Guangyong Zou talks with
ScienceWatch.com and answers a few questions about this month's
Emerging Research Front Paper in the field of Materials
Science.
Article: A modified Poisson regression approach to
prospective studies with binary data
Authors: Zou, GY
Journal: AMER J EPIDEMIOL, 159 (7): 702-706 APR 1 2004
Addresses: John P Robarts Res Inst, Robarts Clin Trials, POB
5015,100 Perth Dr, London, ON N6A 5K8, Canada.
John P Robarts Res Inst, Robarts Clin Trials, London, ON N6A
5K8, Canada.
Univ Western Ontario, Dept Epidemiol & Biostat, London, ON,
Canada.
Why do you think your paper is highly
cited?
This paper provides a simple statistical method that can be used to more
precisely estimate effect size. The ratio of risks is a key parameter in
epidemiological and clinical research, but conventional logistic regression
models can only provide estimates of the ratios of odds, which can be
misleading.
For example, a controversy in the US mass media in 1999 was created by odds
ratio estimates in a study published in the February 25, 1999 issue of the
New England Journal of Medicine. This otherwise well-conducted
study falsely concluded, as a result of odds ratio estimates, that race and
gender influence how physicians manage patients with chest pain. The method
presented in my paper could have prevented the controversy.
Does it describe a new discovery, methodology, or
synthesis of knowledge?
The paper describes a new method that is broadly applicable to quantitative
research. Although it has been known for decades that the ratio of odds was
created to approximate the ratio of risks when the latter was not estimable
(for example, in case-control studies), researchers continue to use odds
ratio based methods to estimate risk ratio in prospective studies. This is
due largely to the lack of simple statistical methods.
Would you summarize the significance of your paper in
layman's terms?
"The method described in the paper may be extended
to studies with more complex designs and to the
meta-analysis of clinical trials and
epidemiological studies."
Scientists in clinical and epidemiological research usually speak of risks
or benefits in relative terms. For example, if a healthy diet reduces the
risk of a disease from 30% to 10%, then the ratio of risks is 3. In this
case it is also mathematically correct to say that the diet reduces the
odds of disease from 3/7 to 1/9, which results in an odds ratio of 3.86.
An intrepid investigator would interpret the odds ratio as a risk ratio and
thus unknowingly exaggerate the effect of the diet by 30%. Although such
numerical discrepancies are well-known in epidemiology, statistical
textbooks continue to provide methods for obtaining ratios of odds. My
paper provides a simple approach to directly estimate ratios of risks.
How did you become involved in this research and were
any particular problems encountered along the way?
This is the first article I wrote after I graduated with a Ph.D. in
Biostatistics from the Department of Epidemiology and Biostatistics at the
University of Western Ontario under the supervision of Professor Allan
Donner, Ph.D.
I was intrigued by the unnecessary controversy created by the
aforementioned study on the role of race and gender in influencing how
physicians manage patients with chest pain. I encountered no problems along
the way, owing to the excellent advice provided by my colleagues in Robarts
Clinical Trials of Robarts Research Institute and in the Department of
Epidemiology and Biostatistics.
Furthermore, in addition to the encouragement and generous help from
Professor Allan Donner, I received constructive comments from the editor of
the American Journal of Epidemiology, Professor Donna Spiegelman,
Ph.D., who decided to publish the paper without an external review.
Where do you see your research leading in the
future?
The method described in the paper may be extended to studies with more
complex designs and to the meta-analysis of clinical trials and
epidemiological studies. This paper also provides a relevant example of how
biostatistical research can change the interpretation of research data.
Do you foresee any social or political implications for
your research?
Yes. Data with high quality are expensive to acquire. But data cannot speak
well for themselves without the help of good statistical methods. As a
biostatistician, my aim is to provide investigators with useful tools for
their research, thus helping them draw more valid conclusions.
GuangYong Zou, Ph.D.
Associate Professor and Scientist
Department of Epidemiology and Biostatistics
and
Robarts Clinical Trials of Robarts Research Institute
Schulich School of Medicine and Dentistry
The University of Western Ontario
London, ON, Canada Web