Predicted Risk of Opioid Use Disorder

Revision as of 16:54, 15 June 2020 by Smcdougall (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Student: Sarah McDougall
Mentor: Dr. Praveen Madiraju

Project Description

The prevalence of opioid misuse is rising, as evidenced by the increase in hospital readmissions linked to opioid abuse and dependence and the increase in opioid overdose deaths. This project uses deidentified hospital records to find health-related and social determinants that can be used to assess the risk of a subsequent opioid overdose following an opioid-related overdose. Furthermore, the project will use the hospital data to assess the risk of 30-day and 90-day all-cause readmissions following an opioid-related overdose. Machine learning models will be implemented on the hospital dataset to predict the risk that a patient develops opioid use disorder.

The project can be broken down into two areas of research:

  • Prediction of risk of opioid use disorder and opioid-related overdose
  • Prediction of risk of hospital readmission following opioid-related overdose (30-day and 90-day, all-cause and opioid-related)

Project Goals

  • Gain understanding of the opioid epidemic’s severity and lasting effects on society and the healthcare system
    • This requires learning basic medical terminology surrounding opioid use.
  • Read related works that discuss risk factors for opioid use disorder, opioid overdose, and hospital readmission
  • Perform data cleaning, wrangling, and feature engineering on the healthcare dataset
  • Complete exploratory data analysis and data visualization on the dataset to determine which variables are closely related to heavy opioid use
    • Health-related, demographic, social, and outcome variables
    • Look for variables related to opioid use disorder or opioid overdose
    • Look for variables related to 30-day and 90-day hospital readmission (all-cause and opioid-related)
  • Implement machine learning models to predict risk of developing OUD, overdose, and hospital readmission at different time intervals (30-day and 90-day, all-cause and opioid-related)
    • Possible models: random forest, tree-based models, recurrent neural network
  • Evaluate and compare the machine learning models using quantitative measures: accuracy, precision, recall, area under the curve (AUC)
  • Make conclusions about how providers can utilize the data and results to provide overdose prevention intervention in an ED setting


Week Description
Week 1
  • Set milestones and goals for project
  • Complete Data Science boot camp
  • Start reading related work provided by mentor
Week 2
  • Read related work provided by mentor
  • Read additional related work (search ACM, IEEE, Google Scholar)
  • Do additional research and practice with Python ML libraries
  • Perform exploratory data analysis on hospital dataset (if available)
Week 3
  • Perform exploratory data analysis and preprocessing on hospital dataset
  • Do more research on which ML models have been used to evaluate similar healthcare-related issues
Week 4
  • Start implementation of ML models
Week 5
  • Prepare and deliver mini presentation
  • Implementation of ML models
Week 6
  • Rerun ML models, make improvements based on results
Week 7
  • Begin writing final report
  • Begin preparing poster
  • Evaluate and begin comparing ML models
Week 8
  • Compare ML models
  • Derive insights from the ML models, make conclusions about what next steps should be for PCP and healthcare staff
Week 9
  • Tidy up ML models and data visualizations for poster
  • Consider future work
  • Continue writing final paper
Week 10
  • Prepare and give oral presentation
  • Present poster at poster session
  • Finish and submit final paper