Fig. 2: Predictive model of metastasis based on dermatoscopic features, performance metrics of three models and classification of patients according to Model 1 prediction. | Nature Communications

Fig. 2: Predictive model of metastasis based on dermatoscopic features, performance metrics of three models and classification of patients according to Model 1 prediction.

From: Predictive models of melanoma metastasis based on dermatoscopy in an international retrospective human reader study

Fig. 2

a. Forest plot showing Odds Ratios (ORs) (squares) and 95% Confidence Intervals (CI) (error bars) for the prediction of metastasis based on predictors entered the last step of multivariable analysis in the total cohort (n = 524 patients). b. Boxes show the Area under the curve (AUC) values from 5-fold cross-validation in training set (n = 425) and test set (n = 99) for the prediction of metastasis. The middle line denotes the AUC value and the upper and lower parts demonstrate the upper and lower 95%CI, respectively. Dots indicate AUCs for individual folds. Two-sided p-values derived from DeLong’s test and stars (*) denote pairwise comparisons of AUC values between three models [training set: M1/M2 p = 0.20, M1/M3 p = 0.52 M2/M3 p = 0.10, test set: M1/M2 p = 0.51, M1/M3 p = 0.72, M2/M3 p = 0.31]. Correction for multiple comparisons was not applied. c. Accurate (TP: true positive, TN: true negative) and false predictions (FP: false positive, FN: false negative) according to status of parameters included in model 1 by AJCC stage in training set (n = 425). The order of variables in y axis is Pigmentation (Pigm.) – Ulceration (Ulcer.) - Regression (Regres.) - Blue-white veil (BWV). H: high, M/L: moderate/low, A: absent. Source data are provided as a Source Data file. (Ref.: reference level).

Back to article page