Full set of cofounders for eICU

mchhoy commented 6 years ago

These are the ones which are on the todo list:

apsiii
elix
angus
sofa
fluid_balance
mechanical_ventilative_volume (tidal_high_count, total_count, invasive)

Need to have a look around to find them, e.g. apsiii is here:

harounchahed commented 6 years ago

I tried to do this today following the mimic codes but it's still very vague. I suggest we work on directions on where this data is on eICU (like we did with MIMIC on this doc) and then querying should be easy. What do y'all think?

mchhoy commented 6 years ago

Yes, we should see what sort of things can be extracted, given that we don't have a nice list like we did for MIMIC ...

Anyway has this been implemented already? Otherwise prob too early to close this issue

willemvandenboom commented 6 years ago

I've made a start with a table on where data is in #27, as @harounchahed suggested. I agree with @mchhoy that this issue is best left open until we're done with these eICU confounders. It's no problem if this issue remains open for a while. If desired, we can create more focused issues on specific aspects of obtaining these confounders.

mchhoy commented 6 years ago

Did @harounchahed start this already?

willemvandenboom commented 6 years ago

Not yet to my knowledge. Last time I had contact with @harounchahed was yesterday. Also, if someone starts working on this, it's best if they assign it to themselves here for others to see.

mchhoy commented 6 years ago

OK, have worked on this in #33, some notes:

For apsiii, I used "apachepatientresult", as it has a single scalar score like we used in MIMIC. There's more detail in "apacheapsvar" and "apachepredvar" but this is not being used. I verified that "apacheversion" was either "IV" or "IVa"
Googling "elix site:eicu-crd.mit.edu" has no hits. Likewise for angus.
For "sofa" the only appearance is the following phrase:

APACHE IVa further improved the equations and has been described as having the highest discrimination of any other adult risk adjustment model (SAPS 3, SOFA, MPM III).

I suspect there is a structural relationship between apsiii, elix, angus, sofa, etc - need to clarify with KC.

fluid_balance is based on intakeoutput, has values for ~180K patient stays.
For mechanical_ventilative_volume, I was looking through respiratoryCharting, there are many quantities:

results-20181021-205353.txt

I went with:

TV/kg IBW

and computed the proportion of the time it was over 6.5 (as done for MIMIC). It has data for 40K stay ID's. If you instead do something like:

WHERE respchart.respchartvaluelabel IN ("Tidal Volume (set)", "TV/kg IBW", "Exhaled TV (machine)", "Exhaled TV (patient)")

it only increases to 45K stay ID's.

For ventilation type, I previously did this:

https://github.com/nus-mornin-lab/oxygenation_kc/pull/31/commits/e2324341f7aba48a634bb705e60433462942315e

However I think @willemvandenboom had a better way to do in inside R, so I have omitted here.

Note I got excited when I saw "O2 Device" in respiratoryCharting, but it only has data for 6K stay ID's :(

willemvandenboom commented 6 years ago

I went with:

TV/kg IBW

and computed the proportion of the time it was over 6.5 (as done for MIMIC). It has data for 40K stay ID's. If you instead do something like:

WHERE respchart.respchartvaluelabel IN ("Tidal Volume (set)", "TV/kg IBW", "Exhaled TV (machine)", "Exhaled TV (patient)")

it only increases to 45K stay ID's.

For ventilation type, I previously did this:

e232434

However I think @willemvandenboom had a better way to do in inside R, so I have omitted here.

Yes, I agree this might be better done in R. Also, ventilation type is covered by #29.

I've assigned myself to this issue to work on ventilative volume. Since that is also a rather complex, time evolving quantity, I'll work on it in R, using the the code @mchhoy already provided here.

willemvandenboom commented 6 years ago

The attached picture is what the whiteboard at the datathon contained on confounders. Many are already mentioned here but I've asked KC about "decision life therapy" and the comorbidities mentioned with arrows on the right.

confounders

willemvandenboom commented 6 years ago

Regarding sepsis from Slack:

Based on the supplementary material (https://static-content.springer.com/esm/art%3A10.1038%2Fs41591-018-0213-5/MediaObjects/41591_2018_213_MOESM1_ESM.pdf), “Sepsis, including pneumonia” is determined using ICD-9 codes. We can do the same.

willemvandenboom commented 6 years ago

KC See noted that https://doi.org/10.1038/s41591-018-0213-5 extract the SOFA score from eRI, a database related to eICU. I've asked the author for his SQL code. Please find his response below.

Hi Willem,

Thanks for your interest. Here is my SQL code to compute SOFA in the eICU CRD. You'll need a public table called cohort1 containing a list of all the patientunitstayids. Hope this helps (and works)!

