Exclude pediatric patients (younger than 18) from clinical prediction tasks

[DarylOkeke]

Multiple clinical prediction tasks had TODO comments saying that patients under 18 should be excluded, but the filtering was never implemented.

This adds age filtering to the mortality prediction, length-of-stay, and drug recommendation tasks across MIMIC-III, MIMIC-IV, eICU, and OMOP datasets. Adult prediction benchmarks shouldn't mix in pediatric cohorts b/c the clinical patterns are too different.

Some notes:

• eICU stores age as a string and uses "> 89" for patients over 89, so that case is handled explicitly
• MIMIC-III requires deriving age from dob and admission time since it doesn't store a precomputed age field
• All checks fail-open: if age data is missing or unparseable, the patient is included instead of dropped

[jhnwu3]

Hey nice first PR!

Is it possible to add a Boolean arg for this filtering? I think it'd be nice to be able to quickly ablation study this difference in patient demographics later.

[DarylOkeke]

Thanks! I added exclude_minors to all the affected classes and legacy functions in the 3 task files. Defaults to true so the existing behavior stays the same

Another thing though, I saw that LengthOfStayPredictioneICU doesn't have an age check at all. The original TODOs didn't cover it but the other 3 LOS classes do. Want me to add one for consistency, or no?

[jhnwu3]

Yeah that would be nice

[DarylOkeke]

Finished!

[jhnwu3]

Will check back after unit tests finish this weekend

[jhnwu3]

One last thing before we meet, can you verify it works locally on your machine with our demo dataset?

https://physionet.org/content/mimic-iv-demo/2.2/

[jhnwu3]

closed bc duplicate (it seems?). Reopen if you disagree