patrickthoral
commented
1 year ago 1) Of course this depends on how you would define a surgical patient, but in the definition used to develop the script, the idea was to only consider patients that required surgery leading to an ICU admission. When you look at the actual problems (value) those are mainly medical problems (sepsis, respiratory failure after surgery), e.g.
non_surgical_sql = """
SELECT * FROM listitems WHERE itemid IN
(
13141, --D_Algemene chirurgie_Algemeen
16642 --DMC_Algemene chirurgie_Algemeen
)
"""
diags = pd.read_sql(non_surgical_sql,con)But it could be argued that when those diagnostic categories were developed, we did consider them as 'surgical' patients, so it's reasonable to include them.
2) You are right that one of the main reasons was to reduce false positives. In the Netherlands we routinely adminster antibiotics (cefotaxim and SDD paste/suspension) and take cultures for selective digestive tract decontamination (SDD) to prevent ventilator-associated pneumonia. The means taking sputum, oropharyngeal and perineal cultures does not necessarily signify sepsis, at least not on Dutch ICUs. The list of cultures is perhaps slightly more restrictive than needed (taking the SDD process into account), so you could consider adding some of them, but it would be quite uncommon to draw these cultures without starting antibiotics:
8418, --Urinekweek afnemen
9191, --Drainvochtkweek afnemen
9192, --Faeceskweek afnemen -> Clostridium
9194, --Liquorkweek afnemen: --> especially Some combinations of antibiotics are also very often as prophylaxis, and not for empirical treatment of sepsis. For sepsis: metronidazol should be combined with ceftriaxon (or another antibiotic), co-amoxiclav normally would have been changed to ceftriaxon/metronidazol (and or other antibiotics) and co-trimoxazol is basically never used empirically as the sole antibiotic. In the rare case that (for our institution) prophylactic antibiotics have been used exclusively, the combination with cultures drawn should lead to a sepsis flag, but if the culture information is missing (i.e. not documented) you indeed might miss patients suffering from sepsis.
This is definitely an area to improve upon, and for the next version of AmsterdamUMCdb we are looking into getting both the actual culture results and the (billing) diagnoses at discharge.
In 'concepts/diagnosis/reason_for_admission.ipynb' script:
1) The following are marked as diagnoses variables, but not surgical. Is this correct? And if so, why are they not surgical? --Not surgical: 13141, --D_Algemene chirurgie_Algemeen --Not surgical: 16642, --DMC_Algemene chirurgie_Algemeen
2) Most cultures (as well as antibiotics including metronidazole, co-trimoxazol, co-amoxiclav) are excluded from the definition of sepsis because they are 'routinely used'. These seem relevant to a diagnosis of sepsis, so I was wondering how this was determined? Including these may result in false positives in the data, but excluding them may perhaps lead to many more false negatives. Is it just the case that specificity was deemed more important in this script here than sensitivity?