This could significantly underestimate the real impact. A single point measurement is perhaps a pretty noisy measure of long term average. If we had lifetime averages, the quintiles would be more purely differentiated by the variable of interest, and the risk would be as well.
Or overestimate?
Holding all else equal, noisier estimates bias us towards the null. This is attenuation bias.
However, the estimates are still probably overestimated. Confounding, p-hacking, publication bias, all move us towards larger estimates.
“Still probably” is classic statistical science