Yes, that is the ATC anomaly in the CONCEPT_ANCESTOR table.
Usually, you can rely on the fact that the children of your children are you grandchildren. With ATC, it doesn’t work like that, because those concepts are defined using a number of attributes, which don’t allow a hierarchical mapping into RxNorm (they are picked up much lower in the hierarchy there). Therefore, we created an exception from that rule.
In your case, ATC concept 21603535 has the attributes “certain surfactant ingredient”, “non enteral dose form” and “combination”. Which means, not all children of the ingredients are also grandchildren of this particular ATC. When we had that, we had a gigantic permutational explosion of wrong descendants.
So, don’t double jump. Your query against the DRUG_EXPOSURE table is right on the money. Or are you expecting something else to show up that is not there?