What is an acceptable way to present patient summaries for labs descriptive summaries for a n <= 2 in an early phase trial.
Using standard tables will yield very sparse cells. Alternatively, developing listings for over a 100 lab parameter study will be unwieldy especially to present at most 2 patients.
I am leaning towards a case summaries written by the clinical team for these few patients but is this an acceptable approach for FDA submissions?
I don’t know of an official FDA line on this but would highly recommend against computing anything, even the mean, when n < 3. As a slight aside, it takes n=70 to adequately estimate a standard deviation (to with a multiplicative margin of error of 1.2 with 0.95 compatibility).
Hi,
Can you offer some additional details on the study design and also on the timing, since you reference an FDA submission.
If the study has already been done (presuming at least IRB approval) and you are now, on a post hoc basis, trying to understand what may be acceptable to the FDA, it would seem reasonable to search for other similar studies and how those data were reported, at least to have some kind of precedent to reference. It is not clear if this study is to be the primary means of establishing the basis with the FDA for continuing with additional investigative research, or if this study is to supplement a more typical phase I submission.
If the study has not yet been done and you are looking for designs and reporting approaches that can make sense, such that you can then take the proposed design and analysis plan to the FDA for discussion and approval of the design a priori, you might consider so-called “N of 1” designs, which have been used in the setting of rare diseases (e.g. oncology) and other scenarios where more typical phase I study sample sizes would not be apropos (e.g. pediatric studies).
N of 1 studies tend to be a series of individual subject cross-over studies, and there have been numerous publications and extensions of guidelines for those. As an example, there was a special issue of HDSR here:
Personalized (N-of-1) Trials: Methods, Applications, and Impact
with one paper focused on reporting guidelines:
N-of-1 Trials, Their Reporting Guidelines, and the Advancement of Open Science Principles
In the above, there are references to the SPENT extension to SPIRIT and to the CENT extension to CONSORT that may be helpful to you.
Broadly speaking, I agree with Frank’s view above, which is that usual descriptive statistics in this setting would not typically be of value given their inherent instability.