MPOG DataDirect:
QI and Advanced Approaches
Vikas O’Reilly-Shah, MD, PhD, FASA
Director, Center for Perioperative and Pain initiatives in Quality Safety and Outcomes (PPiQSO)
Associate Chief for Perioperative Informatics and Outcomes, Seattle Children’s Hospital
Professor of Anesthesiology & Pain Medicine, University of Washington
Are the point of care hemoglobin values significantly
different from those sent to the lab?
QI Use Case:
POC vs Lab Hemoglobin
• Approach:
– Use DataDirect exclusively
– Find cases where there were more than one hemoglobin lab value within 10 minutes of each other.
POC vs Lab Hemoglobin
• Approach:
– Use DataDirect exclusively
– Find cases where there were more than one hemoglobin lab value within 10 minutes of each other.
POC vs Lab Hemoglobin
• Approach:
– Use DataDirect exclusively
– Find cases where there were more than one hemoglobin lab value within 10 minutes of each other.
– Sleuthing: Use ‘same minute’ data for other blood gas data points (pO2, etc) to determine which we POC
Hgb
– “As our lab system cannot easily differentiate point of care values from central lab values, those
hemoglobin values with an
– exact timestamp match to a blood gas measurement of arterial PaO2 were considered as point of
care values. Those hemoglobin values
– without an exact timestamp match were considered as central lab run values.
– Then, only matched samples within 10 minutes of each other where there was exactly one point of
care value and exactly one central lab value were used.”
Bland-Altman plot generated from this data
POC vs Lab Hemoglobin
MPOG DataDirect: Limitations and Workarounds
• An AWESOME tool for
– cohort identification
– rapid extraction of data for QI questions
– rapid delivery of data for projects
• But only if
– … if the filter you need is available in DataDirect
– … making use of data entirely available in the MPOG data extract
– … that data is mapped AND available as a “Report Column” in DataDirect
– … if less than 100,000 cases
– … if less than 1,000,000 rows in the secondary tables (e.g., physiologic data)
ASPIRE Reports to Individual
Providers
What is the incidence, demographic pattern, and airway
management strategies associated with patients
presenting for surgery taking a GLP1-RA (e.g. how many
get an RSI for elective surgery)?
Advanced Use Case:
GLP1-RA
• Home medications are a mappable concept in MPOG, but not filterable by the value
GLP1-RA
• Approach
–Leverage the powerful MPOG phenotypes for
induction approach (e.g. RSI) which are NOT stored in
the local copy of the MPOG database
–SQL query against our local MPOG database to identify
those on GLP1 receptor agonists
–Join and compare cases with GLP1-RA listed vs not
* - The final query used
included brand names as
well.
SQL Server Management Studio
GLP1-RA
• Minimum necessary problem
– Can’t filter in DataDirect on a manually identified set of
MPOG Case IDs
– As the honest broker for the project, we used
DataDirect to iteratively pull this data for our entire
case list of patients (100,000 case row limit) within the
timeframe of interest
All MPOG
Cases
GLP1
How did you write that beautiful SQL?
• I didn’t. Claude did. Trick: download the MPOG schema to use
as context for the LLM
How did you write that beautiful SQL?
This SQL query was, of
course, buggy to start
with. Had to iterate with
the LLM and hand edit
in places. But provided
a quick and
comprehensive starting
point for writing the
query.
Does primary language of care influence rates of
premedication, parental presence on induction, and on
ratings of difficulty of induction?
Advanced Use Case:
Language and Induction Strategy
• Approach
–Use DataDirect for primary extraction of
demographic/case characteristics (premedication)
–Use the MPOG Application Suite to join to MRN
–Use MPOG local SQL query for unmapped data
elements (induction difficulty, parental presence)
–Use enterprise data warehouse for language of care
Language and Induction Strategy
Thank You
