Welcome to powerROC
What is powerROC?
powerROC is a Python-based web application that helps researchers determine the sample size required to estimate the area under the receiver operating characteristic curve (AUROC) with a desired level of precision or to compare the AUROCs of two models with a desired level of power.
Why is sample size calculation important for evaluating prediction models?
In the design of a study, it is crucial to accurately calculate the sample size to effectively allocate resources, ensure statistical power for detecting true differences, and minimize Type II errors. For the evaluation of clinical prediction models, the TRIPOD+AI statement mandates reporting the process of determining the sample size and justifying its sufficiency in addressing the research question.
How can I use powerROC for sample size calculation?
- For evaluating a single prediction model
- For comparing two prediction models
• With access to a pilot test set (or a pilot dataset where you wish to adjust the prevalence),
click Sample size for comparing AUROCs: using a pilot test set
and calculate sample size using the resampling with replacement method for computing multiple DeLong p-values.
• Without access to a pilot test set,
click Sample size for comparing AUROCs: without a pilot test set
and calculate sample size by specifying joint two distributions (for cases and for controls) to compute multiple DeLong p-values via Monte Carlo simulations.
click Sample size to precisely estimate the AUROC
and calculate sample size by leveraging the asymptotic properties of Mann-Whitney U-statistics.
The research paper can be accessed at http://arxiv.org/abs/2501.03155. For the introductory tutorial and source code, visit our GitHub repository.
For any inquiries or further assistance, please contact François Grolleau at grolleau [ a t ] stanford.edu.
© 2024 PowerROC. All rights reserved.