She walked out of jail after serving a six-month sentence and reached into her pocket.
She had 73 cents to her name.
She tossed a coin into the fountain outside the Ventura County Courthouse and prayed.
It was at this precise moment, on Aug. 8, 2014, that Morgan Thorpe realized the tremendous cost of her opioid addiction.
A clean-living Ohio native who moved with her family to Southern California when she was 16, Thorpe never imagined the depths she would sink to after enjoying a marriage, home and well-paying career as a sales manager at Warner Bros. Entertainment.
She lost it all after becoming a heroin addict, shooting up $3,000 worth of smack a month. At one point, Thorpe had racked up so many drug offenses that she ended up on the Most Wanted list of the Ventura County Sheriff’s Department.
The seeds of Thorpe’s disease were sown when her doctor prescribed her two powerful opioids, oxycodone and fentanyl, after she underwent spinal fusion in 2009 for lifelong back issues. Two weeks after her back surgery, she was hooked.
“It was hell,” Thorpe said of her five-year descent into opioid abuse. “The addiction attacks you. It got to the point where my whole day was spent getting drugs for my next high.”
At USC Viterbi, a handful of engineering and computer science researchers are working on initiatives they hope will, in the coming years, help prevent opioid addiction and provide better therapy and other services for those who succumb to addiction.
The need has never been greater.
In October, President Trump declared a national public health emergency as the number of opioid overdose deaths reached epidemic proportions. Opioids were involved in 42,249 deaths in 2016, and opioid overdose deaths were five times higher in 2016 than 1999, according to the Centers for Disease Control and Prevention.
Thorpe, 39, is one of the lucky ones. She’s been in recovery since she got out of jail in August 2014.
“People are dying out there, and it’s sad,” said Thorpe, who now works as a human resources manager at a substance abuse treatment center in Orange County. “I see a lot of clients who never used drugs become opioid addicts. These drugs take over your brain. Unfortunately, not everyone recovers. Sobriety is a gift.”
USC Viterbi researchers are employing advanced technology to address an addiction that dates back to ancient civilizations, when the opium poppy was discovered as a medicinal plant to blunt pain, elevate mood and induce sleep.
Their research ranges from using deep learning methods to identify likely candidates for addiction, applying signal processing to improve the performance of therapists who specialize in addiction, and using an algorithm to more effectively group addicts in recovery to increase their chances of staying clean.
“This isn’t just an American problem,” said Shrikanth Narayanan, the Niki and C. L. Max Nikias Chair in Engineering and professor of electrical engineering and computer science. “Addiction is something that is human, so we feel this type of technology will have far-reaching impact.”
Yan Liu, a USC Viterbi associate professor in computer science, recalled having to take her 4-year-old daughter recently to a doctor for abdominal pain. She was surprised at the ease at which the doctor prescribed pain medication — in this case, high-strength Tylenol.
In Liu’s native China, painkillers of any sort are seldom prescribed, she said. Opioids — much stronger medications than the Tylenol her daughter was prescribed — typically are used only for late-stage cancer patients. In fact, Liu’s grandfather was prescribed opioids when he was dying from colon cancer some 30 years ago, she said.
“Doctors in China prescribe antibiotics to try to solve the cause of pain, and [the patients] endure the pain,” said Liu, a native of Changchun in Jilin Province who came to the U.S. in 2001. “When my daughter was hurting, I didn’t even think of getting her Tylenol.”
Liu’s research addresses the common practice of prescribing opioid medications, and is aimed at combating addiction before it begins. In November 2017, she and Zhengping Chi a colleague in the Department of Computer Science at USC, along with two collaborators at the Mayo Clinic in Rochester, Minnesota, presented a groundbreaking study that applied a deep learning model to the electronic medical records of 150,000 Mayo Clinic patients.
Because of the high-powered and complex mathematics involved, Liu’s yearlong preliminary study, the first of its kind, classified patients who previously had been prescribed opioid medication into three groups: short-term users, long-term users and opioid-dependent users.
Such modeling has the potential, she said, to help doctors predict a patient’s likelihood of developing an addiction to opioid-based pain medication before such medication is prescribed. In the study, for example, patients who were prescribed opioids for more than 90 days or who received 10 or more prescriptions were classified as long-term users. Others were classified as short-term users or opioid-dependent users. In the latter case, such high-risk patients would be prescribed alternative medication.
