Development of a miRNA signature highly associated with cGVHD. (A) Based on the relative expression of the 5 miRNAs miRNA-365a-3p, miRNA-148a-3p, miR-122-5p, miR-378a-3p, and miR-192-5p, we performed a hierarchical cluster analysis (Euclidean correlation and complete linkage). This cluster analysis clearly demonstrates that high serum levels of these miRNAs were associated with high frequency of patients with cGVHD. Using LASSO analysis, we identified miR-365a-3p, mir-148a-3p, and miR-378a-3p to be sufficient for the prediction for the diagnostic model. We performed an internal fivefold crossvalidation for our material, and ROC curves were compared. (B) ROC curve for miR-378a-3p without crossvalidation (red) and with crossvalidation (blue); the AUCs were 0.732 and 0.719, respectively. (C) ROC curve for miR-148a-3p without crossvalidation (red) and with crossvalidation (blue); the AUCs were 0.732 and 0.719, respectively. (D) ROC curve for miR-365a-3p without crossvalidation (red) and with crossvalidation (blue); the AUCs were 0.720 and 0.692, respectively. (E) Based on a linear regression model to improve the diagnostic model by use of the 3 miRNAs miR-378a-3p, miR-148a-3p, and miR-365a-3p, the highest AUC was reached. ROC curve without crossvalidation (red) and with crossvalidation (blue) are shown; the AUCs were 0.778 and 0.740, respectively.