As the opioid epidemic turns millions of lives upside down, Dr. Hawre Jalal believes the power of mathematics will help point to solutions — and the state and health insurers agree.
Jalal stands at a growing and crucial intersection between public health and public policy. He’s a medical doctor and mathematician, a specialist in mathematical simulation of disease, and an assistant professor at the University of Pittsburgh’s Graduate School of Public Health. He and Pitt colleagues have been using machine learning to get a clearer picture of the nation’s drug epidemic and its spread by developing models that can test the effectiveness of policy approaches before they are implemented to find out which work best.
“The epidemic is a series of sub-epidemics of different tomographies and different geographical locations,” said Donald Burke, dean of the Pitt Graduate School of Public Health. “So we have to be smart about how we intervene to make sure that we target effectively.”
That’s why Pitt is looking to its earlier work in using algorithms to predict the spread of infectious diseases like influenza or measles to apply to the opioid epidemic.
“All of these diseases have commonality with the opioid epidemic with the way it progresses with an individual and the way it spreads among individuals,” Jalal said.
While there are hundreds of models that can predict the course of infectious diseases, there are only a handful that can do what Pitt’s does. The algorithm brings in a wide stream of data sources, and then the computer models provide a clearer picture of the epidemic’s past, present and future. The model then allows researchers to simulate a policy intervention and whether it will ultimately drive down overdoses and overdose deaths. That could be, for instance, the effect of marijuana legalization, prescription drug monitoring programs or the availability of naloxone, an opioid overdose reversal medication.
“What we are finding is the epidemic is highly dynamic,” Jalal said. That makes quick, efficient modeling of policy interventions even more important because it won’t necessarily work in every locale.
Pitt’s mathematical modeling was tapped earlier this year by the Pennsylvania Department of Health to supply the state’s opioid-abuse information system. Secretary of Health and Physician General Dr. Rachel Levine said the data will be invaluable in providing feedback on local and state efforts.
“We want to make that data even more robust, and we want to use that data to inform policy and strategy as well as to be more nimble in our responses to issues that come up,” Levine said.
‘A very useful … model’
Pennsylvania and the rest of Appalachia are ground zero in the number of overdose deaths by opioids, with 4,627 drug overdose deaths in 2016, more than twice the national average rate. Heroin, methamphetamine and cocaine abuse deaths are more numerous elsewhere.
Beyond the emotional toll, the opioid epidemic has a devastating financial cost. More than $1 trillion was spent on it nationwide between 2001 and 2017, with another $500 billion in costs anticipated between 2018 and 2020, according to estimates from health care analytics firm Altarum.
That’s why the region’s biggest health care providers, including Highmark Health, UPMC and Gateway Health Plan, are using big data to change the practice of doctors oversubscribing opioids, transform the lives of members who are addicted to opioids and shift the culture that uses opioids instead of other pain management efforts.
Highmark estimates 1 in 250 of its commercial members are addicted to opioids, up 25 percent in just two years. Opioid addiction cost Highmark $93 million in its three-state coverage area in 2017.
Before Highmark went public in February with its “war on opioids,” the insurer had been quietly, behind the scenes, fighting on multiple fronts. Its efforts yielded 4,100 fewer prescriptions and a 181,623 reduction in opioid pills, tablets and other applications. Highmark Health has donated $600,000 and Highmark Foundation another $120,000 to local efforts against opioid abuse. And in West Virginia, Highmark has been employing a Nashville company’s expertise.
That company is axialHealthcare, a firm that marries analytics, clinical knowledge and evidence-based practice into a platform that can use the power of machine learning to find patterns and associations amid disparate streams of data. AxialHealthcare was founded in 2012 by a team of doctors, scientists, pharmacists and technicians with two goals in mind: Promoting better, evidence-based pathways for pain management and using machine learning and big data to tackle opioid-use disorder. Highmark, an early investor in axialHealthcare, was intrigued by the company’s results.
“One of their most valuable capabilities is their analytics,” said Dr. Charles DeShazer, Highmark’s senior vice president and chief medical officer. “They’ve employed data science and experts in this field to develop this very useful predictive model.”
After a successful trial program in West Virginia, Highmark and axialHealthcare have begun to roll out the program in Pennsylvania and Delaware. DeShazer hopes to see similar results: Through axialHealthcare, more than 250 providers received pain-management consultation, and Highmark saw a 28 percent drop in the amount of members who got opioids from multiple prescribers.
“(Highmark) was one of the first health plans in the country that saw what we saw, that this travesty of opioid-related overdoses had a very specific cause downstream, which is the initial presentation of a patient in pain,” said John J. Donahue, chairman and CEO of axialHealthcare.
The company’s strength is not only in its brainpower and machine learning, but also in its data. Donahue said axialHealthcare has the largest repository of claims data of patients in pain and on pain medication.
Its machine-learning platform can identify doctors who are at the highest risk of overprescribing opioids, and provide consultancy services to guide them to nonopioid pain management therapies, said Dr. Stacey Grant, axialHealthcare’s vice president of clinical services. It also works, along with the insurer and provider, with patients who are using opioids to potentially switch to other pain management options.
But axialHealthcare’s value is much more than just the ability to discover patterns in prescribing or getting prescriptions. Axial combines volumes of data from participating insurers, including Highmark, and state and federal data and its research into the opioid epidemic to take the next step: Finding out who might next be at risk and getting them the help they need. It could be someone who is taking a high morphine equivalent dose as part of treatment, or are on a type of opioid or may have a mental health issue. The quicker that person gets help, the less likely they are to develop opioid-use disorder.
“We’re getting increasingly good at identifying when we see patients who are at high risk of opioid-use disorder who haven’t gotten help yet,” Donahue said.
This has an immediate impact, DeShazer said.
“We’re able to identify and reach out to patients and engage them with our case managers,” he said. “We can apply a comprehensive approach to addressing the need.”
Another Pittsburgh-based health insurer, Gateway Health, also is working with axialHealthcare — as well as in-house — to help identify people who are at risk. It phased in prior authorization for opioids during 2017, which has helped drive down the number of Gateway members with opioid prescription fills per month by 26 percent. And it has removed barriers to medication-assisted treatment.
Across town, UPMC Health Plan also has been taking both big and small steps to tackle the opioid crisis.
“The biggest challenge for us as well as other insurers is that we don’t see most of the people who are at risk of opioid overdose,” said Dr. James Schuster, vice president of behavioral integration at UPMC Insurance Division. There’s a significant number of opioid overdoses who have never been in treatment, never been prescribed opioids or prescribed opioids not covered by insurance. Like Highmark, UPMC Health Plan is using algorithms to predict people at risk for opioid abuse using hospital admissions data, reaching out to patients who may have gone to an ER with symptoms of an overdose or who did overdose, and connecting them with followup services.
“We’re using that information to try to identify these individuals and follow up pretty close to real time,” Schuster said.
For the past three years, UPMC Health Plan also has been working with its network by using a benchmarking system, determining among the providers the baseline for opioid prescriptions and who either prescribes a higher percentage of opioids or at a higher dose, and why.
“We’re trying to identify providers who are prescribing in ways that are significantly different than their peers,” Schuster said.
UPMC shares that information with its providers, starting conversations about prescription best practices and offering additional resources and pain management tools, and, if needed, referring people for addiction treatment. These and other efforts are working. Schuster said the rate of opioid prescription is down significantly.
And there’s plenty more room for big data to help more.
“I think machine learning can identify clearly unrelated factors that we may not have thought of ourselves,” Schuster said.