Tackling America’s opioid epidemic with data, Deloitte and DataStax

The opioid epidemic in the United States of America is devastating the lives of thousands and thousands of citizens every year. In the late 1990s big pharma assured healthcare providers that opioids would not become addictive, leading to them being prescribed on mass for years on end.

We know now that this was not true. And the stats around the epidemic make for a depressing read. For example, in 2017 more than 47,000 Americans died as a result of an opioid overdose, including prescription opioids, heroin, and illicitly manufactured fentanyl, a powerful synthetic opioid. That same year, an estimated 1.7 million people in the United States suffered from substance use disorders related to prescription opioid pain relievers, and 652,000 suffered from a heroin use disorder.

According to the National Institute on Drug Abuse, roughly 21 to 29 percent of patients prescribed opioids for chronic pain misuse them; between 8 and 12 percent develop an opioid use disorder; an estimated 4 to 6 percent who misuse prescription opioids transition to heroin; about 80 percent of people who use heroin first misused prescription opioids; and opioid overdoses increased 30 percent from July 2016 through September 2017 in 52 areas in 45 states.

In other words, people are dying and the situation is only getting worse. Deloitte and DataStax are looking to help solve this national healthcare energency by providing healthcare professionals with an analytical tool - called Opioid360 - that collects de-identified data from a variety of datasets and helps to build profiles of people that may be at higher risk of developing an opioid addiction, that have barriers to treatment, and also offer suggestions on what the best interventions may be.

Speaking at DataStax’s user event in Washington DC this week, Meera Kanhouwa, managing director at Deloitte Consulting & MissionGraph, who is also a clinician with over 20 years experience, said:

Part of the reason I'm here is because I really believe that opiates touch every single one of us. And I bet if you thought about it, one or two degrees of separation from every single person in this room, has an opiate story.

The challenge

Kanhouwa explained that the problem for a physician is that if two people walk into her emergency department, both the same age, have the same back pain - it’s very hard to know who is going to end up being an opioid addict and who has “resiliency”. She added that the healthcare industry needs a data driven approach that can actually create interventions and care for that patient. Because the interventions work. Kanhouwa said:

We need to really understand why it is that people aren't getting the treatment that they need? There are medication treatment protocols that really significantly help and they work. But only two out of 10 people that would benefit from medical treatment, to help their addiction, are actually getting it. That means 80% of people who would benefit from the treatment are not getting the treatment. So it'd be really great to understand what the barriers to that treatment are.

How Opioid360 works is by using DataStax’s Graph technology to analyse the relationships between a number of datasets that traditionally have remained disparate - or would have required someone manually going through and making the connections themselves (which clearly is impossible at scale).

For example, by collecting data from not only government agencies, such as law enforcement, or human services, but also data that government agencies have collected from Google Analytics when running campaigns, or demographic data on people living in certain areas (e.g. homeowners, or financial information), profiles can be built on individuals and their likelihood of becoming an opioid addict.

It’s worth reiterating that DataStax and Deloitte are only using de-identified data to create these profiles - so it is anonymised. The below picture gives a clearer idea of how the data is used to build a profile:

Sean Conlin, Principal, Deloitte Consulting & MissionGraph, was also speaking at the event and he explained:

And so you now have, with de-identification, the ability to look at a very powerful timeline view of a de-identified individual. You can roll that up and look across hundreds of thousands and millions of individuals and look at that over time, and roll that up into observed patterns that are positive in terms of disruption of people ending up in a bad state, or conversely, patterns that lead to very serious outcome. Being able to recognize those patterns, and being able to bring those forward to inform what are the best interventions for groups of people is what we're talking about here.

A potential solution

Kanhouwa and Conlin explained that this meant that if two people came in with similar problems (e.g. back pain), using the Graph technology, Opioid360 could make it easier to identify which of them may be at risk of addiction. If, for example, one was a high earner, owned their own home, their demographic information suggested a healthy lifestyle, and they had good access to transport - they are at lower risk. However, a patient would be identified at much higher risk if they have to travel far to their healthcare provider, don’t have access to transport, live in an area that is typically made up of people on a lower income, and are making use of state-benefits. Opioid360 can help to not only identify that person as ‘at risk’ but then can also suggest what intervention would be best suited for that individual, based on their barriers to treatment (e.g. not having access to transport).

Kanhouwa explained:

Look at this akin to Amazon. When you buy one children’s book, it will then tell you that the people who bought that book also bought these other books. You’re getting recommendations and it is 100% de-identified. That’s big data in the cloud, changing behavior. There is no reason you can’t use that for healthcare. So it's the same idea, right? We know not only the social determinants, the medical data, claims data, or Medicare claims data, plus local geographic search and social media data.

Why couldn't you, and why shouldn't we, as a matter of protocol, ingest that data to improve our knowledge and have information nudges about the patients that we are taking care of on a daily basis?