Multicenter study of the diagnostic value of erythrocyte morphology assessment by the EH-2090 for differentiation of glomerular and non-glomerular hematuria

To address the limitations of microscopy and current automated instruments, we analyzed 11 red blood cell (RBC) morphological parameters generated by the EH-2090 analyzer. We aimed to identify dysmorphic RBCs associated with glomerular hematuria (GH), establish optimal diagnostic thresholds, and develop a predictive model for GH based on urinalysis features. A total of 597 hematuria samples were collected from patients across eight hospitals. The 11 RBC morphological parameters were compared with the final clinical diagnosis. The parameter yielding the largest area under the curve (AUC) was identified as being associated with GH, and its optimal diagnostic cutoff value was determined. Nine machine learning (ML) models were constructed. Receiver operating characteristic (ROC) and precision-recall (PR) curves were used to evaluate the models’ performances, and the SHapley Additive exPlanations (SHAP) method was used for visual analysis of the model with the optimal performance. The results confirmed that acanthocytes, jagged RBCs, annular RBCs, and other dysmorphic RBCs offered the greatest clinical value for diagnosing GH. When the combined proportion of these four dysmorphic RBC types exceeded 26% of the total erythrocytes, the sensitivity and specificity for diagnosing GH were 92.2% and 81.7%, respectively, with an AUC of 0.907. Among the constructed diagnostic models for GH, the random forest (RF) model demonstrated the best performance, with an AUC of 0.960. Confusion matrix analysis of the validation set showed a sensitivity of 96.6% and a specificity of 85.9%. The EH-2090’s identification of dysmorphic RBCs demonstrated high sensitivity and accuracy in distinguishing GH from non-glomerular hematuria (NGH), offering a rapid, automated, and standardized detection method.