Question about how a blood gas sample is defined as arterial

[snowgato]

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.mit.edu/
  • Checked that your issue isn't already addressed: https://github.com/MIT-LCP/mimic-code/issues?utf8=%E2%9C%93&q=

Description

Hello,
I get confused about how a blood gas sample is classified as an arterial blood gas sample in mimic iv database.
In mimic_derived, there's a script, bg.sql, that defines mimic_derived.bg table containing specimen field supposed to designate the nature of the blood gas sample, wether it is arterial or not.
This field is actually always null in mimic_derived.bg table.
According to bg.sql, specimen field corresponds to labevents records with itemid equal to 52028.
But strangely, the label associated to itemid in d_labitems is Hematocrit, and actually there's not a single labevents record with itemid equal to 52028.

In bg.sql,
specimen_pred is set to specimen if not null,
else set to 'ART.' if specimen_prob>0,75
else it is null.
As specimen is always null,
finally specimen_pred is only set to 'ART.' if specimen_prob>0,75, else null.

Finally mimic_derived.first_day_bg_art is defined out of mimic_derived.bg table where specimen_pred is equal to 'ART.'
As a consequence, all recordings defined as arterial blood gas in first_day_bg_art actually rely on a probability computation specimen_prob and never on specimen tagged as 'ART.' before being sent to the laboratory for analysis.

If someone has an explanation about that, I'd greatly appreciate it.
Thanks

[alistairewj]

Hello, I get confused about how a blood gas sample is classified as an arterial blood gas sample in mimic iv database. In mimic_derived, there's a script, bg.sql, that defines mimic_derived.bg table containing specimen field supposed to designate the nature of the blood gas sample, wether it is arterial or not. This field is actually always null in mimic_derived.bg table. According to bg.sql, specimen field corresponds to labevents records with itemid equal to 52028. But strangely, the label associated to itemid in d_labitems is Hematocrit, and actually there's not a single labevents record with itemid equal to 52028.

Yes, this is a bug. I retained the wrong itemid here and it should instead be 52033.

In bg.sql, specimen_pred is set to specimen if not null, else set to 'ART.' if specimen_prob>0,75 else it is null. As specimen is always null, finally specimen_pred is only set to 'ART.' if specimen_prob>0,75, else null.

This is based off of a regression model I built to impute the specimen type if missing.

Finally mimic_derived.first_day_bg_art is defined out of mimic_derived.bg table where specimen_pred is equal to 'ART.' As a consequence, all recordings defined as arterial blood gas in first_day_bg_art actually rely on a probability computation specimen_prob and never on specimen tagged as 'ART.' before being sent to the laboratory for analysis.

If someone has an explanation about that, I'd greatly appreciate it. Thanks

This is definitely unideal! And in fact, I don't think the model is needed anymore, as we should always have the specimen now as we improved data collection in MIMIC-IV.

[alistairewj]

See also #923 .. kinda slow on the fix, sorry.

[alistairewj]

Comparison of the model's predictions with the ground truth:

Row	specimen_pred	specimen	count
1	null	null	14909
2	null	ART.	9169
3	null	CENTRAL VENOUS.	4479
4	null	GV.	3
5	null	MIX.	6803
6	null	VEN.	41833
7	ART.	null	14874
8	ART.	ART.	411402
9	ART.	CENTRAL VENOUS.	2993
10	ART.	GV.	2
11	ART.	MIX.	4043
12	ART.	VEN.	50702

Unfortunately still ~30,000 specimens without a label. I'll still remove the model and let individuals handle the missing data as they prefer.

[snowgato]

Hi,
Thanks for your answer and bug fix.
Have you modified accordingly first_day_bg_art in mimic_derived ?