Check SaO2 values in eICU

[joamats]

In eICU, SaO2 comes from "O2 Sat %", which does not clearly state whether it's venous or arterial. We may need to clamp some of the lowest values to make sure we're only considering SaO2 and not SvO2.

[mirkompcr]

Normal venous/SvO2 range is 60-70% Normal arterial/SaO2 range is 95-100%

Studies on SaO2 are hard to come by. But most people seem to be above 86%. As we have more than enough SaO2-SpO2 pairs, I would go for a very specific cut-off, accepting a low sensitivity. As such I would discard all SaO2 values < 80% because they could be venous.

References:

https://journals.physiology.org/doi/full/10.1152/japplphysiol.00394.2016?rfr_dat=cr_pub++0pubmed&url_ver=Z39.88-2003&rfr_id=ori%3Arid%3Acrossref.org

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0250740#sec012

https://bmjopenrespres.bmj.com/content/bmjresp/8/1/e000939.full.pdf

[mirkompcr]

Identifying patients with arterial lines

Use pivoted tables first:

pivoted vital https://github.com/MIT-LCP/eicu-code/blob/master/concepts/pivoted/pivoted-vital.sql

pivoted infusion -> all medications in there are vasopressors (except heparin) and are usually used with arterial catheters https://github.com/MIT-LCP/eicu-code/blob/master/concepts/pivoted/pivoted-infusion.sql

Then use original tables: infusiondrug -> see above nursecharting -> everyone with an invasive blood pressure SELECT distinct nursingchartcelltypevallabel, count(*) FROM physionet-data.eicu_crd.nursecharting where lower(nursingchartcelltypevallabel) like "%arter%" group by nursingchartcelltypevallabel order by nursingchartcelltypevallabel

medication --> SELECT distinct routeadmin, count(*) FROM physionet-data.eicu_crd.medication where lower(routeadmin) like "%arterial%" group by routeadmin order by routeadmin

vital aperiodic table SELECT distinct cardiacoutput, count(*) FROM physionet-data.eicu_crd.vitalaperiodic where cardiacoutput is not NULL AND cardiacoutput >0.8 group by cardiacoutput order by cardiacoutput vital periodic table SELECT patientunitstayid, observationoffset, systemicmean, pamean FROM physionet-data.eicu_crd.vitalperiodic where (systemicmean is not NULL and systemicmean > 0) OR (pamean is not NULL and pamean > 0)

[joamats]

Shared by "Barros, Andrew J (ajb5d)" ajb5d@virginia.edu Paper: https://doi.org/10.1016/j.chest.2021.01.038 Code: https://github.com/ajb5d/ArterialLineBacteremia Doesn't seem to be as exhaustive as Tristan's approach though

[joamats]

If we exclude patients without arterial line, we're excluding roughly 25% of the pairs. However, this may not make sense because we're not sure if the arterial line information is accurate and may be missing for some patients/hospitals. Further, a patient with an arterial line can have a "O2 Sat %" value that is actually venous. If arterial_line == 0 OR arterial_line == 1 -> Mean SaO2 = 93.96% If arterial_line == 0 -> Mean SaO2 = 94.13% If arterial_line == 1 -> Mean SaO2 = 93.91% The arterial line information does not seem to subset the data to a cohort where SaO2 values are higher. I'd discard this approach. We can simply consider SaO2 values over 80%