Ann Yellowlees, Francis Bursa, and Kelly J. Fleetwood are affiliated with Quantics, in Edinburgh, Scotland, United Kingdom. Sue Charlton is affiliated with Public Health England (PHE), in Porton Down, Salisbury, United Kingdom. Karie J. Hirst, Ronggai Sun, and Peter C. Fusco are with PharmAthene, Inc., in Annapolis, Maryland. PCF is also affiliated with VaxInnate Corporation, in Cranbury, New Jersey
Auto-generated profile, based on publicly available information
Peter C. Fusco
Ann Yellowlees, Francis Bursa, and Kelly J. Fleetwood are affiliated with Quantics, in Edinburgh, Scotland, United Kingdom. Sue Charlton is affiliated with Public Health England (PHE), in Porton Down, Salisbury, United Kingdom. Karie J. Hirst, Ronggai Sun, and Peter C. Fusco are with PharmAthene, Inc., in Annapolis, Maryland. PCF is also affiliated with VaxInnate Corporation, in Cranbury, New Jersey
Outliers within a bioassay are not uncommon, especially with animal models. Ordinary linear regression is sensitive to outliers; even a single outlier may compromise parallelism and other assay suitability criteria, skew the estimate of relative potency, or inflate its uncertainty. In this article, robust regression, which is less sensitive to outliers, is considered as an alternative. A comparison of ordinary linear with robust regression was conducted for 32 immunopotency assays; we also conducted a simulation study. Under ordinary linear regression, outliers flattened the slopes and inflated their variance. Under robust regression, slopes were less sensitive to outliers. Assay failures due to lack of dose response were reduced from 9% to 1%, failures due to nonparallelism were reduced from 18% to 2%, and the precision of the relative potency estimate improved by 9%. We conclude that robust regression is a more appropriate approach for bioassay data containing outliers than ordinary linear regression or excluding outliers.