HgbNet: Predicting Hemoglobin Level/Anemia Degree From Irregular EHR

Abstract

Predicting a patient’s hemoglobin level or degree of anemia using Electronic Health Records (EHRs) is a non-invasive and rapid approach. However, it presents challenges due to the irregular multivariate time series nature of EHRs, which often contain significant amounts of missing values and irregular time intervals. To address these issues, we introduce HgbNet, a model specifically designed to process irregular EHR data. HgbNet incorporates a NanDense layer with a missing indicator to handle missing values. Inspired by clinicians’ decision-making processes, the model employs three kinds of attention mechanisms to account for both local irregularity and global irregularity. We evaluate the proposed method using two real-world datasets across two use cases. HgbNet outperforms the best baseline results across all test scenarios, achieving an R2 score of $0.867~\pm $ 0.003 and $0.861~\pm $ 0.003 for hemoglobin level prediction, and an F1 score of $0.855~\pm $ 0.005 and $0.843~\pm $ 0.005 for anemia degree prediction under usecase 1 across two datasets. Additionally, we analyze the effect of the length of irregular time intervals on prediction performance and improve HgbNet’s performance at long intervals in usecase 2. These findings highlight the feasibility of estimating hemoglobin levels and anemia degree from EHR data, positioning HgbNet as an effective non-invasive anemia diagnosis solution that could potentially enhance the quality of life for millions of affected individuals worldwide.