Hmm, one example off the top of my head is COVID-19 vaccines. Most of the vaccine RCTs' inclusion criteria were healthy people with stable medical conditions. This means Covid-19 vaccines were initially shown to be safe and effective in healthy people only. And it was further observational evidence from post-marketing surveillance that tracked the vaccine's safety and effectiveness in other demographics, such as elderly populations, cancer patients, etc.

This is similar to the shingles vaccine evidence, where I find the retrospective studies of Pomirchy et al. and Taquet et al. to be supportive of the prospective study by Eyting et al.

But of course, this is under the condition that we have prior prospective or RCT evidence. If we reverse the order and start with retrospective studies, I agree that we would probably reach erroneous conclusions.