Amsterdam UMC on Implementation: @drPaulElbers Conference conflicts of interest : chair of data science edition.
"Working AI to Implementation"
#criticalcare #DataScience #ai #aitobedside #LIVES2022
@ESICM
"Working AI to Implementation"
#criticalcare #DataScience #ai #aitobedside #LIVES2022
@ESICM
@drPaulElbers @ESICM PacMed Model to help discharge patients from ICU. Autokinetics for antibiotic dosing.
These 2 are currently implemented. @patrickthoral also involved in discharge models.
These 2 are currently implemented. @patrickthoral also involved in discharge models.
@drPaulElbers @ESICM @patrickthoral - ICU natural habitat
- large vol of data
- frequent decision making
BUT there is a problem in ICU and data.
- high variability
pubmed.ncbi.nlm.nih.gov/36173632/
- large vol of data
- frequent decision making
BUT there is a problem in ICU and data.
- high variability
pubmed.ncbi.nlm.nih.gov/36173632/
@drPaulElbers @ESICM @patrickthoral Help recommend Readiness scale. Shockingly most of the research remains in model development phase. see image
WHAT are the barriers for getting to bedside
1. you cannot develop model without data. - data availability problem
2. compliance with legislation and reporting
3. integration into work flow and EHR
4. local stakeholder engagement
5. model generalisability and data drift
1. you cannot develop model without data. - data availability problem
2. compliance with legislation and reporting
3. integration into work flow and EHR
4. local stakeholder engagement
5. model generalisability and data drift
6. Unpredictable interaction
7. all of this results in "prohibitive cost"
very difficult for individual hospitals to do. require national or private sector in his view.
Now discussing Dutch Data Warehouse combining Meta Vision , HiX and EPIC systems into 1 warehouse
7. all of this results in "prohibitive cost"
very difficult for individual hospitals to do. require national or private sector in his view.
Now discussing Dutch Data Warehouse combining Meta Vision , HiX and EPIC systems into 1 warehouse
bmj.com/content/377/bm…
Framework for DECIDE-AI consensus. The goal is to minimise paperwork and help with compliance, GDPR, Medical Device Regulation, Artificial Intelligence Act(coming). MDR is needed right at the very end.
Framework for DECIDE-AI consensus. The goal is to minimise paperwork and help with compliance, GDPR, Medical Device Regulation, Artificial Intelligence Act(coming). MDR is needed right at the very end.
Amount of docs required for the pipeline 
Pipeline
To get to bedside model - you need a lot more stakeholders - EHR , IT, Software engineers, Network engineers, Data Scientists, Ethical Boards.
Liu(2020) article : future of AI in critica care is augmented, not artificial.
Liu(2020) article : future of AI in critica care is augmented, not artificial.
Impressive piece of work overall in @amsterdamumc Incredible timeline and Gantt chart
@amsterdamumc FINAL TAKEHOME : cannot leave this for multi-national industry to solve this. Need ownership by patients and doctors. Hence perhaps we need "department of clinical artificial intelligence" in each hospital.
Insightful !!
Do you think we are there yet ^^ ?
Insightful !!
Do you think we are there yet ^^ ?
@amsterdamumc Question : difficulty in regulator, some regulators require submission even before model evaluation 😳 .
Challenge even within EU which is supposed to be more "uniform"
Q: Difficulties in collecting data in the context of GDPR.
Challenge even within EU which is supposed to be more "uniform"
Q: Difficulties in collecting data in the context of GDPR.
@amsterdamumc In his view, GDPR is strict but need to demonstrate that your data is anonymous. Aim for less than 1 in 20 and aim for GDPR anonymity framework. No body has been ever re-identified so far.
@amsterdamumc @threadreaderapp unroll
• • •
Missing some Tweet in this thread? You can try to
force a refresh
