We are planning a study on anti TB treatment: daily vs. alternate day regime. The patients will be divided into two groups: one will receive the daily regime and the other will receive the alternate day regime. The parameters that we aim to study are sputum conversion rates at the end of 2 months, cure rates at the end of treatment and relapse rates till 18 months after completion of treatment. What would be the ideal sample size for the study?
What difference in the treatment outcome between the two regimens do u (as the PI) consider as significant clinically? that would give you the answer to your "clinical" effect size ! for e.g.: if treatment arm 1 has sputum conversion of 80% and treatment regime two has a sputum conversion rate of 85%…(literature review would give a probable number unless you r doing a pilot study)….So in the e.g. you have to say that the clinical effect size of 5% is "clinically meaningful" and if you would put that number in the online sample size calculator , it would tell you the size of the sample (minimum) which is powered (80%) to detect the difference which is not by chance (p
It sounds like your study design is fairly simple with only two groups. And the the response variables are binary (Although maybe sputum conversion rate are not). I would recommend simulating your data and analyzing it with the statistical model you intend to analyze the real data with. This would be a fairly simple process in R or even in excel. The following is a link to a blog that I wrote about performing this procedure with R code that you could modify for your particular problem.
Good luck!
http://dgwinn.wordpress.com/2014/04/06/virtual-ecologist-approach-using-simulations-to-evaluating-experimental-designs/
do u plan to include children as well?
What is the clinical effect size that you are choosing as "significant"? the sample size would depend on that
Ramesh Menon
No, we are excluding children from the study since the diagnostic criteria for inclusion is "sputum positive PTB". How do we determine the clinical effect size? We have the cure rates in the two groups as primary outcomes. The power of the study is chosen at 80.
What difference in the treatment outcome between the two regimens do u (as the PI) consider as significant clinically? that would give you the answer to your "clinical" effect size ! for e.g.: if treatment arm 1 has sputum conversion of 80% and treatment regime two has a sputum conversion rate of 85%…(literature review would give a probable number unless you r doing a pilot study)….So in the e.g. you have to say that the clinical effect size of 5% is "clinically meaningful" and if you would put that number in the online sample size calculator , it would tell you the size of the sample (minimum) which is powered (80%) to detect the difference which is not by chance (p
Hello Mishra, I would like to share one study that I was involved in, with similar conditions that you mentioned. For that study, the sample size was determined on the basis of confidence level 95% and precision 10%. We also assumed the possible relapse rate of 30% and in order to be representative the final sample size estimated was minimum of 81 patients.
Further, you should consider follow-up losses and withdrawals that also need to be adjusted.
All the best
You will need to expect the difference to be yielded by the two regimens for the sputum conversion rate after two months, for the cure rate at the end of the trial and the relapse rate until 18 months. Then calculate the sample size for the three alternatives as illustrated by P Ramesh Menon and Umesh Ghimire . You will get three sample sizes, then select the larger one because it will include the three alternatives. Good luck.
1) First download "power and sample size" http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize#Windows
2) Open the program, look for "Dichotomous" folder
3) In the output section, in the blank section set for "Sample size"
4) In design section set for "Independent" because you have two independent groups, then in case control blank set for "Prospective"
5) In design section new blank spaces will arise, for how is the alternative hypothesis expressed? look for two proportions and in the next space ser for fisher exact test, which is a better estimation in sample size.
6)In the input section
alpha= 0.05
power= 0.8
p0= (proportion expected in traditional treatment) 0.15 if 15%
p1= (proportion expected in experimental treatment) 0.40 if 40%
m= is the control/case ratio, 2 control for case m=2. or 1 control for case m=1
Considerations.
you have two objetives (cure rates and relapse) so yow have to do two sample size calculations, power should be 80% at least but 90% its my suggestion. You should consider that patients with alternated regimen will be less than the daily regimen so you should try to increase "m" value to 2 or 3 but no more than 3 because you woul increase the sample the cost of the study with no real benefit.
I hope this point of view will help you.
P.S. Bigger the sample size more insignificant differences will appear to be statistically significant but no clinical meanning and a lot of resources used.
If you have access to a very large population then engage a statistitian to help you; if not, you may have to use all the patients
all the best for your project. the outcome would be relevant and timely
Thanks
Mishra, try to cosider the prevalences also while determining sample size. Post study effects or outcomes selected will depend upon prevalences of various categories of Tb and relapses. Better take different independent sample sizes depending upon their prevalences .
- Badges
- Science topic
- Similar topics
- Medicine
- Public Health