How to learn Biostatistics?

Lots of clinicians and practicing doctors feel difficulty in understanding the biostatistics. As a result, the research papers we see, published in our journals, lack good statistical methods and so contain many errors. Because of poor knowledge, our papers loose value.

Is there any way, through this forum, users can learn about the important biostatistics necessary for a clinician to know?

I know biostatistics is a separate subject, but still its better to learn it then to relying on others for simple statistical problems.

I suggest to work on following example study and lets proceed it till end.

Topic: Drug A is a new blood pressure lowering medicine.

Hypothesis: Drug A significantly lowers the blood pressure by 20 mmHg systolic than placebo.

Study design: Cohorot study
Cases: Those given Drug A
Controls: Those given Placebo

Statistical design & Sample Size: Yet to be decided.

Please comment ......

[Edited by docosama on 06-06-2005 at 11:12 AM GMT]

Posted by: docosamaPosts: 333 :: 06-06-2005 :: | Reply to this Message

Re: How to learn Biostatistics?

If I understood your question correctly, you are asking about a clinical trial of a BP lowering drug.

Generally speaking, we have two options for study design here. A randomized placebo-controlled trial or a cross-over trial. Both should give the answer that this question poses. There are, ofcourse, advantages and disadvantages of both designs and choice usually depends on the resources available.

Student't test will probably be the easiest test to detect a statistical difference between the normally-distributed means of the two groups.

For calculating sample size, atleast two things are needed. One is the size of BP lowering effect that you want to detect, and the second is the standard deviation of the population blood pressure response when given this drug.

For example, a total of 12 patients will be needed for a randomised clinical trial study design, if we set power at 80 percent and that the study will detect a treatment difference of 20 mm Hg at a two sided 5.000 percent significance level. This is based on the assumption that the standard deviation of the BP response is 10.000 mm Hg. Results will be different if we choose different numbers.

[Edited by rqayyum on 06-08-2005 at 08:10 PM GMT]

Posted by: rqayyumPosts: 199 :: 06-06-2005 :: | Reply to this Message

Re: How to learn Biostatistics?

I was just thinking why have you selected Randomized Clinical Trial?

Posted by: yasirPosts: 90 :: 08-06-2005 :: | Reply to this Message

Re: Re: How to learn Biostatistics?

Docosama can give the answer that why did he choose such a study design (I would have choosen the same study design for an example!).

However, for the same power (80%) and significance level (5%), and assuming that within-subject standard deviation is 10 mm Hg and the true difference before and after the treatment is 20 mm Hg, 7 patients will be needed for a cross-over study design.
Pair-wise t-test can be used for statistical analysis provided we assume a normal-distribution, otherwise a non-parameteric test will be needed.

quote:

---

yasir wrote:
I was just thinking why have you selected Randomized Clinical Trial?

---

Posted by: rqayyumPosts: 199 :: 08-06-2005 :: | Reply to this Message

Harvard U.'s Sample size calculator

Harvard University's Biostatistics center has a simple online sample size calculator.
" http://hedwig.mgh.harvard.edu/sample_size/size.html "

[Edited by rqayyum on 06-08-2005 at 08:20 PM GMT]

Posted by: rqayyumPosts: 199 :: 08-06-2005 :: | Reply to this Message