“Health practitioners have to walk a very fine line in terms of giving the right amount of medication to patients so they don’t become addicted,” Liu said. “That’s essentially the medical motivation behind what we’re doing. We’re looking at data to accurately identify and personalize what is the right amount of opioid dosage, or to have doctors rule it out as an option.”
Deep learning models allow computers to analyze huge amounts of data in a brief period of time.
“With machine learning, [opioids] can be prescribed much more precisely and allow for more personalized medical treatment,” Liu said.
Liu and her colleagues have used similar artificial intelligence models in research projects for Children’s Hospital Los Angeles to help predict the mortality rate of critically ill children, as well as to determine how long they should stay on ventilators without compromising the natural development of their lungs.
“Doing anything we can do to provide better care and less pain for patients makes this very worthwhile,” Liu said.
She hopes to feel the same when it comes to helping doctors more accurately prescribe — or not — opioid-based medications based on the deep learning solutions she and her colleagues are working on.
Their recent research paper is the first step in a process that will require clinical trials and approval from the Food and Drug Administration before it is applied as a standard.
“I feel we’re doing something that’s really important,” Liu said.
A Tool for Therapists
There are certain phrases a therapist uses that indicate empathy toward a patient:
It sounds like …
What I’m hearing …
So you feel …
There are certain phrases that have the opposite effect:
Please answer the …
During the past …
First of all …
Narayanan and some colleagues have been conducting research that involves automatic speech recognition and machine learning-based models to help therapists provide more effective care to patients fighting any type of addiction, including opioids.
After making audio recordings of more than 1,300 drug and alcohol counseling therapy sessions, they designed a machine learning algorithm to generate an empathy score for each session, based on what was said by the therapist during the session. The phrases listed above were some of several that factored into the empathy rating.
For the past 70 years, the labor-intensive and error-prone method for evaluating patient-provider interactions in psychotherapy — basically, observing and describing a subject’s behavior — has remained unchanged, Narayanan said. Each year, $100 billion is spent on mental health treatment in the U.S. That translates to roughly one billion counseling sessions, according to Narayanan.
“Right now, we really don’t have any way to measure the quality of these sessions or therapists,” Narayanan said. “We want to see how AI can help solve this problem, and one of the things we’re building on is how to provide better feedback to therapists based on a qualitative measure of how a session went.”
Narayanan and two colleagues in the Signal Analysis and Interpretation Lab (SAIL) at USC Viterbi, in collaboration with researchers at the University of Utah and the University of Washington, recently published a paper titled, in part, “Rate My Therapist.” They hope to use the machine learning algorithm to train aspiring psychotherapists to be as empathetic as possible toward patients, which is one way to make sure the sessions are as effective as possible to achieve the goal of long-term recovery from addiction. As Narayanan’s paper puts it, empathy is important because it demonstrates that the counselor has a deep understanding of the client’s point of view, as opposed to showing no interest in the client’s perspective.
Long-term, Narayanan and his team hope to create software that gives therapists with real-time feedback.
Narayanan and his fellow researchers are in the first year of a second, five-year grant from the
National Institutes of Health.
“We’re trying to quantify things that, up until now, have largely been qualitative,” he said. “It’s a fairly ill-defined domain, and we’re trying to bring technology and apply it broadly across therapy, not just addiction-related therapy.”
In another recent paper, Narayanan and three colleagues at SAIL (Daniel Bone, Theodora Chaspari and James Gibson), along with USC alumnus Chi-Chun Lee, now an assistant professor at National Tsing Hua University in Taiwan, discussed what they foresee as the results of applying signal processing techniques to mental health research and the clinical field.
“An achievable dream of ours is to see engineering technologies integrated within and supporting all aspects of mental health research and care, helping to fill scientific knowledge gaps, connecting dots, and supporting novel interventions,” the researchers wrote. “Signal processing will enable access to truly dynamic, patient-centric care.”
A Great Need
In the mid- to late 19th century, the first national opioid crisis occurred, partly as a result of liberal use of opioid-based treatments for injuries and diseases suffered by Civil War combatants. That first epidemic eventually was contained.
Whether the current crisis can be contained, and when, is up for debate. But research going on at USC Viterbi is addressing the “how” part of the equation, which for Thorpe and other addicts is great news. Addicts, Thorpe said, need all the help they can get fighting their addiction and staying clean.
“I didn’t even know I was addicted to painkillers until I didn’t have them anymore,” she said. “My skin was crawling and I couldn’t stay still. I needed [drugs] just to stay normal, and that’s a real scary feeling.”
Two years after her back surgery in 2009, Thorpe checked into a detox facility to get off fentanyl. By then, her husband had left her and she was spending most of her days isolated inside her home.
