Using Local Likelihood Density Estimation for Bias Correction at and Near a Boundary in the Context of Diagnostic Test Evaluation

• Activity Number: 281
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Norberto Pantoja*+ and Mary E Thompson
• Companies: FDA and University of Waterloo
• Keywords: Density estimation ; Density with discontinuity ; Diagnostic tests

Abstract

An important component in the evaluation of diagnostic tests is the estimation and visualization of the distributions (or probability density functions) of the diagnostic test results for the diseased and non-diseased subjects in a study population. The intuitive appeal of kernel density estimation makes it an attractive candidate to obtain nonparametric estimates of such probability density functions. However, in spite of the advantages presented by standard kernel density estimation techniques, a problem of increased bias at and near a known boundary is present in certain scenarios. We develop a local likelihood estimation method that is suitable for a density which is believed to have a discontinuity at and near a known boundary. We produce an estimator that corrects boundary bias. Potential applications in the context of diagnostic test evaluation are discussed.