A Neurotransmitter-Based Model for Identifying Significant Depressive Symptoms in Patients with End-Stage Renal Disease

Abstract Depressive symptoms are common in patients receiving maintenance hemodialysis, but their identification is complicated by symptoms that overlap with uremia. We examined whether serum neurotransmitters, together with demographic and lifestyle variables, could identify significant depressive symptoms in a single-center, cross-sectional study of 250 adults with end-stage renal disease. Significant depressive symptoms were defined by a score greater than 7 on the depression subscale of the Hospital Anxiety and Depression Scale. Serum neurotransmitters and metabolites were measured by liquid chromatography-tandem mass spectrometry. Least absolute shrinkage and selection operator regression was used to select predictors, and eight machine learning classifiers were compared. Significant depressive symptoms were present in 86 participants (34.4%). The six selected predictors were age, household income, alcohol use, tea consumption, epinephrine, and 5-hydroxyindoleacetic acid. The random forest model achieved the highest area under the receiver operating characteristic curve (0.92), with 84% accuracy, 87% sensitivity, and 80% specificity in the study testing set. These findings support further evaluation of a biomarker-informed approach to identifying depressive symptoms in hemodialysis patients. External validation is required before clinical use.