Sample size calculation in stata, please help

[Madu Abuchi]

I want to recalculate a sample size presented in a published study using stata but struggling with it.

The study mentioned that cumulative event rates will be calculated by survival analysis methods and compared between groups by the logrank test. Given this information I was trying to use the power logrank sample size calculator of stata but struggling.

The study states:
"To be able to detect, at the 5% level of significance and with a power of 90%, a 26% difference between the treatment groups in total cardiovascular events over 5-year follow-up will require approximately 6000 patients (3000 in each group). This sample size will includes an allowance of at least 700 patients (i.e. 15%) for dropouts and crossover to the alternative regimen. The basis for this sample size calculation is number of cardiovascular events (about 20 per 1000 patient-years) observed in the group receiving active drug treatment ..."

Can anyone help me out here please.

Thanks

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Also, if you refer to a paper, provide a full citation and perhaps a link to the paper. Without your code and output (and that of the paper you cite) we can't see what your problem is. If the paper you're citing just did a simple comparison of means, then the power calculator your chose is not the right one.

Stata 15 has a full range of power calculators one of which probably is what you need.

[Madu Abuchi]

Hi Phil,

thanks.

The paper is Wing, L. M. H., Reid, C. M., Ryan, P., Beilin, L. J., Brown, M. A., Jennings, G. L. R., . . . West, M. J. (2003). A Comparison of Outcomes with Angiotensin-Converting–Enzyme Inhibitors and Diuretics for Hypertension in the Elderly. New England Journal of Medicine, 348(7), 583-592. doi: 10.1056/NEJMoa021716. http://www.nejm.org/doi/full/10.1056...1716#t=article

The protocol of this paper states that "cumulative event rates will be calculated by survival analysis methods and compared between groups by the logrank test" See Second Australian National Blood Pressure Study (ANBP2): Australian Comparative Outcome Trial of ACE inhibitor- and Diuretic-based Treatment of Hypertension in the Elderly:Management Committee on behalf of the High Blood Pressure Research Council of Australia http://www.tandfonline.com/doi/abs/1...41969709083186


In this paper, alpha is 0.05, power 0.90, difference 25% (0.25), assuming a rate of 21 events per 1000 person-years in the diuretic group (0.021)

My initial attempt was:

power twoproportions 0.021, test(chi2) diff(0.25) power(0.9)

My result is attached and this is not close to what was mentioned in the articles

[Marcos Almeida]

Madu, you should consider the command - power exponential - for that matter.

As explained in the Stata Manual, you may select the HRs (or the difference), the studytime, follow-up period, the probability of losses (overall or by group), the accrual period and - shall you be hungry for more - the time when losses are considered.

Really awesome, this command. To get acquainted with it, please read the manual carefully.

[Philip Ryan]

This is really testing my ageing memory, but we wrote the protocol for ANBP2 back in late 1993 or early 1994. We had very little to go on, but took (plucked?) the figure of 21 events/1000 pyrs for the controls from a previous UK trial in approximately the same domain. We figured a 25% improvement in outcome would be clinically important. We were looking at a 5 year trial. So (and I do not need reminding of the assumptions and limitations of this!) we figured that the cumulative proportion of outcomes in the control group at the end of the study would be 5 x 21/1000 = 0.105. A 25% reduction would yield a cumulative proportion of 0.07875 in the treated group.

power twoprop  .105  .07875,  a(.05)  p(.9)  contin

yields 2619 in each group and, after a 15% inflation to account for dropouts etc, you get to about 3000 in each group.

You get similar results if you use the power logrank command on the complementary proportions. Of course back in 1993 there was no power command, we would have used something like sampsi (was it sampsiz back then?) or something else outside of Stata.

I post this only to assist the original poster to see how the numbers arose. Of course, if you have better information on hazards, accrual, dropout and other parameters then you would use something more sophisticated like power exponential.

[Madu Abuchi]

Thanks Marcos and Phillips