A Nonparametric Method for Combining Multilaboratory Data
- PDF / 616,828 Bytes
- 12 Pages / 504 x 719.759 pts Page_size
- 92 Downloads / 214 Views
0092-8615/2002 Copyright 0 2002 Drug Information Association Inc.
A NONPARAMETRIC METHOD FOR COMBINING MULTILABORATORY DATA JIE HUANG,PHD, AND Rocco BRUNELLE,MS Statistical and Mathematical Sciences, Lilly Research Laboratories. Indianapolis, Indiana
In multilaboratory clinical studies, laboratory differences cannot be ignored when analyzing data from different laboratories together: In this paper, we propose a new method to transform data across laboratories. The method uses established empirical distributions of assay results from different laboratories for the transformation. This transformation does not rely on either the distributional assumption of the assay reading data or the reference ranges that are required by commonly-used standardization methods. A simulation study shows that the mean and standard deviation of the transformed data are close to the true values. A clinical example is given to compare the usual standardization method to our method. In general, our method is shown to be useful in combining data from multiple laboratories, especially when the data are nonnormal and when the reference ranges are not reliable. Finally, the pros and cons of the method are discussed. Key Words: Cumulative density function; Standardization; Transformation; Reference ranges; Empirical distribution
INTRODUCTION AN ISSUE THAT OFTEN arises in clinical trials involves the combination of data from multiple laboratories. This can occur when various investigators do not utilize the same laboratory. It is common for these laboratones to use different techniques and instruments, thus leading to different underlying distributions of the same measurement. Even with a single central laboratory, assay techniques could be modified during the study, which could also affect the assay results. This is more likely to happen in long-term studies. In either situation, a standardization method is required to combine the data. The International Conference on Harmonization E9 guideline Statistical Principles for Clinical Trials specifies that, “if different units or different reference ranges appear in Reprint address: lie Huang, Eli Lilly and Co., Lilly Corporate Center, Indianapolis, IN 46285.
the same trial (eg, if more than one laboratory is involved), then measurements should be appropriately standardized to allow a unified evaluation” (1). However, no standard method has yet been suggested by the Food and Drug Administration on how to combine data from different laboratories. In practice two methods are typically used. The first method is based on incorporating the laboratory effect into the statistical model. However, clinic or investigator effect is often confounded with laboratory effect. The second method, proposed by ChuangStein (I), is aimed at removing the laboratory difference in the data. In this method, raw data are standardized according to the respective reference range. They are processed by subtracting the lower limit or the mean of the upper and the lower limit of the reference range and then divided by the wid
Data Loading...