Difference between revisions of "User:ZFarahany"
From REU@MU
| Line 91: | Line 91: | ||
*Among the factors analyzed race was the strongest influence over the prevalence of Covid | *Among the factors analyzed race was the strongest influence over the prevalence of Covid | ||
*List of specific limitations of EHR data | *List of specific limitations of EHR data | ||
| + | |||
| + | ===The Opioid Crisis: a Comprehensive Overview=== | ||
| + | |||
| + | Useful info | ||
| + | *Comprehensive history of the opioid crisis | ||
| + | *Driving forces of the crisis | ||
| + | *Groups at risk (Middle-aged Women, Pregnant women, Veterans, children in sports) | ||
| + | *Comprehensive list of adverse events | ||
| + | *Discussion of legislature used to combat the epidemic, criminalization | ||
Revision as of 20:20, 14 June 2021
Zach Farahany's profile
About me: Hi, I'm Zach. I'm a Data Science and Computational Mathematics major at Marquette University. My research project is Predicting Risk of Opioid Use Disorder with a focus on the impact from Covid-19.
Contents
- 1 Work Log
- 2 Week 2
- 3 Literature Summaries
- 3.1 Predicting Opioid Overdose Readmission and Opioid Use Disorder with Machine Learning
- 3.2 Predicting Opioid Use Disorder using Random Forest
- 3.3 A clash of epidemics: Impact of the COVID-19 pandemic response on opioid overdose
- 3.4 COVID-19 risk and outcomes in patients with substance use disorders: analyses from electronic health records in the United States
- 3.5 The Opioid Crisis: a Comprehensive Overview
Work Log
Week 1
Tuesday
- Attended REU Orientation and learned beginner python
Wednesday
- Learned data visualization and basic machine learning
Thursday
- Good research practices talk from Dr. Brylow
- Meeting with Dr. Praveen for project expectations
Friday
- Made personal webpage and completed research plan
Sunday
- Read "Predicting Opioid Overdose Readmission and Opioid Use Disorder with Machine Learning"
Week 2
Monday
- Attended data ethics talk from Dr. Brylow
- Read "Predicting Opioid Use Disorder using Random Forest" by Wadekar et. al.
Tuesday
- Meeting w/ Dr. Praveen
Wednesday
- Completed Basic RCR CITI module
Thursday
- Completed Biomedical CITI module
- Read "A clash of epidemics: Impact of the COVID-19 pandemic response on opioid overdose" by Linas et al.
- Read "COVID-19 risk and outcomes in patients with substance use disorders: analyses from electronic health records in the UnitedStates" by Wang et al.
Friday
- Researched ML terms such as association rule, support, confidence, lift, conviction
Some information below may be improperly paraphrased from an article so do not copy
Literature Summaries
Predicting Opioid Overdose Readmission and Opioid Use Disorder with Machine Learning
Objectives
- Use multiple machine learning models and multiple data types to predict the likelihood of hospital readmission following an opioid overdose and diagnosis of opioid use disorder after being prescribed an opioid
Useful Info
- AUC value is a rate of correct prediction
- Hospital info is meant to be anonymized, patient identifiers must be removed
- T40 codes are used as identifiers of various conditions including Covid and OUD
- Various methods of cleaning and compiling hospital records into more useful data frames
- Various viable machine learning models that could be used on my data
- 10 fold cross-validation methods of machine learning
- SMOTE used for class balancing
- Various limitations of hospital data, the data does not include non-registered opioid use or addiction
- "Black box" structure of machine learning models
- Deep learning models such as RNN GRU and LSTM
- Doctor AI used for EHR(Electronic Health Record) data
Questions
- Why is SMOTE necessary?
- What is "gain"?
- How to fix "data noise"?
Predicting Opioid Use Disorder using Random Forest
Objectives
- To use Random Forest on a public dataset to make a predictive model for determining OUD diagnosis
Useful info
- First age of marijuana consumption, mental illness status, and age in that order are the biggest predictors of OUD from this study
- Useful references for the existence of the Opioid crisis
Questions
- What is downsampling?
A clash of epidemics: Impact of the COVID-19 pandemic response on opioid overdose
- Created a simple model (RESPOND - researching effective strategies to prevent opioid death) to show how social distancing could harm people with OUD and worsen the OUD population
- Discussed possible problems pandemic poses for people with OUD such as drug supply shortages, mental health problems from social isolation, relapses from lack of community, people not seeking medical help because of distancing, etc.
- Concluded that the pandemic will have a disproportionate effect on the OUD population because of compounding mental, physical, economic, and social problems
- Rough estimate of how much covid harms the OUD population
COVID-19 risk and outcomes in patients with substance use disorders: analyses from electronic health records in the United States
Useful info
- OUD is the worst substance abuse problem in terms of the additional likelihood of getting covid
- Opioids and Covid both weaken the respiratory system, Opioid overdose deaths are from failures in the respiratory system
- Other SUD(Substance Use Disorders) target cardiovascular, pulmonary, and metabolic all of which are risk factors of Covid
- DSM-5 contains standardized definitions of OUD and other SUD
- OUD often have comorbidities from their drug use. Many of these are risk factors for Covid
- Use of MOUDs do not have a significant effect on the prevalence of Covid
- Among the factors analyzed race was the strongest influence over the prevalence of Covid
- List of specific limitations of EHR data
The Opioid Crisis: a Comprehensive Overview
Useful info
- Comprehensive history of the opioid crisis
- Driving forces of the crisis
- Groups at risk (Middle-aged Women, Pregnant women, Veterans, children in sports)
- Comprehensive list of adverse events
- Discussion of legislature used to combat the epidemic, criminalization
