The research has already been done historically without convincing results. That's why they want NIH large dataset (you need large dataset to find a small effect size).

But NIH understably will not share a dataset that the contributors wouldn't want to used in this way.

But if you're a researcher in this domain just go get your own dataset. You'll have to tell the potential contributors you will allow research that might stigmatize them.

Good luck with recruitment.

It you don't think you could recruit for that study because participants would find it objectionable then you can't demand access to NIH dataset.

Ethically it's that simple.