Update 1

[swei1999]

1. Do you plan for your final project to be an extension of the mid-semester project?

YES

• We plan to find & use additional data that will include the overall prescriptions of opioids (by state, year, and drug) and naloxone (basically an opioid overdose epipen). Our goal here is extending from our previous project because now we are trying to identify issues with prescriptions to identify why all these opioid deaths are happening.

• Potentially from this website: https://data.medicaid.gov/State-Drug-Utilization/State-Drug-Utilization-Data-2020/va5y-jhsv)

• We will incorporate spatial data to create a map of MME (morphine milligram equivalent) per Capita prescribed per state.

• We decided to look at MME because it is the industry standard for comparing opioids by state. Also, this allows us to not worry about the drug type variable, which overcomes one obstacle we had for the last project of not being able to find national data on deaths for specific opioid types.

• We will also attempt to incorporate unsupervised learning in some fashion.

2. Describe what you hope to deliver as a final product.

Obviously, these are initial ideas, but we are thinking to include:

• A link to our original Shiny application (with potential touch-ups) - this would serve to give the user a sense of opioid deaths and some still shots from the original Shiny app that will help tell the story of the opioid epidemic.

• An interactive map that shows which states have prescribed the most / lowest MME per Capita - we will calculate MME per Capita based on prescriptions by state, the relative strength of the product (anhydrous conversion factor), and the population by state

• We are currently thinking to look at specific cases to observe prescription issues by including static visuals of this map and then including an additional Shiny app link to give the user a way to interact and look at the map in more depth (this is subject to change).

• Use unsupervised learning to try to predict future values for MME per Capita (if possible)

• Text throughout providing an overview of the topic, what we are trying to answer (why inappropriate prescribing), explaining our processes and thinking, analysis of these three components, and concluding remarks.

3. Outline a schedule for your group’s progress that will take you from now (ideas phase) to final blog post and presentation at the end of the semester.

Week 1 (next week) 11/1-11/7

• Complete search for data (on our own before Tuesday)

• Reconvene Tuesday for 45 minutes to make sure data is good to go

• Wrangle data on Thursday for 45 minutes (together)

Week 2 11/8-11/14

• Start putting together the blog post throughout the week - probably will assign one component to each person

• Touch up original Shiny app

• Make sure interactive map is working (by Thursday)

• Will have to discuss aesthetics of the overall blog post

• Implement the unsupervised learning model

• Research qualitatively (and put into writing) how / why inappropriate prescribing

Week 3 11/15-11/20

• Make sure everything all parts of blog work and narrative / logic flows nicely

• Finalize blog post & presentation (by Tuesday)

[katcorr]

@swei1999 @cmurphy28 @tamersullivan

Great ideas, team! I do think it makes sense to expand on your shiny app topic further for this blog project. Particularly if you're getting state-level population data - and if your blog post links to your shiny app - I would recommend updating your shiny app visualizations in response to the feedback given for that project.

• to compute per capita, need population for each state in relevant year(s). have you identified a data source for getting that information? Or is that included in one of the data sources you already have or mention here?
• "We will also attempt to incorporate unsupervised learning in some fashion." hopefully after class on Tuesday you can identify a more specific idea around this (or, you don't have to include it, since you will have the spatial visualization)
• "Use unsupervised learning to try to predict future values for MME per Capita (if possible)" -- don't think unsupervised learning is appropriate for that but you can decide after Tuesday's class
• ". . . additional Shiny app link to give the user a way to interact and look at the map in more depth (this is subject to change)." I don't think you'd need an additional Shiny app unless you wanted to allow the user to change something about what the map was displaying. Rather, you could incorporate a Leaflet plot in your blog post directly that would allow the user to click on a state and get more information and zoom in, out and around in the map. is that what you're thinking would be useful interactivity? or something else?
• seems like you have lots of ideas and potential directions; I would recommend finalizing your decisions around what your main focus/deliverables will be by the next update
• great detail in your schedule!

Update 1: 10/10

[tamersullivan]

Update 2

https://www.cdc.gov/drugoverdose/maps/rxrate-maps.html

All the data that we will use comes from the CDC link above (exporting .csv files). Unfortunately, we were unable to track down any data that contained by-drug prescription data (We could purchase by drug prescription data from IQVIA, but that would cost $$$ and we are concerned that our computers would not be able to handle all of the data). The data that we will use contains overall opioid prescription data by state. We will use this data to create a choropleth of overall opioid prescription density and update our previous shiny app to include a choropleth of overall opioid overdose deaths. We will cluster the states (will use an elbow plot to determine the optimal number of clusters) and classify them based on opioid overdose death rate and prescription rate.

Take home points: we are on track, but our plan is slightly modified (no by-drug prescription/overdose data available)

[katcorr]

Glad to hear you're on track, with slight modifications to your plan.

Update 2: 5/5

[swei1999]

Update 3

We exported all the csv files and wrangled the dataset. We decided on doing a Shiny app with spatial data and unsupervised learning and including still images and text throughout our blog post to tell our story. Tamer will be doing the prescription map, Chris will do the overdose map, and I will be doing the unsupervised learning portion of it. We assigned each member of the group a certain section of the blog to complete, and we are planning to meet sometime later this week to put everyone’s pieces together and write up the report/get ready for our presentation.

[katcorr]

Sounds good!

Update 3: 5/5