Matthieu

-- SOFA-cardiovascular

create table public.sofacv as
(
with t1 as -- MAP
(

with tt1 as
(
select patientunitstayid,
min( case when noninvasivemean is not null then noninvasivemean else null end) as map
from eicu.vitalaperiodic
where observationoffset between -1440 and 1440
group by patientunitstayid
), tt2 as
(
select patientunitstayid,
min( case when systemicmean is not null then systemicmean else null end) as map
from eicu.vitalperiodic
where observationoffset between -1440 and 1440
group by patientunitstayid
)

select pt.patientunitstayid, case when tt1.map is not null then tt1.map
when tt2.map is not null then tt2.map
else null end as map
from eicu.patient pt
left outer join tt1
on tt1.patientunitstayid=pt.patientunitstayid
left outer join tt2
on tt2.patientunitstayid=pt.patientunitstayid
order by pt.patientunitstayid
), t2 as --DOPAMINE
(
select distinct  patientunitstayid, max(
case when lower(drugname) like '%(ml/hr)%' then round(cast(drugrate as numeric)/3,3) -- rate in ml/h * 1600 mcg/ml / 80 kg / 60 min, to convert in mcg/kg/min
when lower(drugname) like '%(mcg/kg/min)%' then cast(drugrate as numeric)
else null end ) as dopa
from eicu.infusiondrug id
where lower(drugname) like '%dopamine%' and infusionoffset between -120 and 1440 and drugrate ~ '^[0-9]{0,5}$' and drugrate<>'' and drugrate<>'.'
group by patientunitstayid
order by patientunitstayid

), t3 as  --NOREPI
(
select distinct patientunitstayid, max(case when lower(drugname) like '%(ml/hr)%' and drugrate<>''  and drugrate<>'.' then round(cast(drugrate as numeric)/300,3) -- rate in ml/h * 16 mcg/ml / 80 kg / 60 min, to convert in mcg/kg/min
when lower(drugname) like '%(mcg/min)%' and drugrate<>'' and drugrate<>'.'  then round(cast(drugrate as numeric)/80 ,3)-- divide by 80 kg
when lower(drugname) like '%(mcg/kg/min)%' and drugrate<>'' and drugrate<>'.' then cast(drugrate as numeric)
else null end ) as norepi

from eicu.infusiondrug id
where lower(drugname) like '%norepinephrine%'  and infusionoffset between -120 and 1440  and drugrate ~ '^[0-9]{0,5}$' and drugrate<>'' and drugrate<>'.'
group by patientunitstayid
order by patientunitstayid

), t4 as  --DOBUTAMINE
(
select distinct patientunitstayid, 1 as dobu
from eicu.infusiondrug id
where lower(drugname) like '%dobutamin%' and drugrate <>'' and drugrate<>'.' and drugrate <>'0' and drugrate ~ '^[0-9]{0,5}$' and infusionoffset between -120 and 1440
order by patientunitstayid
)

select pt.patientunitstayid, t1.map, t2.dopa, t3.norepi, t4.dobu,
(case when dopa>=15 or norepi>0.1 then 4
when dopa>5 or (norepi>0 and norepi <=0.1) then 3
when dopa<=5 or dobu > 0 then 2
when map <70 then 1
else 0 end) as SOFA_cv –-COMPUTE SOFA CV

from public.cohort1 pt
left outer join t1
on t1.patientunitstayid=pt.patientunitstayid
left outer join t2
on t2.patientunitstayid=pt.patientunitstayid
left outer join t3
on t3.patientunitstayid=pt.patientunitstayid
left outer join t4
on t4.patientunitstayid=pt.patientunitstayid
order by pt.patientunitstayid
);

-- SOFA-RESPI

create table public.sofarespi as
(
with tempo2 as 
(
with tempo1 as
(
with t1 as –-FIO2 from respchart
(
select *
from
(
select distinct patientunitstayid, max(cast(respchartvalue as numeric)) as rcfio2
-- , max(case when respchartvaluelabel = 'FiO2' then respchartvalue else null end) as fiO2
from eicu.respiratorycharting
where respchartoffset between -120 and 1440 and respchartvalue <> '' and respchartvalue ~ '^[0-9]{0,2}$'
group by patientunitstayid
) as tempo
where rcfio2 >20 -- many values are liters per minute!
order by patientunitstayid

), t2 as –-FIO2 from nursecharting
(
select distinct patientunitstayid, max(cast(nursingchartvalue as numeric)) as ncfio2
from eicu.nursecharting nc
where lower(nursingchartcelltypevallabel) like '%fio2%' and nursingchartvalue ~ '^[0-9]{0,2}$' and nursingchartentryoffset between -120 and 1440
group by patientunitstayid

