Precise time of in_hospital_mortality

[pszolovits]

I am trying to figure out more precisely when a patient dies during an ICU stay, but am somewhat confused by how to interpret the data that occur in MIMIC-IV. Here are some observations. I would appreciate any insights on these:

1. Time of death can be viewed from the dod field of the patients table, but only to the nearest date, not a more precise time. It can also be seen from the deathtime field of admissions, which is a timestamp.
2. According to hospital_expire_flag in admissions, 8609 patients died in the hospital. However, deathtime is non-null for only 8598 patients, so 11 died but have no specific time of death.
3. Comparing deathtime and dod, for nearly all of the 8598 patients who died in the hospital, in all but five cases, deathtime falls within the day specified in dod. However, there are five instances in which deathtime comes up to 94 hours after dod.
4. Although admissions frequently have multiple icustays associated with each, presumably a patient can only die during or after their latest ICU stay. Thus, for those who died, we should be able to align the patient, admission and icustay to see various ways one could precisely estimate how much time elapsed from ICU intime (my index time for most of these analyses) until death. However, death does not necessarily occur in the ICU, and there seem to be 1,639 patients who die during a time when they are not in the ICU, though they are in the hospital. 5,319 die during an ICU stay.
5. I wondered how ICU outttime relates to deathtime for those who die in the ICU. Oddly, although most dead people seem to have outtimes within a day of their deaths in the ICU, there are a few outliers who are not discharged from the ICU until as much as 140 hours after death. I don't understand exactly what ICU outtime really means in these cases.
   
6. More generally, it seems like it should be the case that any ICU stay (stay_id) that occurs during a particular hospitalization (hadm_id) should have its intime and outtime within the span of the hospitalization's admittime and dischtime. This is true in the data.
7. So, to return to my original goal, for those 5,319 patients who died in the ICU, I can get a reasonable estimate of how long they survived from the index time (intime) until their deaths. For these patients, the mean is 125 hours (a bit over 5 days), and the median is 61 hours, so half die within a very few days.
   

[XiaoYaHui]

Hi! I am wondering how you solve this problem? I am also confused about this.