How to handle“>1 million mL”fluid records in inputevents?

[dbcoffee]

Problem Description

While analysing the inputevents table in MIMIC-IV v2.2 we noticed a non-negligible number of rows with extremely high fluid amounts (>10 000 mL, sometimes even >1 000 000 mL). These outliers severely bias daily intake calculations.

Example of the query used

SELECT 
    starttime, endtime, itemid, amount, amountuom, rate, rateuom,
    ordercategoryname, ordercategorydescription, statusdescription
FROM mimiciv_icu.inputevents 
WHERE amount > 10000 AND amountuom = 'mL'
ORDER BY amount DESC;

One extreme row (among many)

starttime	endtime	itemid	amount	amountuom	rate	rateuom	ordercategoryname	ordercategorydescription	statusdescription
2178-10-30 07:41	2178-10-30 23:00	225828	1 000 400 mL	mL	65 314 mL/h	mL/hour	02-Fluids (Crystalloids)	Continuous IV	FinishedRunning

That is more than one million millilitres supposedly infused in ~15 h.

Questions

1. Do these rows represent:

• simple data-entry errors,
• conversion unit bugs (e.g. micro-units accidentally scaled to mL), or
• legitimate ICU procedures such as continuous bladder irrigation, pre-admission fluid totals, machine priming volumes, etc.?

2. What is the recommended strategy for daily-fluid-balance studies?

• Drop any row above a physiologic threshold (e.g. >1 000 mL/h)?
• Keep them but cap the rate?
• Flag and exclude only when ordercategoryname / itemid matches a known “non-intravascular” purpose?

3. Are there official data-quality notes or cleaning scripts from the MIMIC team that address this issue?

Any experience, institutional protocol, or code snippet you can share will be highly appreciated!