Discussion: View Thread

I would like to ask advice from this statistical consulting group. Both at grant work shop and inital meetings with physicians that I have participated, I often been asked about study design . My question is why physicians often seek advice about study design from biostatistician.

Thank you !
Cindy Weng

-------------------------------------------

Cindy Weng, Biostatistician

Pediatrics Research Enterprise

Study Design and Biostatistics Center 

Department of Pediatrics

School of Medicine

University of Utah

Phone: (801) 213 3753

Statistics Without Border (SWB)

American Statistical Association (ASA)

 

-------------------------------------------

Simple question which has a simple answer.  Research projects usually involve statistical analysis.  Sample size becomes an issue especially for the investigators.  How you conduct the study and how many patients you need has everything to do with statistics.  Design questions of a statistical nature include:
1. What are the study endpoints and how are they determined?
2. How are endpoints analyzed or compared?
3. If you have treatment groups, how many are there?
4. Do you construct this as a parallel treatment designor cross-over?
5. Do you need randomization and or blinding?
6. If blinding, what level of blinding
7. Is it retrospective, prospective or cross-sectional?
8. Is it to be a fixed sample size design, group sequential or adaptive?
9. If there are multiple primary endpoints how do you handle multiplicity in the inference.
10. Are you trying to demonstrate superiority of one treatment over another, or just equivalence or noninferiority?
11. If there will be missing data how will that be handled (e.g. type of imputation method and/or sensitivity of results to the approach done inthe analysis).
12. What is the sample size required to have a specific power for rejecting the null hypothesis?

These are some of the many design questions that are statistical in nature.

-------------------------------------------
Michael Chernick
Director of Biostatistical Services
Lankenau Institute for Medical Research
-------------------------------------------

Study design is to a great extent dependent on statistical principles, and requires familiarity with not only medicine and other areas of science but statistics as well.  For example, the selection and definition of endpoints; size, duration and rules for monitoring of clinical trials; adaptive design; dose finding; case-control vs. cohort studies; etc.  Ideally, the study design should be developed through collaboration between the statistician and the clinician.  A good statistician functions as a full-fledged study team member collaborating with the rest of the team at every stage of the study, from the initial design stage to writing the final study report.  Those who simply take responsibility for sample size calculations and the statistical analysis of the data are functdioning as data analysts rather than statisticians.

Edith Zang
Independent Consultant
NYCASA
-------------------------------------------