Adjusting ROC Curve for Covariates with AROC R Package

Abstract

The ability of a medical test to differentiate between diseased and non-diseased states is of vital importance and must be screened by statistical analysis for reliability and improvement. The receiver operating characteristic (ROC) curve remains a popular method of marker analysis, disease screening and diagnosis. Covariates in this field related to the subject’s characteristics are incorporated in the analysis to avoid bias. The covariate adjusted ROC (AROC) curve was proposed as a method of incorporation. The AROC R-package was recently released and brings various methods of estimation based on multiple authors work. The aim of this study was to explore the AROC package functionality and usability using real data noting its possible limitations. The main methods of the package were capable of incorporating different and multiple variables, both categorical and continuous, in the AROC curve estimation. When tested for the same data, AROC curves are generated with no statistical differences, regardless of method. The package offers a variety of methods to estimate the AROC curve complemented with predictive checks and pooled ROC estimation. The package offers a way to conduct a more thorough ROC and AROC analysis, making it available for any R user.

Neuroinflammation and Alzheimer’s Disease: A Machine Learning Approach to CSF Proteomics

“…The area under the curve (AUC) and AUC sensitivity and specificity confidence intervals were calculated by using 2000 bootstrap replicates. Age-adjusted AUC were instead computed with the R package AROC, according to the semiparametric Bayesian approach described by F. Machado e Costa and A. C. Braga [16]. The significance of z-score differences between groups was assessed by two-way analysis of covariance (ANCOVA) by assuming age as a covariate.…”