Linking Data Sets Critical for Unlocking Big Data’s Potential for Improving Addiction Treatment

The potential of big data to improve addiction treatment is high. Thus far, however, it has not delivered. Breaking down silos that are separating data sets has been particularly challenging.

Recently at the Rx and Illicit Drug Summit, Scott Weiner, MD, MPH, an emergency physician with Brigham and Women’s Hospital, Michelle Hendricks, PhD, a senior research analyst for Comagine Health, and Sanae El Ibrahimi, PhD, MPH, a biostatistician for Comagine, presented a session on the use of big data to gain insights on the effects of opioid prescribing.

The trio spoke with Addiction Professional on site at the meeting in Atlanta, Georgia, on the topics of why big data has fallen short of its potential to improve addiction treatment thus far, how public data sets can be linked, and findings from their research evaluating the harms related opioid prescribing at the individual, household, and community level.

Editor’s note: This interview has been edited for length and clarity.

Addiction Professional: Has big data delivered on its potential to improve addiction treatment? Why or why not?

Scott Weiner: I started off our session with just a slide about TED Talks. They talk about big data and the promises that big data will deliver. I have to say, in a lot of healthcare, it hasn't come to fruition. I think the key reason for that is that you can bring together all sorts of data sets, which are completely disparate and use them as one data source, but right now, everything is separate, whether it be prescription drug monitoring programs, death data, emergency medical service (EMS) transports, hospital data. Sometimes they're housed in the same department of public health, but they're often siloed even there. There are different data use agreements for each of those data sources, and it really prevents us from doing what big data does, which is merging all the data together.

Michelle Hendricks: Some of the bureaucratic requirements and the barriers that have been put up, I think, have made it really hard to see if big data can meet its promise. I don't know if we had a good opportunity yet because of all the challenges around that data.

Sanae El Ibrahimi: And then you have challenges in terms of, is there all the information that you need to answer your questions? What type of data is missing? We had to input race and ethnicity for 50% of all our cohorts, for example. Artificial intelligence (AI) and big data go together. Hopefully, in the future, we'll be able to use things like that to help us clean up the data.

AP: Can you talk a little bit about how public health data sets can be linked to address the opioid crisis?

SEI: I think a very interesting topic that we addressed with this dataset that, looking at individual level factors, we did find quite a bit of that information in the literature, but there's not much about household implications and community level implications. An interesting part of the study that we did, we looked at if there are leftover pills in the household, how does that affect someone's overdose risk? How does the community you live in also contribute? We looked at social vulnerability index—what type of socioeconomic status in the community is also related to someone's overdose risk? I think the way we analyzed the data made it interesting for public health purposes, and we came out with some interesting recommendations in the data analysis, as well.

AP: What can you tell us about your research evaluating the harms related to opioid prescribing at the individual, household, and community level? And related to that, how did you source your data? What have been your findings? What are your next steps?

SW: The project was divided up into the 3 phases: your individual risk factors, your household risk factors, and then your community risk factors. We found results from all 3 of those. For the individual risk factors, one of the novel findings we found was that elderly patients—those who are 75 and older—who are receiving a first prescription, they were the most at risk compared to the other age groups. And that's different than what you usually hear, which is that younger patients are more at risk of overdose. But if you take an opioid-naive individual, they're particularly at risk. There are some other factors that greatly increased risk, such as the dually enrolled patients with both Medicare and Medicaid, which we know is a surrogate for social determinants of health sometimes as well. And so, we found that at the individual level.

And there were other factors that were interesting, too. We found that prescriptions for tramadol or for oxycodone, if they were the first prescription versus codeine, were associated with a much greater risk. The household level was also pretty eye-opening to show the risks of an overdose that would happen compared with a household that didn't have any prescriptions for opioids. As you can imagine, if someone gets a prescription themselves, it increases the risk. If someone in the household gets the prescription, it also increases the risk. And then, of course, the highest risk is if multiple people in the household have prescriptions. The more opioids that are around, the more chance of overdose.

AP: That makes sense.

SW: They’re simple questions. They make sense. But without doing this data merging, they’re tough to answer.

MH: At the community level, we found that certain community vulnerabilities are associated with risk of overdose.

SEI: Specifically, if patients lived in communities with, basically, housing insecurity. You have those who are experiencing homelessness or people who live in communities with high levels of multilevel housing. A lack of transportation was also associated with a higher risk of overdose. And again, this is in addition to the individual's own risk because we were adjusted for all this information.

Another interesting thing that we found is, still at the level of the community, if you are living in a single family home or in communities where there's more children, elderly, or more people who have low efficiency of English—which is an indicator of minorities—that was predictive. We just analyzed this. We haven't delved into looking into the literature to understand what these findings mean, but there were some interesting things that we found. Having more buprenorphine fills, for opioid use disorder specifically, was also predictive.

AP: Was there anything else anybody wanted to add that we haven't touched on yet?

SW: The work that we did was very static. We took data that was already existing, and then we did a lot of work to bring it all together, and we found these very interesting conclusions. But hopefully, the work going forward will be something that's more real time. The data providers, which are usually departments of public health, can work to break down their silos of data, and they can figure out a way to merge them with people with increasing accuracy of data around race and ethnicity. This would allow us to learn from it in a way that would help deliver the promise of big data.

Reference

Weiner S, Hendricks M, El Ibrahimi S. Using big data to gain insights on the effects of opioid prescribing. Presented at Rx and Illicit Drug Summit; April 10-13, 2023; Atlanta, Georgia.