), t3 as –-sao2 from vitalperiodic
(
select patientunitstayid,
min( case when sao2 is not null then sao2 else null end) as sao2
from eicu.vitalperiodic
where observationoffset between -1440 and 1440
group by patientunitstayid

), t4 as –-pao2 from lab
(
select patientunitstayid,
min(case when lower(labname) like 'pao2%' then labresult else null end) as pao2
from eicu.lab
where labresultoffset between -1440 and 1440
group by patientunitstayid

), t5 as –-airway type combining 3 sources (1=invasive)

(

with t1 as –-airway type from respcare (1=invasive) (by resp therapist!!)
(
select distinct patientunitstayid,
max(case when airwaytype in ('Oral ETT','Nasal ETT','Tracheostomy') then 1 else NULL end) as airway  -- either invasive airway or NULL
from eicu.respiratorycare
where respcarestatusoffset between -1440 and 1440

group by patientunitstayid-- , respcarestatusoffset
-- order by patientunitstayid-- , respcarestatusoffset
),

t2 as –-airway type from respcharting (1=invasive)
(
select distinct patientunitstayid, 1 as ventilator
from eicu.respiratorycharting rc
where respchartvalue like '%ventilator%'
or respchartvalue like '%vent%'
or respchartvalue like '%bipap%'
or respchartvalue like '%840%'
or respchartvalue like '%cpap%'
or respchartvalue like '%drager%'
or respchartvalue like 'mv%'
or respchartvalue like '%servo%'
or respchartvalue like '%peep%'
and respchartoffset between -1440 and 1440
group by patientunitstayid
-- order by patientunitstayid
),

t3 as –-airway type from treatment (1=invasive)

(
select distinct patientunitstayid, max(case when treatmentstring in
('pulmonary|ventilation and oxygenation|mechanical ventilation',
'pulmonary|ventilation and oxygenation|tracheal suctioning',
'pulmonary|ventilation and oxygenation|ventilator weaning',
'pulmonary|ventilation and oxygenation|mechanical ventilation|assist controlled',
'pulmonary|radiologic procedures / bronchoscopy|endotracheal tube',
'pulmonary|ventilation and oxygenation|oxygen therapy (> 60%)',
'pulmonary|ventilation and oxygenation|mechanical ventilation|tidal volume 6-10 ml/kg',
'pulmonary|ventilation and oxygenation|mechanical ventilation|volume controlled',
'surgery|pulmonary therapies|mechanical ventilation',
'pulmonary|surgery / incision and drainage of thorax|tracheostomy',
'pulmonary|ventilation and oxygenation|mechanical ventilation|synchronized intermittent',
'pulmonary|surgery / incision and drainage of thorax|tracheostomy|performed during current admission for ventilatory support',
'pulmonary|ventilation and oxygenation|ventilator weaning|active',
'pulmonary|ventilation and oxygenation|mechanical ventilation|pressure controlled',
'pulmonary|ventilation and oxygenation|mechanical ventilation|pressure support',
'pulmonary|ventilation and oxygenation|ventilator weaning|slow',
'surgery|pulmonary therapies|ventilator weaning',
'surgery|pulmonary therapies|tracheal suctioning',
'pulmonary|radiologic procedures / bronchoscopy|reintubation',
'pulmonary|ventilation and oxygenation|lung recruitment maneuver',
'pulmonary|surgery / incision and drainage of thorax|tracheostomy|planned',
'surgery|pulmonary therapies|ventilator weaning|rapid',
'pulmonary|ventilation and oxygenation|prone position',
'pulmonary|surgery / incision and drainage of thorax|tracheostomy|conventional',
'pulmonary|ventilation and oxygenation|mechanical ventilation|permissive hypercapnea',
'surgery|pulmonary therapies|mechanical ventilation|synchronized intermittent',
'pulmonary|medications|neuromuscular blocking agent',
'surgery|pulmonary therapies|mechanical ventilation|assist controlled',
'pulmonary|ventilation and oxygenation|mechanical ventilation|volume assured',
'surgery|pulmonary therapies|mechanical ventilation|tidal volume 6-10 ml/kg',
'surgery|pulmonary therapies|mechanical ventilation|pressure support',
'pulmonary|ventilation and oxygenation|non-invasive ventilation',
'pulmonary|ventilation and oxygenation|non-invasive ventilation|face mask',
'pulmonary|ventilation and oxygenation|non-invasive ventilation|nasal mask',
'pulmonary|ventilation and oxygenation|mechanical ventilation|non-invasive ventilation',
'pulmonary|ventilation and oxygenation|mechanical ventilation|non-invasive ventilation|face mask',
'surgery|pulmonary therapies|non-invasive ventilation',
'surgery|pulmonary therapies|non-invasive ventilation|face mask',
'pulmonary|ventilation and oxygenation|mechanical ventilation|non-invasive ventilation|nasal mask',
'surgery|pulmonary therapies|non-invasive ventilation|nasal mask',
'surgery|pulmonary therapies|mechanical ventilation|non-invasive ventilation',
'surgery|pulmonary therapies|mechanical ventilation|non-invasive ventilation|face mask'
) then 1  else NULL end) as interface   -- either ETT/NiV or NULL
from eicu.treatment
where treatmentoffset between -1440 and 1440
group by patientunitstayid-- , treatmentoffset, interface
order by patientunitstayid-- , treatmentoffset
), t4 as
(
select distinct patientunitstayid,
max(case when cplitemvalue like '%Intubated%' then 1 else NULL end) as airway  -- either invasive airway or NULL
from eicu.careplangeneral
where cplitemoffset between -1440 and 1440
group by patientunitstayid -- , respcarestatusoffset

