Claims Data Preprocessing

[DanielYWu]

Hi,

Thank you for the paper on using self-attention for medical cost prediction; learned a lot on how to apply DL to claims data! I had a quick question; in 4.2 you mentioned that in the claims database "facility and pharmacy visits were grouped by month." I'm assuming this was done only for the sequential medical feature baseline models, and not for any models that accept irregularly temporal input?

In addition, I was curious as to how you aggregated claims data to the visit level. Did you use service date? Given that some facility claims for Inpatient hospital stays may span multiple days, curious if the dataset if you use had another identifier for a visit.

Thank you in advance!