Package: adjROC 0.4

adjROC: Computing Sensitivity at a Fix Value of Specificity and Vice Versa as Well as Bootstrap Metrics for ROC Curves

This software assesses the receiver operating characteristic (ROC) curve at adjusted thresholds, enabling the comparison of sensitivity and specificity across multiple binary classification models. Instead of comparing different models with varied cutoff values in their risk thresholds, all models can be compared at a fixed threshold of sensitivity, a fixed threshold of specificity, or the crossing point between sensitivity and specificity. If a threshold for specificity is given (e.g., specificity = 0.9), sensitivity and its confidence interval are computed, and vice versa. If the threshold for either sensitivity or specificity is not provided, the crossing point between the sensitivity and specificity curves is returned, along with their confidence intervals. For bootstrap procedures, the software evaluates the mean and CI bootstrap values for sensitivity, specificity, and the crossing point between specificity and sensitivity. This allows users to discern whether the performance of a model (based on adjusted sensitivity or adjusted specificity) is significantly different from other models. This software addresses the issue of comparing different classification models with varying predefined cutoff thresholds, which often leads to inconclusive results due to the fluctuating values of both sensitivity and specificity.

Authors:E. F. Haghish