)

select pt.patientunitstayid,
case when t1.airway is not null or t2.ventilator is not null or t3.interface is not null or t4.interface is not null then 1 else null end as mechvent --summarize
from public.cohort1 pt
left outer join t1
on t1.patientunitstayid=pt.patientunitstayid
left outer join t2
on t2.patientunitstayid=pt.patientunitstayid
left outer join t3
on t3.patientunitstayid=pt.patientunitstayid
left outer join t4
on t4.patientunitstayid=pt.patientunitstayidorder by pt.patientunitstayid

)

select pt.patientunitstayid, t3.sao2, t4.pao2, 
(case when t1.rcfio2>20 then t1.rcfio2 when t2.ncfio2 >20 then t2.ncfio2  when t1.rcfio2=1 or t2.ncfio2=1 then 100 else null end) as fio2, t5.mechvent
from public.cohort1 pt
left outer join t1
on t1.patientunitstayid=pt.patientunitstayid
left outer join t2
on t2.patientunitstayid=pt.patientunitstayid
left outer join t3
on t3.patientunitstayid=pt.patientunitstayid
left outer join t4
on t4.patientunitstayid=pt.patientunitstayid
left outer join t5
on t5.patientunitstayid=pt.patientunitstayid
-- order by pt.patientunitstayid
)

select *, -- coalesce(fio2,nullif(fio2,0),21) as fn, nullif(fio2,0) as nullifzero, coalesce(coalesce(nullif(fio2,0),21),fio2,21) as ifzero21 ,
coalesce(pao2,100)/coalesce(coalesce(nullif(fio2,0),21),fio2,21) as pf, coalesce(sao2,100)/coalesce(coalesce(nullif(fio2,0),21),fio2,21) as sf
from tempo1
)

select patientunitstayid, 
(case when pf <1 or sf <0.67 then 4  --COMPUTE SOFA RESPI
when pf between 1 and 2 or sf between 0.67 and 1.41 then 3
when pf between 2 and 3 or sf between 1.42 and 2.2 then 2
when pf between 3 and 4 or sf between 2.21 and 3.01 then 1
when pf > 4 or sf> 3.01 then 0 else 0 end ) as SOFA_respi
from tempo2
order by patientunitstayid
);

-- SOFA-RENAL

create table public.sofarenal as
(
with t1 as --CREATININE
(
select pt.patientunitstayid,
max(case when lower(labname) like 'creatin%' then labresult else null end) as creat
from eicu.patient pt
left outer join eicu.lab
on pt.patientunitstayid=eicu.lab.patientunitstayid
where labresultoffset between -1440 and 1440
group by pt.patientunitstayid

), t2 as --UO
(

with uotemp as
(
select patientunitstayid,
case when dayz=1 then sum(outputtotal) else null end as uod1
from
(

select distinct patientunitstayid, intakeoutputoffset, outputtotal,
(CASE
WHEN  (intakeoutputoffset) between -120 and 1440 THEN 1
else null
end) as dayz
from eicu.intakeoutput
where intakeoutputoffset between 0 and 5760
order by patientunitstayid, intakeoutputoffset

) as temp
group by patientunitstayid, temp.dayz
)

select pt.patientunitstayid,
max(case when uod1 is not null then uod1 else null end) as UO
from eicu.patient pt
left outer join uotemp
on uotemp.patientunitstayid=pt.patientunitstayid
group by pt.patientunitstayid

