Penalized Regression Spline Modeling of Dose-Response Functions and Its Application to Monitoring Malaria Drug Resistance in Drug Assays

• Activity Number: 352
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Samiha Sarwat*+ and Jaroslaw Harezlak and Clarissa Valim
• Companies: and Indiana Univ Fairbanks School of Public Health and Harvard School of Public Health
• Keywords: dose response ; penalized regression ; semi-parametric regression

Abstract

Dose-response assays describe the effect of changes in an organism growth caused by exposure to increasing drug concentration. The analysis of such experiments frequently relies on parametric sigmoidal (logistic) models. However, often dose-response data do not follow the specified parametric form and semi-parametric methods are necessary. We propose a semi-parametric approach, an extension of penalized regression spline method of Ruppert et al. (2003) that allows modeling of the smooth dose-response relationship with correlated data via linear mixed model representation. This method preserves the hierarchy of the technical and biological replicates while letting the data guide the mean model estimates. The quantities of interest, e.g. half maximal inhibitory concentration (IC50) and their properties are obtained. We investigate the proposed method in simulation studies assuming different underlying dose-response curves. Our model has multiple advantages over parametric models, reducing the bias in fitted response curves, IC50 estimates and gaining efficiency. We use our method to analyze data arising in the studies monitoring malaria drug resistance through ex vivo assays.