The Sense and Nonsense of Exponential Models in Clinical Research

Abstract

Exponential models are very popular for the assessment of clinical research. Nature is unpredictable and exponential models may be adequate for simple organisms but less so for human beings. Theobjective was to review the potential advantages and limitations of this approach. The following models were reviewed: Cox proportional hazard regression, Markow modeling, polynomial regression for Laplace transformed multi-exponential models, and logistic regression. Real and hypothesized data were used as examples.

In conclusion, the advantages include: the possibility to conveniently study subgroup effects, otherwise called confounders; the possibility to derive pharmacokinetic parameters; andthe possibility to make long-term predictions from observational data. Limitations include: the exponential models may not fit the data well enough; they may give rise to some serious misinterpretations of the data; most of the statistical software for exponential models assumes that confounders are homogeneous, while, in reality, this is virtually never true; and the exponential models are a major simplification for the description of research data in human beings. We underline that exponential modeling, although a very interesting tool for exploratory research, is not adequately reliable for confirmational research. It is to be hoped that this paper will help investigators to more cautiously interpret the results of their exponential data analysis.

Key words: :

• Clinical Research
• Exponential Models
• Logistic Regression
• Cox Proportional Hazard Regression
• Markow Models
• Polynomial Regression
• Laplace Transformed Multi-Exponential Models

Notes

[7] BUGS y WinBUGS. http://www.mrc-bsu.cam.ac.uk/bugs http://cran.r-project.org