Potential of surface-enhanced Raman spectroscopy of blood serum in predicting mortality in patients undergoing maintenance hemodialysis

BACKGROUND: Predicting outcomes in chronic kidney disease remains challenging in modern medicine. It may be addressed using stratification systems based on biomarkers, including metabolic, electrolyte, inflammatory, and instrumental indicators. AIM: This study aimed to assess the prognostic value of surface-enhanced Raman spectroscopy of blood serum in evaluating all-cause mortality in hemodialysis patients with end-stage chronic kidney disease. METHODS: This prospective study included 58 patients of both sexes, aged 33–73 years (mean age: 57.0 ± 12.9 years) on maintenance hemodialysis. Over the 3-year follow-up, 13 deaths were recorded. An additional comparison group was formed, comprising 75 individuals (mean age: 51.33 ± 13.12 years; p 0.01) with estimated glomerular filtration rate corresponding to chronic kidney disease stages I–IIIa, to identify spectral characteristics associated with the mortality phenotype. According to current criteria, chronic kidney disease is diagnosed based on persistent signs of renal dysfunction, including a specific estimated glomerular filtration rate level, present for ≥3 months. However, the duration of asymptomatic stages of chronic kidney disease cannot be determined. Multivariate analysis was used to evaluate the statistical association between serum spectral characteristics and survival in 26 hemodialysis patients. To develop the prognostic model, least squares discriminant analysis was applied, which is a machine learning technique used for classification. RESULTS: Data of the cohort of patients undergoing maintenance hemodialysis was analyzed: 13 individuals who died within 3 years following blood sampling and 5 groups of 13 individuals each, randomly selected from the remaining 45. Each group was formed independently. The model was tested over five iterations and the results averaged. The most prognostically significant spectral peaks were 731, 839, 1240, 1391, and 1578 cm-1. The model demonstrated an 83% sensitivity, 79% specificity, and 81% accuracy and an area under the ROC curve of 0.86. Notably, two of the five frequencies significant for survival prediction overlapped with those characteristic of creatinine and urea (637, 724, 1001, 1095, 1238, and 1393 cm−1), which enable differentiation between stages I–IIIa of chronic kidney disease and end-stage renal disease, yielding a 71% sensitivity, 95% specificity, and 83% overall accuracy. CONCLUSION: Combined surface-enhanced Raman spectroscopy of blood serum and mathematical modeling presents high predictive accuracy with minimal labor input.