Retain for Heart Failure

[asoroosh]

Hello!

Really great work and quite a useful description of the code -- thank you!

I have two questions which might sound naive:

1. What event to_event refers to? I initially thought it is heart failure, but it seems to be mortality. Then the to_event refers to the time (in days) until 2025-1-1. Is that a fair understanding?

2. Choi et al. uses heart failure as an event; could a similar thing be done using your keras version and MIMIC-III? The target will be changed to whether a subject has a ICD-9 for heart failure (e.g. 428.0), removing any Dx after the heart failure and recalculating all to_dates to 2025-1-1 similar to the existing code?

Thank you in advance,

[jstremme]

@asoroosh, all thanks go to @tRosenflanz!

You're correct that to_event is the number of days until the event we're predicting, in this case, mortality.

Yes, you could predict any event using the approach provided by swapping out the target variable.