In vivo Raman spectroscopy of skin in predicting mortality in patients with heart failure.

INTRODUCTION Heart failure (HF) is an age-related condition that complicates heart disease. Currently the number of patients with HF continues to increase, representing a significant burden on the healthcare system. Prognostication of patients with HF may be performed with modern spectroscopic approaches.

METHODS This study proposes to utilize conventional spontaneous Raman spectroscopy and autofluorescence for the in vivo analysis of skin tissues to create a prognosis for patients with HF. We collected skin spectral data from 160 HF patients. 29 patients died during the one-year observation period. After the preprocessing of spectral data, we proposed a classification model for the prediction of mortality. This model utilizes projection on latent structures combined with discriminant analysis (PLS-DA). Stability of the model was demonstrated during division of the data into training and test.

RESULTS Analysis of full spectral data provided only 55% accuracy in a one-year mortality prediction, while analysis of autofluorescence and Raman spectral data provides 66% and 70% accuracy, respectively. The combination of autofluorescence and Raman spectroscopy provides an accuracy of 74% (68% sensitivity and 80% specificity) and an ROC AUC (Receiver Operating Characteristic - Area Under Curve) of 0.80 for the prediction of 1-year mortality in patients with HF. The most important Raman bands for the prediction of one-year mortality appeared at 1086-1180, 1320, 1465, and 1770 cm-1.

CONCLUSION Raman spectroscopy could be a powerful tool for the prognosis of patients with HF; however, further studies on a larger cohort are required to demonstrate the applicability of the proposed spectral in vivo analysis.