD patients, aiming to develop accurate and interpretable prediction tools that clinicians can use for decision support. We hypothesized that a set of routinely collected clinical variables could be effectively represented by a low-dimensional set of latent variables, capturing the essential clinical information while reducing complexity. To this end, we explored manifold learning (MnL) algorithms for EPO dose adjustment. We generated a retrospective dataset which comprised 146 CKD patients undergoing EPO therapy for an average duration of 162 . 9 ± 122 . 2 weeks. A contractive autoencoder was used to generate latent variables summarizing the clinical data for each patient. Results demonstrated that autoencoders and two-dimensional manifolds can effectively capture the underlying clinical patterns relevant to EPO dosing. In the resulting latent space, EPO dose prediction achieved an error of 797 . 88 ± 929 . 50 EPO units, with the model showing good generalization capabilities. We conclude that MnL-based models operating in low-dimensional latent spaces can adequately represent clinical data for guiding EPO treatment in CKD patients. These findings have positive implications for the development of ML-based decision support tools in the management of CKD-associated anemia.
No takes yet. Share an insight, caveat, or question.
Gil-Casares-Casanova et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: