Opioid abuse takes a huge toll on human life and the problem is getting worse, Marr states. Just last year, overdose from misuse of prescription or illegal opioids overtook road accidents as the most common cause of accidental death in the US. President Obama (former president after January 20) identified tackling the problem as a prioryt and allocated $1.1 bln for developing and providing solutions In Canada, the situation has been called a ‘national health crisis’.
Highlight risk factors
The big data project that data scientists at BCBST are working on with Fuzzy Logix, used years’ worth of pharmacy data, along with claims data, both from BCBS’s customers and others, through a third-party sharing agreement. Analysts quickly found they were able to highlight risk factors that could indicate whether a person could be in danger of developing an opioid abuse problem. This will hopefully allow doctors and other specialists to step in and offer preventative care before the problem gets out of hand.
By examining the data, analysts were able to determine that certain traceable behaviors – such as frequent use of different prescribers and dispensaries, in combination with each other, were predictive of a chance of later presenting with an abuse or misuse issue. BCBST data scientist Brandon Cosley says this is often “a good indication that someone is trying to feed a problem – and that’s what helps us to create that target – those are opioid abusers, these are non-abusers who take opioids.”
After running analysis on the dataset using random forest models they found that predictions can be made with “something like 85% accuracy.” Cosley adds: “It’s not like one thing – ‘he went to the doctor too much’ – is predictive … it’s like ‘well you hit a threshold of going to the doctor and you have certain types of conditions and you go to more than one doctor and live in a certain zip code…’ Those things add up.”
In total 742 predictor variables are fed into the Fuzzy Logix system, running on Teradata. Data and algorithmic models used in the project will also be used to inform further analytic work carried out by BCBST, such as an initiative aimed at increasing patients’ medication compliance rates.
Preventative work could include interventions such as amnesty boxes for excess medication being appropriately placed, and educational campaigns, as well as the chance for direct intervention by a patient’s healthcare provider. Of course this can be a touchy subject with patients,Marr writes. It isn’t always easy to tell someone “the computer say’s you’re going to become a drug addict.”
So after potential addicts have been identified, a carefully constructed outreach campaign will need to be put in place. “It’s about getting them the information they need that could potentially change their decision making,” Cosley believes. In addition, prescribers may feel that insights from the system mean they are obliged to moderate or re-think their delivery of service to the individual, in order to comply with their duty of care.
At the moment only structured data is used but there are plans to increase the scope of the analytics by bringing in data from customer services, such as telephone calls. Cosley states they are now in the process of bringing that data into this environment, structuring it, pulling it into their database environment and starting to use some of that for our models.
“And that’s just scratching the surface of some of the analytics that we’re starting to roll out. The theory is that there will be information in what people have a tendency to call about, or engage our customer service representatives, that will be predictive of their behavior.”
Prevention more cost effective
Opioid abusers are 59% more likely to be high-cost claimants, with healthcare costs of greater than $50,000 per year. By rolling out predictive, preventative treatment nationwide, in theory insurance costs should come down for everyone, since prevention is far more cost efficient, both in terms of finance and on the person’s health, than attempting to fix it reactively.
But it does raise ethical concerns – telling someone they are in danger of becoming an addict is something that must be done with care and respect. So, Marr writes, it’s fair to say that thanks to the ongoing advancements being made in the fields of Big Data, advanced analytics and predictive modelling, healthcare professionals and those in related industries such as insurance providers are now facing decisions which they always knew they may have to make one day – but probably considered still to be some distance in the future.
Such questions will include whether it’s ethical to price someone’s healthcare based on problems that a computer says they are likely to develop in the future. In the short term, if initiatives such as this can be shown to save lives while cutting costs, it is likely to be something we will see continued interest and investment in.