)

select pt.patientunitstayid, -- t1.creat, t2.uo,
(case –-COMPUTE SOFA RENAL
when uo <200 or creat>5 then 4
when uo <500 or creat >3.5 then 3
when creat between 2 and 3.5 then 2
when creat between 1.2 and 2 then 1
else 0
end) as sofarenal
from public.cohort1 pt
left outer join t1
on t1.patientunitstayid=pt.patientunitstayid
left outer join t2
on t2.patientunitstayid=pt.patientunitstayid
order by pt.patientunitstayid
-- group by pt.patientunitstayid, t1.creat, t2.uo

);

-- SOFA- GCS, liver, platelets

create table public.sofa3others as
(
with t1 as --GCS
(
select patientunitstayid, sum(cast(physicalexamvalue as numeric)) as gcs
from eicu.physicalexam pe
where (lower(physicalexampath) like '%gcs/eyes%'
or lower(physicalexampath) like '%gcs/verbal%'
or lower(physicalexampath) like '%gcs/motor%')
and physicalexamoffset between -1440 and 1440
group by patientunitstayid, physicalexamoffset
), t2 as
(
select pt.patientunitstayid,
max(case when lower(labname) like 'total bili%' then labresult else null end) as bili, --BILI
min(case when lower(labname) like 'platelet%' then labresult else null end) as plt --PLATELETS
from eicu.patient pt
left outer join eicu.lab
on pt.patientunitstayid=eicu.lab.patientunitstayid
where labresultoffset between -1440 and 1440
group by pt.patientunitstayid
)

select distinct pt.patientunitstayid, min(t1.gcs) as gcs, max(t2.bili) as bili, min(t2.plt) as plt,
max(case when plt<20 then 4
when plt<50 then 3
when plt<100 then 2
when plt<150 then 1
else 0 end) as sofacoag,
max(case when bili>12 then 4
when bili>6 then 3
when bili>2 then 2
when bili>1.2 then 1
else 0 end) as sofaliver,
max(case when gcs=15 then 0
when gcs>=13 then 1
when gcs>=10 then 2
when gcs>=6 then 3
when gcs>=3 then 4
else 0 end) as sofacns
from public.cohort1 pt
left outer join t1
on t1.patientunitstayid=pt.patientunitstayid
left outer join t2
on t2.patientunitstayid=pt.patientunitstayid
group by pt.patientunitstayid, t1.gcs, t2.bili, t2.plt
order by pt.patientunitstayid
);

-- SOFA: COMBINE ALL SUBSCORES 

Select pt.patientunitstayid, --  sofacv.sofa_cv, sofarespi.sofa_respi,sofarenal.sofarenal,sofa3others.sofacoag,sofa3others.sofaliver,sofa3others.sofacns, 
sofacv.sofa_cv+sofarespi.sofa_respi+ sofarenal.sofarenal+sofa3others.sofacoag+ sofa3others.sofaliver+sofa3others.sofacns as sofatotal
From public.cohort1 pt
Left outer join sofacv
On pt.patientunitstayid=sofacv.Patientunitstayid
Left outer join sofarespi
On pt.patientunitstayid= sofarespi.Patientunitstayid
Left outer join sofarenal
On pt.patientunitstayid= sofarenal.Patientunitstayid
Left outer join sofa3others
On pt.patientunitstayid= sofa3others.Patientunitstayid

Dr Matthieu Komorowski Locum consultant in Intensive Care, Charing Cross Hospital PhD student, Dept of Surgery and Cancer & Dept of Bioengineering, Imperial College London Visiting Scholar, Laboratory of Computational Physiology, Harvard-MIT Health Sciences and Technology Please use matthieu.komorowski@imperial.nhs.uk for patient-related matters. Tel: +44 (0) 757 897 5175 ; SkypeID matthieukomorowski

willemvandenboom commented 5 years ago

From Slack conversation with KC, we need to add the following comorbidities:

Chronic liver disease Chronic obstructive pulmonary disease Cerebrovascular accident (stroke) Cancer (malignancy) Diabetes mellitus Hypertension

willemvandenboom commented 5 years ago

We forgot to extract the cofounder whether it is decided that someone is no longer on life therapy. I don't know how to interpret the carePlanEOL table. I've therefore asked in https://github.com/MIT-LCP/eicu-code/issues/65.

willemvandenboom commented 5 years ago

fd007cc addresses the life therapy question, using carePlanEOL amongst others. I therefore close this issue again.

nus-mornin-lab / oxygenation_kc

Full set of cofounders for eICU #18