SAMPLE SIZE is a very interesting topic! If all you want is to reach what is considered to be the standard level of statistical significance (i.e: p < 0.05) - here (below) is a useful illustration showing how even a small sample size of 20 can still do it: you just need a large difference in the treatment effect, while a small difference requires massive number. Source:
https://doc-aids.com/a-lesson-on-sample-size/ (with thanks)!
The AIM in our clinical trials is to prove a
DIFFERENCE : Per hypothesis, we believe and we are saying that;
ISLA 101 prevents infection, while placebo adds nothing to affect natural disease acquisition! ISLA 101 treats (or prevents full blown) disease, while placebo adds nothing to affect natural disease progression once infected! The
challenge trial design will provide us with the most uncontaminated data we can get from human trials. It eliminates lots of uncertainties as we will know who got infected, when they got infected, etc!
Also, deliberately infecting all subjects gives us a 100% infected rate to start with! This matters (as
@Davisite already explained). We have been given a sample with a higher infection burden that what we might get in the streets of New York or Melbourne. Its a dream situation! Let me try and show how.
First, lets note our Primary Endpoint: It's Viraemia: Although this may constitute a continuous variable, for illustration and simplicity, lets have this as a binary outcome (a YES or NO): Low (or negative) versus High (or positive). The rates presented will therefore be the Positives at a chosen time point.
I have done some Sample size Calculations to show why/how the starting point matters! Please Note: the numbers entered on the "Incidence" column of the table below are for illustration only, and have been chosen specifically to give us an equivalent treatment effect: a 12.5% relative risk reduction in the outcome rate following ISLA 101.
These stats were calculated using an online Sample Size calculator at
https://clincalc.com/stats/samplesize.aspx . Try it!
As can be seen: A Clinical trial in a moderately low infection burden population (8%) requires a lot more patients than one done in a high infection burden population. In many cases, the starting rate might even be in lower - below 5%!
In reality, I do not know if anyone would attempt a 1% ARR from a baseline of 8% - its too small, and my be considered not clinically meaningful. The number needed to treat is very high: . This is for illustration only - to show why our numbers are nothing to laugh at!
What if the effect of ISLA 101 is even larger: say it reduces the outcome rate from 80% down to 40%? We would need just 44 patients, for a p value of < 0.05 at 80% statistical power!
------
Noting that there is no need to achieve the 80% statistical power in a Phase 2 trial - guess what numbers we need? Just 16 will achieve the required p-value (I set the power to 50%).
-----
ANY VALUE IN INCREASING THE SAMPLE SIZE BEYOND WHAT THE SS CALC SAYS?
We would generally like an inflation of a certain % (such as 3%) over and above what gets us across the line, to cover for withdrawals or some unexpected events that might produce some un-interprettable data!
But, simply increasing the numbers for the purpose of saying its a big trial is, from a (Frequentist) statistics point of view, its POINTLESS! The p-value is already a problematic statistic, and while making it smaller makes one feel comfortable and happier, it adds little to the usefulness of the results. Catching more fish from the same fish pond adds nothing to the body of evidence, and certainly does not help with generalisability of the results! So, why waste money doing it?
A more logical and certainly more valuable approach would be to repeat the trial in a different setting. Repeatability, after-all, is what science is about! This is not only a much more valid way to prove effect, but it does what's on the tin: it adds generalisability of the data and would allow us to say: ISLA 101 reduces the rate of this outcome by x% (END OF)!
-----
There was mention of Bayesian methods having been used by AXT, and we know it's planned for use by RAC! Perhaps it could have been utilised here as well due to the abundance of previous data (to calculate the prior probability). However, for simplicity - a frequentist approach still rules (the use of p-values). Although Bayesians will argue that theirs is the superior of the two approaches! I guess, what the FDA agree to, goes!
-----
Here is Professor Stephen Thomas talking about human challenge studies - starting from 12:23.
VIDEO 
(20min delay)