Randomized non-comparative trials: an oxymoron?

ESMD:

Pocock was suggesting that, if certain strict criteria were met, it might be possible to “augment” the control arm of an RCT with historical data. But this is different than ignoring the RCT control arm data altogether and, rather than borrowing data from a historical control

Correct. Borrowing historical information to inform RCT comparisons is defensible. Comparing with historical controls also actually has a rationale. What RNCTs do is in a league of its own. For example, some RNCTs will randomize patients to treatment or placebo and then not compare the two. Instead, they benchmark each separately against external or prespecified values. It is just so odd and inverts the usual logic by Pocock etc.: a historical-control design reaches for an external comparator because it has no concurrent one, whereas the RNCT discards the better (concurrent, randomized) control it already paid to obtain in favor of a weaker external benchmark. What is the point of the placebo arm if not to be compared with treatment?

Thus, the mental model is not that “RNCT = swap a historical control in for the RCT control arm”. RNCTs are actually stranger than that: they usually keep a concurrent randomized control and then decline to use it…