A Comparison of Statistical Methods to Estimate Descriptive Statistics for Left-Censored Data

• Activity Number: 341
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Maya Sternberg*+
• Companies: CDC
• Keywords: left censored ; limit of detection ; ROS ; MLE ; maximum likelihood

Abstract

It is not uncommon in laboratory and epidemiologic literature to have a left censored variable. These types of variables often result from quantitative bioassays. Bioassays are essential tools in the evaluation and development of drugs and to monitor biomarkers, as well as pollutants in the environment and water. Bioassays tend to have a limit of detection and/or a limit of quantification. The detection limit is often a function of the concentration chosen for method calibration, of laboratory practices. Therefore, the practice of using ½ the detection limit can reflect lab methodology and arbitrary choices, not real data. In this paper, we plan to compare several statistial alternatives for the estimation of left censored data , such as: maximum likelihood estimation (MLE), multiple imputation, regression on order statistics (ROS), and the Kaplan-Meier method.