Unit conversion with the existing tasks

[rvandewater]

Hi, @mmcdermott @gkondas

I am looking to fit the tasks in different datasets to the following tasks:

"abnormal_lab/vital/hypotension/first_24h"
"abnormal_lab/blood_chemistry/metabolic_acidosis/first_24h"
"abnormal_lab/blood_chemistry/hyponatremia/first_24h"
"abnormal_lab/blood_chemistry/elevated_creatinine/first_24h"
"abnormal_lab/cbc/leukocytosis/first_24h"
"abnormal_lab/cbc/thrombocytopenia/first_24h"
"abnormal_lab/cbc/anemia/first_24h"

The issue is that several datasets report mmol/L instead of g/dL. Firstly, this is non-trivial, as we need to know the molecular mass (although we can use common conversions, such as those for hemoglobin).

Second: currently, the thresholds are in both the predicates.yaml and the task.yaml: see https://github.com/Medical-Event-Data-Standard/MEDS-DEV/blob/main/src/MEDS_DEV/tasks/abnormal_lab/blood_chemistry/elevated_creatinine/first_24h.yaml. I would think the preferred place to put them is in the predicates only, such that the unit can be dataset-agnostic. Is this a bug? Why is this redundancy here?

[mmcdermott]

A few things:

1. The right way to handle this is to have separate predicates for different units, that way the ranges can be specified in the relevant units, and a dataset only need have one defined to work. This may pose an issue as right now a predicate that is not defined will throw an error, so we'd need to define the predicate to point to a non-existent value, which is awkward, but viable.
2. The reason they are specified twice is because we want them to, in the end, be only specified in the task file, so that the range can be fixed and a dataset just needs to align the code, but right now that isn't well supported, so datasets have to define the value range too, and we want them standardized to something, so the task file has it as well just as guidance. This is related to Abnormal Lab Tasks permit 100% accuracy #237 which may be solved now with the task definitions being reverted to their prior definitions.
3. Ultimately, we'd like to improve ACES to support some more nuanced forms of inheritance to support this, but it isn't fully well supported right now. We should use this issue as a forcing function to address it.

[rvandewater]

@mmcdermott So currently it is defined like:

  # Plain predicates
  creatinine:
    code: ???
  abnormally_high_creatinine:
    code: ???
    value_min: 1.3 # mg/dL
    value_min_inclusive: False
    value_max: null

What you suggest is that we get:

  # Plain predicates
  creatinine:
    code: ???
  abnormally_high_creatinine_mg_dl:
    code: ???
    value_min: 1.3 # mg/dL
    value_min_inclusive: False
    value_max: null

  abnormally_high_creatinine_mmol_l:
    code: ???
    value_min: 1.3 # mg/dL
    value_min_inclusive: False
    value_max: null

?

[mmcdermott]

You'd need two base creatinine as well; one for each unit, then we'd need to include a derived predicate that is an OR of all of the different supported units.