Medication administration times

Prerequisites

• [X ] Put an X between the brackets on this line if you have done all of the following:
  • Checked the online documentation: https://mimic.physionet.org/about/mimic/
  • Checked that your issue isn't already addressed: https://github.com/MIT-LCP/mimic-code/issues?utf8=%E2%9C%93&q=

Description

I am currently working on a project where I would require (somewhat) accurate medication administration of loop diuretics. It is currently unclear which table should be used in order to get this - I would appreciate any help/input on how to reconcile this problem.

Here is what I have:
1 - a table with all diuretics administered based on icustay_id, and charttime/starttime pulled from inputevents_cv, inputevents_mv, and prescriptions table. There appears to be some repeated measured (albeit inconsistent) between the input tables and the prescription table, with the latter being less accurate re: times (as per the documentation) but with the advantage of capturing PO doses. I have not yet included the carevue data because I am not 100% clear how the medication administration (in terms of IV bolus, PO, etc.) is arranged.

How would I go about ensuring that I have the most accurate doses and times from loop diuretics?

Here is a sample:
with t1 as
(
select
icustay_id, startdate, enddate, drug_name_generic as drug, prod_strength as drugStr, dose_val_rx as dose, dose_unit_rx as units, route
, case
-- LD's
when lower(drug_name_generic) like '%furosem%' then 1
when lower(drug_name_generic) like '%bumetan%' then 1
when lower(drug_name_generic) like '%torse%' then 1
when lower(drug_name_generic) like '%ethacryn%' then 1
-- TZ
when lower(drug_name_generic) like '%thiazide%' then 1
when lower(drug_name_generic) like '%indapa%' then 1
when lower(drug_name_generic) like '%metolaz%' then 1
-- K-sparing
when lower(drug_name_generic) like '%amilori%' then 1
when lower(drug_name_generic) like '%epleren%' then 1
when lower(drug_name_generic) like '%spirono%' then 1
when lower(drug_name_generic) like '%triamt%' then 1
-- CARB ANH inh
when lower(drug_name_generic) like '%acetazo%' then 1
when lower(drug_name_generic) like '%methazo%' then 1
else 0
end as diuretic
, case when prod_strength like '%CAP%' and CAST(dose_val_rx as float64) < 10 then -- extract the caps into their actual doses
STRUCT((CAST(dose_val_rx as float64)*CAST(regexp_extract(prod_strength, r"[.0-9.]+") as float64)) as d1
, (CAST(dose_val_rx as float64)*CAST(regexp_extract(prod_strength, r"[.0-9.]+", regexp_instr(prod_strength, r"[/]")) as float64)) as d2)
else (null, null)
end as doseCorr
from physionet-data.mimiciii_clinical.prescriptions
)

, mv as
(
SELECT icustay_id, starttime, endtime, round(amount,1) as amount, amountuom, ordercategoryname
from physionet-data.mimiciii_clinical.inputevents_mv
where itemid = 221794 and lower(ordercategoryname) like "%bolus%" and icustay_id is not null and amount >1 and STATUSDESCRIPTION != "Rewritten"
order by icustay_id, starttime
)

select
icustay_id, timestamp(startdate) as startTime, timestamp(enddate) as endTime, drug, drugStr, dose, units, doseCorr.d1, doseCorr.d2, route
from t1
where diuretic = 1 and icustay_id is not null and route not like '%DRIP%' and dose is not null
UNION ALL
select
icustay_id, timestamp(starttime) as startTime, timestamp(endtime) as endTime, "Furosemide", null, cast(amount as string) as dose, amountuom as units, null,null, ordercategoryname as route
from mv
order by icustay_id, startTime, drug

This is joined to a furosemide infusion table that is generated mostly like the vasopressor infusions (did not include given the length of code and availability on the repository).

Here is a sample output for icustay_id 205385:

It can been seen that there are repeated doses (which are the additions of the prescription table and inputevents_mv). Any input would be greatly appreciated (I can provide further clarification if needed).

Joey

Start with the MIMIC documentation and compare the prescriptions, inputevents_cv, and inputevents_mv tables referenced in the issue, including the sample query. Determine how medication routes, doses, and administration times should be reconciled without duplicate records; done means the source tables and reconciliation approach are clearly documented for loop diuretics.