Bioequivalence and Bioavailability Forum • Sample size for the pilot study by the PASS2021 software

dennis001
☆

Hong Kong,
2022-07-24 12:56
(802 d 08:44 ago)

Posting: # 23159
Views: 2,941

Sample size for the pilot study by the PASS2021 software [Power / Sample Size]

Dear colleagues,
We want to conduct a pilot study about the cardiovascular disease.
The aim of pilot study is to compare the postoperative levels of a biomarker between three treatment groups.
We try to estimate the sample size for the pilot study by the PASS2021 software.

PASS2021 software provide this routine presents rules of thumb for finding an appropriate sample size for a pilot study in which the outcome is a continuous measurement (postoperative levels of a biomarker in our study).
These results are for the total sample size of two-group studies in which the sample sizes of each group will be equal. Researcher should pick a sample size for the pilot study.
This procedure named Pilot Study Sample Size Rules of Thumb could generate reports which give the recommendations of statistical researchers as to the sample size of pilot studies. These rules of thumb for two groups can be adapted for single-group or multi-group studies by multiplying the recommended pilot study sample size by an appropriate adjustment factor (e.g., 0.5 for a study with only one group or 1.5 for a study with three groups, etc.).
The procedure based on the flat rules of thumb, which information provided in Machin et al. The results shown the sample size is no less than 10 for a study with only one group, no less than 20 for a study with two groups or no less than 30 for a study with three groups.
So, it seems that we could recruit 10 patients in each treatment group and a total of 30 patients for three groups in our pilot study?

We have uploaded the results estimated by PASS2021 Pilot Study Sample Size Rules of Thumb (see figure 1) [image]

However, could we conduct the Pilot Study based on the sample size? (eg. A total of 30 patients, 10 in each group, were included in our study.)
Any help would be much appreciated. Many thanks!

Helmut
★★★

Vienna, Austria,
2022-07-24 14:27
(802 d 07:13 ago)

@ dennis001
Posting: # 23160
Views: 2,528

Pilot study (independent from software)

Post reply

Hi dennis001,

❝ However, could we conduct the Pilot Study based on the sample size? (eg. A total of 30 patients, 10 in each group, were included in our study.)

We need more information. Are you interested in all pairwise treatment comparisons (i.e., A vs B, A vs C, and B vs C) or is one of them a standard treatment acting as a reference (i.e., A vs R and B vs. R)? Are you interested in superiority or equivalence? At least, since this is a pilot study, you must not be worried about multiplicity issues.
Biomarkers are [image]

lognormal distributed. Hence, you should apply a multiplicate model \(\small{\mu_\text{T}/\mu_\text{R}}\) or – easier – an additive model \(\small{\log_{e}\mu_\text{T}-\log_{e}\mu_\text{R}}\). For any sample size estimation you have to assume treatment difference(s) and variance(s). Note that all versions of PASS require the standard deviation of the treatment difference as input.

The purpose of a pilot study is to obtain estimates of the treatment difference(s) and variance(s) in order to design a pivotal study with a desired power. Estimates are always uncertain. The larger the pilot study, the more certain (i.e., reliable) the estimates will be (the uncertainty decreases roughly with \(\small{1/\sqrt{n}}\)).
In the sample size estimation of the pivotal study you must take these uncertainties into account. Hence, the smaller the pilot, the larger the pivotal. For background in the equivalence setting see this article. Finding an ethical / economic optimum \(\small{(n_\text{pilot}+n_\text{pivotal}\rightarrow \text{min})}\) can be tricky.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

dennis001
☆

Hong Kong,
2022-07-24 16:27
(802 d 05:13 ago)

@ Helmut
Posting: # 23161
Views: 2,403

Pilot study (independent from software)

Post reply

Dear Helmut Schütz,
Thank you very much for your suggestions.
Your paper is very interesting and so professional.
And we will try to calculate the sample size with all pairwise treatment comparisons.
Notablly, you said that all versions of PASS require the standard deviation of the treatment difference as input. However, the paper of the Pilot Study Sample Size Rules of Thumb in PASS Sample Size Software Documentation did not present the standard deviation or any other parameter (see figure2 and figure3)

[image]

Did we ignore any important information?

Many thanks

BEST regards

Dennis

Helmut
★★★

Vienna, Austria,
2022-07-24 16:55
(802 d 04:45 ago)

@ dennis001
Posting: # 23162
Views: 2,439

Rules of Thumb (PASS)

Post reply

Hi Dennis,

❝ Thank you very much for your suggestions.

Welcome.

❝ Your paper is very interesting and so professional.

Just a collection of desultory thoughts – not a peer-reviewed paper. Maybe it’s all crap. :-D

❝ And we will try to calculate the sample size with all pairwise treatment comparisons.

OK. Still: Are you targeting superiority or equivalence? Yin and Yang.

❝ Notablly, you said that all versions of PASS require the standard deviation of the treatment difference as input.

That was my experience before I deleted it from my machine two years ago.

❝ However, the paper of the Pilot Study Sample Size Rules of Thumb in PASS Sample Size Software Documentation did not present the standard deviation or any other parameter (see figure2 and figure3)

That’s not a paper but part of the manual, which is available online. In the future please don’t post screenshots in JPG-format, use PNG instead. JPEGs are blurred and full of compression artefacts making them difficult to read.

❝ Did we ignore any important information?

No idea. I don’t own the book of Machin et al. and hence, can’t tell. Will check Whitehead et al.,¹ time allowing.
That’s the problem with software (not limited to commercial ones). If the documentation is not exhaustive, you never know what’s going on under the hood² when you click the button.

Whitehead AL, Julious SA, Cooper CL, Campell MJ. Estimating the sample size for a pilot randomised trial to minimise the overall trial sample size for the external pilot and main trial for a continuous outcome variable. Stat Meth Med Res. 2016; 25(3): 1057–73. doi:10.1177/0962280215588241 Free Full text.
For some observations see this thread.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

dennis001 ☆ Hong Kong, 2022-07-24 19:48 (802 d 01:52 ago) @ Helmut Posting: # 23164 Views: 2,464	Rules of Thumb (PASS) Post reply
	Dear Helmut, Thanks a lot ❝ OK. Still: Are you targeting superiority or equivalence? Yin and Yang. we might calculate the sample size with all pairwise treatment comparisons targeting equivalence. ❝ 2. For some observations see this thread. unbelievable！unbelievable！ We seriously doubt the accuracy of method for PASS software.