Skip to content

Development & validation of clinical early warning models using multi-centred data.

Notifications You must be signed in to change notification settings

MLGlobalHealth/Early_Warning_Scores

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

77 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Development and validation of EWS systems 🚑

Contents

  • preprocessing

    Contains info on initial pre-processing of Electronic Health Records consisting of blood tests, diagnoses, various procedures, and intensive care data for individuals residing in Denmark, with a general admission to the hospitals in the region of Zealand, Denmark, between 2018-2023.

    • Original_Preprocessing.ipynb contains python code with an initial pre-processing of all datasets with clinical information.

    • Intensive_Care.R contains R code with more thorough analysis of intensive care data.

    • Diagnoses.R contains R code performing categorization/grouping of various ICD-10 diagnoses of patients.

    • Blood_Tests.py contains python code on imputation of blood tests containing string values not suitable for analysis.

    • Procedures.R contains R code on various medical procedures for each individual. Categorization of procedures (SKS-Codes) has been performed + text mining/topic modelling for the characterization of them.

  • Merging

    • EWS_Blood.py contains python code on merging of EWS (Early Warning Score) data of individuals with blood tests.

    • EWS_ITA.py contains python code on merging EWS + Blood Tests with Intensive Care data

    • EWS_Blood_ITA_Procedures.R contains R code on merging EWS + Blood Tests + Intensive Care with Procedures data

    • EWS_Final_Merging.py contains python code on the final merging of the datasets (diagnoses included)

  • modelling

    • Evaluating_NEWS2_Only.r :
      • Contains R code for the validation of NEWS2 system in terms of predictive performance
      • 🔗 Internal-External Cross-Validation (IECV) based on hospitals
      • 🔗 AUC, Brier Score, Calibration, Net Benefit
      • 🔗 Thresholds added in the Decision Curve Analysis
    • IECV_NEWS2.r :
      • Contains R code on IECV with a meta-analysis approach
    • Development_Comparisons.r :
      • Contains R code comparing various models and algorithms with the current NEWS2 system
        • 🔗 NEWS2-Light: NEWS2 - Blood Pressure - Temperature
        • 🔗 IEWS: NEWS2 + Age + Sex
        • 🔗 TREE-EWS: XGBoost with Age,Sex,Vital Signs, Previous Hospitalization & Blood Tests
        • 🔗 Weighted performance metrics
  • To do list:

    • Assessment of NEWS2 current system based on predictive performance metrics using data-splitting techniques ✅.

    • De-biasing the dataset with IPW based on intervention scenarios ✅

    • Development of alternative early warning score systems and model comparison ✅

    • Add calibration plots for the newly developed models ✅

    • Assess sustained recovery prediction of NEWS2 ✅

    • Assess performance on various strata of target population / Assess fairness 🔨

    • Try a DL architecture as an additional benchmark 🚩

    • Create a Table 1 ✅

About

Development & validation of clinical early warning models using multi-centred data.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published