Decide focus for the project

[ethankoch4]

Let's use this issue as a thread for a chalkboard to decide on a focus for our project.

[ethankoch4]

https://www.coloradohealthinstitute.org/research/death-drugs

Paul shared this document -- re. drug statistics in Colorado

[ethankoch4]

https://www.scientificamerican.com/article/ai-scans-twitter-for-signs-of-opioid-abuse/

article regarding twitter data & opioid abuse

[ethankoch4]

https://www.colorado.gov/pacific/cdhs/data-and-evaluation-systems-resources

could be used to track opioid addiction numbers?

[lilsummer]

just a thought that addition might be more interesting that drug related overdose -- we might hit data sparsity issue for the later one

[jtillis0317]

looking at what has already been posted, here are some of my initial ideas:

1. a data dashboard that presents social media (twitter) data by geographic region so that public health providers can respond to that data and target their resources
2. if drug overdose rates are significantly higher in counties/towns with lower populations, something to target more rural areas (i know that's vague)

[adriellepowers]

Just sharing my thoughts and findings from today's lunch break. I'd like to delve a little deeper after work to help refine our question.

https://nsduhweb.rti.org/respweb/homepage.cfm - An organization that conducts yearly surveys on drug use and mental health. Tried to do a little bit of digging into how they select people to interview but can't find much info yet.

https://www.samhsa.gov/data/sites/default/files/NSDUH-FFR1-2016/NSDUH-FFR1-2016.pdf - NSDUH 2016 report, they start talking about opioid misuse on page 20

https://www.drugabuse.gov/drugs-abuse/opioids/opioid-summaries-by-state/colorado-opioid-summary - maybe we can add HIV and Hep C diagnoses to the list of proxies/indicators of opioid addiction

It would probably be a good idea to familiarize ourselves with the names of different types of opioids so we can distinguish what is being prescribed for pain and what's being prescribed for treatment. And see if anything in our data indicates how heavily these meds are prescribed, the type of pain their prescribed for, if there are different events leading up to and/or following the prescription of different types of opioids, are different demographics drawn to any particular type...? .... oxycodone, hydrocodone, morphine, codeine, methadone, naloxone, buprenorphine...

Anyway, back to work. Cheers.

[lilsummer]

I like the dashboard idea, and the second one would be some factor/correlation analysis regarding drug overdose (what are the correlated factors, what are the early indicator, what are the driver, things like that)

Lets put together a working google doc to finalize our goals and plans?

[paul-rosen]

So having spent some time yesterday doing more reading, I feel like I’ve moved from knowing nothing about this whole topic to knowing practically nothing about it, which I guess is a step. What follows is my current big-picture understanding of the issue. Caveat 1: my understanding may not be correct. Caveat 2: if you’ve spent more than a few minutes talking to Google about this issue you probably understand it at least as well as I do.

In the early days of the opioid epidemic, before the medical community understood the nature and scope of the problem, a lot of patients were given prescriptions for high doses of opoids and were allowed to continue to use them for long periods of time. These practices led to a lot of people becoming addicted, and because it was relatively easy to get more prescribed drugs by going to multiple doctors or multiple pharmacies, those who because addicted could abuse the drugs without having to do overtly illegal things like buy heroin from a guy on a street corner.

Early steps in the response to the epidemic included educating doctors about the risks of addiction to try to change how they prescribed the drugs and tracking prescriptions better to identify people who were gaming the system to get more pills. While it seems like these steps will be effective in decreasing the number of people who become addicted each year, though, they don’t address the problem of the large population that is already addicted.

In general, it seems like there are two ways of going about finding people who are addicted: looking for addicted individuals and looking for communities with high rates of addiction.

Organizations that have detailed information about individual patients are working to identify people at high risk for addiction (e.g. https://www.jpain.org/article/S1526-5900(15)00985-2/pdf and https://www.ajpb.com/journals/ajpb/2014/ajpb_septemberoctober2014/understanding-predictors-of-opioid-abuse-predictive-model-development-and-validation). It seems like most of this work focuses on patients’ medical histories and not on other factors related to addiction, but it also seems like there is a growing belief that behavioral health approaches that utilize other kinds of data might be even more effective in identifying at-risk individuals (e.g. https://healthitanalytics.com/news/using-machine-learning-to-target-behavioral-health-interventions).

If we had access to detailed information about individual patients it seems like we might be able to do something interesting along these lines by looking for other things that correlate with addiction, but we don’t have that kind of data.

In terms of identifying communities with high rates of addiction, we do have data that are probably correlated with addiction rate (e.g. prescriptions filled, overdose deaths, drug arrests, rates of things like HIV infection and hepatitis that are correlated with intravenous drug use). We could also look for correlations with various social and economic factors (e.g. education, employment, church attendance, participation in elections, social media use and tone as determined by some sort of sentiment analysis).

I don’t know if this is the kind of thing we can do in a few weeks with a few people working an hour or two here and there. That said, it looks like Cassie (I think lilsummer is Cassie) has already started merging data sets based on census tract. Since a census tract encompasses a few thousand people, this seems like a reasonable unit to use to try to address the question of addiction rate in a community, so if we decide that’s the question we want to look at maybe we just keep adding data based on census tract and see what correlations pop out?

Since I don’t have any conclusion I want to draw, I guess I’ll stop typing now.

[lilsummer]

Have you guys look at other dataset yet? I guess we can have a vote on deciding what is our final goal

I would say

• Generate a data dashboard to show overdose statistics on the FIPS level
• Ask a deeper question, such as what are the correlators of drug overdose?

Can we have a online meeting next week potentially wednesday, thursday or Friday?