Methods/Statistics
OptimalCutoff: an interactive web-based platform for ROC-based cutoff selection in epidemiological research Janos Tibor Fekete* Janos Tibor Fekete Fekete Semmelweis University, Dept. of Bioinformatics, Budapest, Hungary
Background: Receiver operating characteristic (ROC) curve analysis is widely used to evaluate the discriminatory performance of biomarkers and predictive models and to define decision-relevant cutoff thresholds. However, accessible platforms supporting advanced ROC frameworks beyond standard binary analysis are limited. We introduce OptimalCutoff, a web-platform for interactive binary, multiclass, and time-dependent ROC analysis with flexible cutoff selection strategies.
Methods: OptimalCutoff enables web-based ROC analysis of user-uploaded data, without the need for local statistical software or programming. The platform supports binary, multiclass, and time-dependent ROC frameworks and offers multiple cutoff selection strategies, including Youden optimization and sensitivity- or specificity-constrained thresholds. Interactive visualizations facilitate real-time exploration of ROC curves, operating points, and decision tradeoffs. As a methodological demonstration, data from the National Health and Nutrition Examination Survey (NHANES) were used to construct illustrative glycemic risk scores based on logistic regression models incorporating age, body mass index, and glycohemoglobin.
Results: In the NHANES demonstration dataset, an age- and BMI-based screening score showed good discrimination for prevalent diabetes (AUC = 0.80). Multiclass ROC analysis of glycemic status demonstrated strong overall discrimination (global multiclass AUC = 0.84), with distinct class-specific performance profiles. OptimalCutoff facilitated assessment of ROC behavior and cutoff sensitivity under alternative decision criteria. OptimalCutoff delivers figures and summary outputs to enhance reproducibility and reporting.
Conclusions: OptimalCutoff (https://optimalcutoff.com/) provides a web-based solution for ROC-based cutoff selection across binary, multiclass, and time-dependent frameworks, supporting transparent and context-aware decision making in epidemiologic research.