Need to agree on a subset of terms for ICU admission source.

Current options:

pd.crosstab(df['hospital_admit_source'],df['data_source'])

Used in pao2/fio2 calculation. Worth checking if fio2 is ever missing for pao2 in the lab table.

Right now nurse charting gets 90% of respiratory rates

select
count(pt.patientunitstayid) as num_total
, count(vnc_d1.resprate_max) as rr
, count(vnc_d1.resprate_max)::numeric / count(pt.patientunitstayid) * 100.0 as percentage_complete
from gossis_cohort pt left join gossis_vital_nc_d1 vnc_d1 on pt.patientunitstayid = vnc_d1.patientunitstayid
where excluded=0;

If we add in respiratory rate from respiratoryCharting then we could prob bump this up from 90%. Then have to decide whether it is worth it to add in vitalPeriodic to "clean up" the rest.

For the paper:

• Need the total dataset sizes for each data_source, regardless if it was included in the final gossis-data-2018-03-20.csv.gz dataset.
• Bookkeeping for ANZICS/eICU to where the point of gossis-data-2018-03-20.csv.gz (e.g., pre-2014 admissions, <18, etc).

For ANZICS we only have APACHE blood gases (pH, PaO2, PaCO2, FiO2). For eICU, we can get min/max (~35% completion), or APACHE (~23% completion). Regardless, we'd need to think of a sensible way of merging APACHE worst with min/max value.

SATIQ admission sources (e.g. icu_admit_source) are coded with integers 0-14, and need to be converted into standardized text.

The file "apache2-definitions.csv" which is referred to in merge-data.ipynb is missing. Without this file, GOSSIS will not properly be able to run on the dataset.

Age is available as age in NICST and should be added in to the final dataset.

eICU

• eICU BMI is all zeros
• eicu po2 max accidentally coded as min?
• eicu maximum diasbp is very high OR anzics diasbp is very low
• eicu missing pao2:fio2 max ratio

anzics

• anzics minimum diasbp usually lower
• missing d1 glucose for anzics
• hco3 anzics units are wrong
• hct for anzics is missing
• anzics potassium min/max units are wrong
• anzics sodium min/max units are wrong
• temperature source in eicu vs anzics?
• fio2 present in anzics?
• apache glucose missing from anzics
• anzics missing hematocrit_apache
• anzics sodium_apache units wrong

mimic

• inr mimic max/min switched?
• mimic ph_apache is it arterial? has high pHs

MIMIC in general: how to define apache variables if score=0? e.g. messes up temperature

worth comment

• hgb difference in mimic
• hct difference in mimic
• mbp high in anzics

We should add a covariate which indicates if the patient was ventilated for the APACHE RR.

Need to add the APACHE comorbidities to the final dataset for ANZICS/eICU

At the moment we don't have a standardized range for valid data. This can be helpful in removing outliers/junk data present.

Philips suggest the following (based on APACHE).

Jesse applies a few different rules to determine first stay, e.g. age, death, etc. We should integrate these into the GOSSIS code.

This can be done in the gossis_cohort.sql code.

They should be replaced by NULL

We need to decide how best to handle missing data.

I've created a spreadsheet which lists all variables currently available in GOSSIS: https://docs.google.com/spreadsheets/d/13PUD9adciV8mCZ25A6MxrUtd9WXd28-ZCyBfpk0DMDc/edit?usp=sharing

Next to each variable is a drop down list with possible options on how to impute data. We can collaboratively contribute/comment/discuss in the spreadsheet.

Should the glucose variable be defined based on only lab measured glucose or should fingerstick measurements be allowed?