Evaluation of an optimal receiver operating characteristic procedure

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Statistical Modeling

Background:

• Receiver operating characteristic (ROC) curves are crucial for evaluating predictive genetic tests.
• The Lu-Elston method was proposed to optimize area under the curve (AUC) using summary data.
• Increased data sharing necessitates evaluating methods with individual-level genetic data.

Purpose of the Study:

• To investigate the performance of the Lu-Elston algorithm with individual-level genetic data.
• To compare the Lu-Elston method against standard ROC curve-building techniques.
• To assess the utility of the Lu-Elston method in genetic analysis.

Main Methods:

• Evaluation of the Lu-Elston method using individual-level genetic data.
• Comparison with standard multivariable logistic regression methods for ROC curve construction.
• Analysis of the Genetic Analysis Workshop 16 rheumatoid arthritis dataset.

Main Results:

• The Lu-Elston method showed minimal advantage over standard methods when individual data were accessible.
• Standard multivariable logistic methods provided greater ease in comparing nested models.
• The Lu-Elston approach encountered difficulties in incorporating covariates and performing likelihood ratio tests.

Conclusions:

• The Lu-Elston method's advantage is limited when individual genetic data are available.
• Standard logistic regression methods are more versatile for complex genetic analyses, including covariate incorporation.
• The Lu-Elston method may not be superior to established techniques for predictive genetic test evaluation.