mimic-code
mimic-code copied to clipboard
MIT-LCP
Regarding the availability of mortality in MIMIC-IV 1.0 data
According to the document at https://mimic.mit.edu/docs/iv/modules/core/patients/ for MIMIC-IV data v1.0 which is the version of data available now, I quote Out-of-hospital mortality is currently unavailable as of MIMIC-IV v1.0. This means that 28-day mortality or, say, 90-day, 30-day mortality data is not available for all patients, or to be specific, not for those stay in hospital for more than 28 days or so. We at our best are able to extract ICU or in-hospital mortality. But I could not help but noticing articles using 28-day mortality or others as outcomes. To list two caught my eyes here:
-
Zhou, S., Zeng, Z., Wei, H. et al. Early combination of albumin with crystalloids administration might be beneficial for the survival of septic patients: a retrospective analysis from MIMIC-IV database. Ann. Intensive Care 11, 42 (2021). https://doi.org/10.1186/s13613-021-00830-8
...This study was a restrictive observation study from the Medical Information Mart for Intensive Care IV (MIMIC-IV version 0.4) ...The primary outcome was 28-day mortality. Secondary outcomes included 60-day mortality...
-
Zhu, Youfeng, et al. SAPS III is superior to SOFA for predicting 28-day mortality in sepsis patients based on sepsis-3.0 criteria. International Journal of Infectious Diseases (2021). https://doi.org/10.1016/j.ijid.2021.11.015
...Authorization was obtained to access the MIMIC IV database (Date of Agreement: January 4, 2021)....Data regarding 28-day mortality after ICU admission...
Also, it may not be the same but I also spotted that a paper listing Alistair Johnson and Tom Pollard as co-authors included 1-year mortality as the primary outcome. MIMIC-III data was used in the paper but I quote again from the document at https://mimic.mit.edu/docs/iii/about/identifiers/, the DOD_SSN column in PATIENTS table from MIMIC-III data contains dates of death up to 90 days in the future for Metavision patients. It contains dates of death up to 4 years in the future for CareVue patients. But it didn't seem like only the CareVue part of the data was used for the analysis, or at least not mentioned explicitly in the paper.
-
Serpa Neto, A., Deliberato, R.O., Johnson, A.E.W. et al. Mechanical power of ventilation is associated with mortality in critically ill patients: an analysis of patients in two observational cohorts. Intensive Care Med 44, 1914–1922 (2018). https://doi.org/10.1007/s00134-018-5375-6
Secondary outcomes included ICU, 30-day, and 1-year mortality....The MIMIC-III database ...of whom 3846 patients admitted from 2001 till 2012 had complete data...
I get confused with all these papers doing something that doesn't seem to be possible with the public available data. Am I missing something here? Any hints are welcome!
With MIMIC-III data we had dates of death outside of the hospital stay. However our source of dates of death became extremely poor sometime in ~2013, and so we no longer have it available for the majority of patients in MIMIC-IV. If papers are using 28-day mortality in MIMIC-IV, then they are misleading their readers unless they somewhere clarify it is 28-day mortality within the hospital.
@alistairewj Alistair,
Did BIDMC stop getting the Social Security Death Index in 2013? Or did they just mess up and not correlate it with their clinical data? Having accurate DOD data has been a real boon to research because it's a more challenging prediction problem post-hospitalization than in-hospital or in-ICU mortality, but somewhat easier than a lot of other clinical prediction problems. Is there some way to get this for MIMIC-IV again? It can't be done by users because one would need access to the SSN, which the hospital has (at least had), but which they certainly can't release.
@alistairewj Alistair,
Did BIDMC stop getting the Social Security Death Index in 2013? Or did they just mess up and not correlate it with their clinical data? Having accurate DOD data has been a real boon to research because it's a more challenging prediction problem post-hospitalization than in-hospital or in-ICU mortality, but somewhat easier than a lot of other clinical prediction problems. Is there some way to get this for MIMIC-IV again? It can't be done by users because one would need access to the SSN, which the hospital has (at least had), but which they certainly can't release.
So.. fun story.. it's actually because of a legislative change which forbid the social security death index from collecting deaths from state databases. The result was a ~40% drop in the capture of deaths - it's no longer of sufficient quality to use as a source of out-of-hospital mortality. I looked into purchasing the data from the CDC, which would cost around $70k USD for MIMIC-IV. Still looking for grant opportunities!
With MIMIC-III data we had dates of death outside of the hospital stay. However our source of dates of death became extremely poor sometime in ~2013, and so we no longer have it available for the majority of patients in MIMIC-IV.
pity. Such information is really helpful for research purpose.
If papers are using 28-day mortality in MIMIC-IV, then they are misleading their readers unless they somewhere clarify it is 28-day mortality within the hospital.
That's what I was thinking. maybe should write a letter to the editors to get a clarify then.
So.. fun story.. it's actually because of a legislative change which forbid the social security death index from collecting deaths from state databases. The result was a ~40% drop in the capture of deaths - it's no longer of sufficient quality to use as a source of out-of-hospital mortality. I looked into purchasing the data from the CDC, which would cost around $70k USD for MIMIC-IV. Still looking for grant opportunities!
That certainly need some effort! wish you good luck and thanks for your hard work again!
Just closing this stale issue; versions of MIMIC-IV from 2.0 onward (now on 2.2) all have out of hospital mortality data - https://physionet.org/content/mimiciv/2.2/#release-notes