Leveraging Electronic Health Records for Predictive Modeling
-
Upload
michigan-state-university-research -
Category
Data & Analytics
-
view
257 -
download
4
Transcript of Leveraging Electronic Health Records for Predictive Modeling
![Page 1: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/1.jpg)
Leveraging Electronic Health Records for Predictive Modeling
Marianne Huebner
![Page 2: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/2.jpg)
Hospital electronic health recordsSources Size [MB] Data types
Cohort 1.21 TS, C, N
Registration 1.07 C,N
Allergies 1.8 TS, N, T
Labs 312 TS, N, T
Pharmacy 174 TS, T, N
Clinical Notes 3.7 TS, T, N
Nurses Flow sheet 830 TS, T, C,N
Orders 405 TS, T
etc
TS= time stamp, C=categorical, N=numeric, T=text
![Page 3: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/3.jpg)
Models for different time intervals
Tables of 80% random sample
of Mayo patients:•Chart events•Cohorts•OP Notes•Allergies•Cllnical notes•Flowsheets•Labs•MAR•OP notes•NLP annotations
evaluated data
Matchedcomplications
info from NSQIP andGold Standard
No match to complications
info(6,740)
Joinedsurgeries with op
notes (proc. group, diag. group, proc.
mode)
Added rules engine features
No match torules engine
data(78)
Surgeries w/ matched
complications info (1,273)
80% random sample of surgeries(8,013)
Completed dataset (1,082)
Pre-surgery model dataset
(1,082)
30 Days Post Surgery
model dataset(898)
Defined surgeries
removed outliers
variable transformation
imputed data
No match toNLP
annotations(106)
Surgery Completion
model dataset (1,004)
Added NLP annotations
![Page 4: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/4.jpg)
Machine learning algorithms
![Page 5: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/5.jpg)
Enhanced recovery pathway
![Page 6: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/6.jpg)
![Page 7: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/7.jpg)
Mobile applications
• Identify potential problems (pain, diet management, complications)
• Clinician effort– Current medical records:
30 mi (95% navigation, 619 mouse clicks)
– New Tool: 4 min (25 mouse clicks)
![Page 8: Leveraging Electronic Health Records for Predictive Modeling](https://reader031.fdocuments.net/reader031/viewer/2022030317/58706f721a28ab48378b7329/html5/thumbnails/8.jpg)
Objective• Develop accessible and accurate guidance in the design and
analysis of observational studies.
70 statisticians from 15 countriesTopic groups: Missing data, Initial data analysis, Variable selection, Measurement error and misclassification, Study design, Evaluating prediction models, Causal inference, Survival analysis, High dimensional data analysis
STRATOS Initiative (STRengthen Analytical Thinking in Observational Studies)
Intended for applied statisticians and other data analysts with varying levels of statistical education, experience, and interests.