NEXT: Who's "high risk" in ICU? Nicolas Bennett - Zurich, Switzerland.
reminds me of NELA score development in UK.
Declaration : he now works for industry (known side effect for medics doing data science )
@ESICM #criticalcare #ai #datascience #LIVES2022
reminds me of NELA score development in UK.
Declaration : he now works for industry (known side effect for medics doing data science )
@ESICM #criticalcare #ai #datascience #LIVES2022
@ESICM Interesting : eventually end up writing R package. cran.r-project.org/web/packages/r…
#sepsis classification using MIMIC-3 - time series data. Time-series classficiation.
#sepsis classification using MIMIC-3 - time series data. Time-series classficiation.
his labels are also earlier. main improvement is due to missing data Mx. "Missingness" is not random as assumption "more data available" in sicker pts.
Despite performance, did not evaluate further as want to move onto real-time risk development rather than retrospective.
Despite performance, did not evaluate further as want to move onto real-time risk development rather than retrospective.
Uses 4 data sets ; MIMIC-III, eICU Pollart et al, HiRID, AUMC Thoral et al.
Time-varying and static variables.
Prediction models considered : NN architectures with transformer models. baseline LGBM, GLM, features: NEWS, MEWS, SIRS,SOFA.
Time-varying and static variables.
Prediction models considered : NN architectures with transformer models. baseline LGBM, GLM, features: NEWS, MEWS, SIRS,SOFA.
AUROC, precision at 80% recall and median earliness.
Outlook : do we want to rely on treatment-related information? Misalignment between time dynamics of label, data availability and expectations of prediction risk. Uses vasopressor adjusted map value.
Outlook : do we want to rely on treatment-related information? Misalignment between time dynamics of label, data availability and expectations of prediction risk. Uses vasopressor adjusted map value.
eth-mds.github.io/icubdc/index.h… GitHub repo
interesting view from 1 of senior physicians :: SOFA score has nothing to do with sepsis. it was designed first to detect "abnormalities". Are we barking up the wrong tree ?
- 2nd Q: does prediction matter or makes any difference?speaker is non clinician so said hard to comment
- 2nd Q: does prediction matter or makes any difference?speaker is non clinician so said hard to comment
@threadreaderapp unroll
• • •
Missing some Tweet in this thread? You can try to
force a refresh
