Add NWICU

[rvandewater]

Note that not all tasks seem to work yet (mainly lab tests), presumably because of the value/unit differences, such that they are not added to be supported in the task definition. This is work in progress. Also made minor adjustments so AUMCdb would work with the new API.

┌───────────────────────────────────────────────┬──────────┬───────────────┬───────┐
│ task                                          ┆ split    ┆ boolean_value ┆ len   │
│ ---                                           ┆ ---      ┆ ---           ┆ ---   │
│ str                                           ┆ str      ┆ bool          ┆ u32   │
╞═══════════════════════════════════════════════╪══════════╪═══════════════╪═══════╡
│ blood_chemistry_elevated_creatinine_first_24h ┆ held_out ┆ false         ┆ 3921  │
│ blood_chemistry_elevated_creatinine_first_24h ┆ train    ┆ false         ┆ 30728 │
│ blood_chemistry_elevated_creatinine_first_24h ┆ tuning   ┆ false         ┆ 4113  │
│ blood_chemistry_hyponatremia_first_24h        ┆ held_out ┆ false         ┆ 2388  │
│ blood_chemistry_hyponatremia_first_24h        ┆ train    ┆ false         ┆ 18214 │
│ blood_chemistry_hyponatremia_first_24h        ┆ tuning   ┆ false         ┆ 2489  │
│ cbc_leukocytosis_first_24h                    ┆ held_out ┆ false         ┆ 48    │
│ cbc_leukocytosis_first_24h                    ┆ train    ┆ false         ┆ 526   │
│ cbc_leukocytosis_first_24h                    ┆ tuning   ┆ false         ┆ 66    │
│ cbc_thrombocytopenia_first_24h                ┆ held_out ┆ false         ┆ 3213  │
│ cbc_thrombocytopenia_first_24h                ┆ train    ┆ false         ┆ 25022 │
│ cbc_thrombocytopenia_first_24h                ┆ tuning   ┆ false         ┆ 3374  │
│ mortality_in_icu_first_24h                    ┆ held_out ┆ true          ┆ 117   │
│ mortality_in_icu_first_24h                    ┆ held_out ┆ false         ┆ 1226  │
│ mortality_in_icu_first_24h                    ┆ train    ┆ true          ┆ 1000  │
│ mortality_in_icu_first_24h                    ┆ train    ┆ false         ┆ 9190  │
│ mortality_in_icu_first_24h                    ┆ tuning   ┆ false         ┆ 1203  │
│ mortality_in_icu_first_24h                    ┆ tuning   ┆ true          ┆ 130   │
│ readmission_general_hospital_30d              ┆ held_out ┆ false         ┆ 1832  │
│ readmission_general_hospital_30d              ┆ held_out ┆ true          ┆ 1175  │
│ readmission_general_hospital_30d              ┆ train    ┆ true          ┆ 9844  │
│ readmission_general_hospital_30d              ┆ train    ┆ false         ┆ 14610 │
│ readmission_general_hospital_30d              ┆ tuning   ┆ true          ┆ 1346  │
│ readmission_general_hospital_30d              ┆ tuning   ┆ false         ┆ 2047  │
│ vital_hypotension_first_24h                   ┆ held_out ┆ false         ┆ 345   │
│ vital_hypotension_first_24h                   ┆ train    ┆ false         ┆ 2747  │
│ vital_hypotension_first_24h                   ┆ tuning   ┆ false         ┆ 353   │
└───────────────────────────────────────────────┴──────────┴───────────────┴───────┘

[rvandewater]

@mmcdermott not sure why the readmission task was removed?

[mmcdermott]

@mmcdermott not sure why the readmission task was removed?

@rvandewater It was removed because it is not actually a useful task on ICU only datasets -- we should design a HRRP readmission proxy task for general hospitalization datasets, but the current task is misleading.

[rvandewater]

@mmcdermott merged dev into this branch. I am not sure when I have bandwidth to verify the lab-dependent tasks which do fit the right codes but where the values might mismatch due to unit differences, so anyone is free to pick this up (Let me know).

[rvandewater]

@mmcdermott if there is no demo, should that command simply not do anything? (see failing test)

[mmcdermott]

@mmcdermott if there is no demo, should that command simply not do anything? (see failing test)

In that case I think there should be no demo section at all. But it looks like you already have that, so clearly there is a MEDS-DEV infrastructure issue. I think this is an issue with the testing infrastructure, where datasets without a demo version should not be tested. I'll need to take a look and fix it in the conftest file -- in the meantime it shouldn't block this PR likely.