@aostropolets this might be another forum post, but I would be weary of any diagnosis codes found on claims to represent “actual onset” of a diagnosis. Physicians add diagnoses they treated the patient for on their bills. For acute conditions, you might be able to INFER that the diagnosis treated was the date of confirmed onset. But for chronic conditions (i.e. diabetes), physician will bill for chronic condition diagnoses treated, regardless if it was confirming a new case or treating an old case. The diagnosis codes are the same between new and existing conditions. So generally, to find “onset”, the analyst needs to use some kind of temporal rule (i.e. look for the first diabetes diagnosis in the dataset or look for a diabetes diagnosis with a clean period of 1 year?) to INFER onset. So your concern, while valid, is not limited to POA, but to all claim diagnoses.
I feel this is outside of the realm of CDM standards and should be left up to the analyst to understand, based on the dataset and inherent biases within that dataset.