How to add `caregiver` concept?

[mlondschien]

I am interested in variables that describe possible heterogeneity in the data. One such variable is cgid (caregiver-id) in table chartevents MIMIC III:

CGID is the identifier for the caregiver who validated the given measurement

(ref).

I would like to add a concept that returns, for each hour where at least one measurement was entered into the system, the list of caregivers involved. I tried the following:

  "caregiver": {
    "category": "misc",
    "sources": {
      "mimic": {
        "table": "chartevents",
        "val_var": "cgid",
        "target": "ts_tbl",
        "class": "col_itm"
      }
    }
  },

However, this raises

Error in eval(assertion, env) : 
  argument "ids" is missing, with no default

How should I go about adding such a concept?

[prockenschaub]

I think you just forgot to wrap the item definitions in a list [].

"caregiver": {
    "category": "misc",
    "sources": {
      "mimic": [{
        "table": "chartevents",
        "val_var": "cgid",
        "target": "ts_tbl",
        "class": "col_itm"
      }]
    }

The hint was in the traceback, where it was trying to create a sel_itm (the default) despite you specifying a col_itm. One further thing to look out for is that you currently defined this as a numeric concept. This means that by default, multiple rows with the same time stamp (= hourly bin) will be aggregated by the median. This is most likely not what you are looking for, so you will want to use unique as an aggreation.

load_concepts("caregiver", src = "mimic", aggregate = unique)

You should be able to specify this in the json too (see "dobu_dur" for an example of how) but it currently throws an error as the string is passed to dt_gforce which has a very narrow definition of what should be considered an aggregation function --- probably something we should change.

[mlondschien]

Thanks for the hint, also with unique :).

Followup: I want to implement a provider concept. In MIMIV-IV (2.2), the prescriptions table has a column order_provider_id I'd like to add as a ts_tbl via "index_var": "starttime". If I define

    "provider": {
        "category": "misc",
        "aggregate": "unique",
        "sources": {
            "miiv": [
                {
                    "table": "admissions",
                    "val_var": "admit_provider_id",
                    "class": "col_itm",
                    "target": "id_tbl"
                },
                ...
                {
                    "table": "poe",
                    "val_var": "order_provider_id",
                    "class": "col_itm",
                    "target": "ts_tbl"
                },
                {
                    "table": "prescriptions",
                    "val_var": "order_provider_id",
                    "class": "col_itm",
                    "target": "ts_tbl",
                    "index_var": "starttime"
                }
            ]
        }

and run ricu::load_concepts("provider", "miiv", aggregate = unique), this raises ! is_target(x = x, dat = res) is not TRUE. If I replace ts_tbl with win_tbl:

                {
                    "table": "prescriptions",
                    "val_var": "order_provider_id",
                    "class": "col_itm",
                    "target": "win_tbl",
                    "index_var": "starttime",
                    "dur_var": "stoptime"
                }

this yields

Error in colnamesInt(x, names(on), check_dups = FALSE) : 
  argument specifying columns specify non existing column(s): cols[3]='stoptime

(stoptime is a column of prescriptions).

What am I doing wrong here?

[prockenschaub]

I am afraid I haven't installed the MIMIC IV version 2.2 needed tor the provider IDs. However, my best guess is that you are trying to combine id_tbls and ts_tbl. The result coming from admissions is an id_tbl, whereas the others you listed are ts_tbls. The provider concept itself expects to receive ts_tbls (this is the default for concepts and can be changed by the "target" attribute, see for example "sex" or "ett_gcs").

You get the exact same error if you remove ts_to_win_tbl(mins(1L)) from the callback of AUMC's "ett_gcs" concept. You likely need another callback function id_to_ts_tbl that specifies how the id_tbl in your example should be upgraded to a ts_tbl.

[mlondschien]

Thanks @prockenschaub!

Together with https://github.com/eth-mds/ricu/pull/40, the following json

    "provider": {
        "category": "misc",
        "aggregate": "unique",
        "sources": {
            "miiv": [
                {
                    "table": "admissions",
                    "val_var": "admit_provider_id",
                    "class": "col_itm",
                    "index_var": "admittime",
                    "target": "ts_tbl"
                },
                ...
                {
                    "table": "poe",
                    "val_var": "order_provider_id",
                    "class": "col_itm",
                    "target": "ts_tbl"
                },
                {
                    "table": "prescriptions",
                    "val_var": "order_provider_id",
                    "class": "col_itm",
                    "target": "ts_tbl",
                    "index_var": "starttime"
                }
            ]
        }

works.

For the resulting miiv table, for 1081583 out of 5816262 rows, the index_var is negative (47508 of these from admissions). Am I correct to assume this is due to admissions / procedures from before the patient was transferred from the hospital to the ICU?

[prockenschaub]

Yes, if you use the ICU stay as your time origin (default), negative times relate to things that happened earlier outside of ICU.

Closing this issue now because I think it's resolved.