At an acute psychiatric facility in Ventura County, she met someone who introduced her to heroin.
“When you’re in that state of mind making the wrong decisions, it become easier to make the wrong decisions,” Thorpe said.
She used heroin for three years, losing everything, until she was locked up.
“In the end,” Thorpe said, “I was a garbage disposal. I was using everything.”
After tossing that coin into the pond outside the Ventura County Courthouse, Thorpe began her long journey to sobriety. She spent six months in a sober-living home, where she met a fellow addict who helped her get her current job at a recovery home in Orange County.
“I’d say having a strong network is about 25 to 30 percent of my sobriety,” said Thorpe, who is active in a 12-step program and recently participated in an opioid-awareness event.
Better Support Networks
Another research project at USC Viterbi aims to improve an addict’s chances of recovery by surrounding him or her with the right people in group therapy sessions.
Milind Tambe, a professor of computer science and the founding co-director of CAIS, the USC Center for Artificial Intelligence in Society, is leading a collaborative pilot project with Urban Peak, a homeless youth shelter in Denver. Assistant Professor Eric Rice, from the USC Suzanne Dworak-Peck School of Social Work, also is involved in the project.
The team, which also includes Phebe Vayanos, an assistant professor of industrial and systems engineering and computer science and associate director of CAIS, has created an algorithm to carefully construct intervention groups based on the addicts’ social networks.
“The algorithm uses information on the drug-using behaviors of the individuals and their social ties to determine the best way of conducting the intervention, the aim being to minimize the number of drug users,” Vayanos said. “The algorithm does so by reasoning on how the social ties will evolve as a result of the intervention and on how individuals tend to influence the drug-using behavior of their social ties.”
The mathematical model, using a type of AI known as optimization, avoids clustering hardcore addicts together to get a mix of users with various levels of addiction. Such variety in group therapy sessions increases the chances of recovery, Tambe said.
“We want to create the right kind of mix of people, and that’s where AI comes in,” said Tambe, the Helen N. and Emmett H. Jones Professorship in Engineering. “It’s not easy, because there are existing relationships that change over time when you put people in groups. AI has to figure it all out. It becomes a very complicated problem for a human being.”
Robin Baker, manager of program evaluation at Urban Peak, said she was a little wary of AI at first. But after meeting in December 2017 with Tambe and Vayanos, the potential of AI being applied to group-based interventions looks promising.
“We’re living in a time when there is so much social media and internet-based everything, and I think this is a tool that will help to ensure that any interventions that are provided have the best chance of success,” Baker said. “Plus, youth today are so tech-savvy, so this project plays into that, too.”
The pilot project is expected to begin this year.
“I’m really excited at the prospect of collaborating with Urban Peak and Robin to help optimize the way in which they conduct interventions,” Vayanos said. “Having visited Urban Peak, we got to witness firsthand that they care deeply about helping the homeless youth population and being [even a small part] of that endeavor feels extremely rewarding.”
Tambe has high hopes the algorithm will increase the effectiveness of group sessions at Urban Peak. He said he was inspired by the good work he witnessed by staff members to turn around the lives of the homeless youth, many of whom suffer from addiction, including opioid addiction.
“It’s not all gloom and doom,” Tambe said. “That we may be able to contribute to their success is really rewarding.”
The Opioid Epidemic
• Opioids were involved in more than 42,000 deaths in the U.S. in 2016. Forty percent of all opioid overdose deaths involve a prescription opioid.
• 115 Americans die every day from an opioid-related overdose.
• Overdose deaths involving prescription opioids were five times higher in 2016 than 1999.
• Every day, over 1,000 people are treated in emergency departments for misusing prescription opioids.
• A study in 2016 estimated that prescription opioid overdoses, abuse and dependence in the U.S. in 2013 cost $78.5 billion. In November 2017, the White House said the true cost of the opioid drug epidemic in 2015 was $504 billion.
Sources: Centers for Disease Control and Prevention, Council of Economic Advisers
What Are Opioids?
Opioids are a class of drugs used to reduce pain. They can cause severe side effects and are highly addictive.
Common prescription opioids are oxycodone (OxyContin), hydrocodone (Vicodin), morphine and methadone.
Fentanyl is a synthetic opioid pain reliever that is many times more powerful than other opioids and is used by doctors to treat severe pain, typically advanced pain in cancer patients.
Heroin is an illegal opioid. It, as well as illegally made and distributed fentanyl, has been on the rise in several states.