Algorithm Aims to Improve Mental Health Outcomes
By Kirsten HeuringMedia Inquiries
- Associate Dean for Communications, MCS
During their senior year at Carnegie Mellon University, Cara Yi and Wani Zhang knew they wanted to do research with a tangible impact.
As undergraduate mathematicians at CMU, they partnered with public health experts to identify ways to assign children with behavior problems to programs best suited to serve their needs.
"From their study of operations research, the students had learned a set of tools for optimization," said David Offner, associate teaching professor of mathematical sciences, who advised their work. "We were looking for an interesting applied optimization problem, preferably involving some data from an applied setting."
Offner connected them with Dan Warner, the executive director of Community Data Roundtable (CDR), a nonprofit organization working to increase the effectiveness of the Pennsylvania public mental health system by finding evidence-based solutions.
CDR provided data about two treatment programs for children with mental health issues, Behavioral Health Rehabilitation Services (BHRS) and Family Based Mental Health (FBMH). Both are community-based programs that provide assistance to children at risk of losing home placement, and they rely on the Child and Adolescent Needs and Strengths (CANS) assessment tool to identify treatment needs and measure effectiveness.
"Typically [CANS] asks a whole bunch of questions on a scale of zero to three and creates profiles of individuals based on how all these items for any individual might be scored from zero," said Warner, who earned a doctorate in psychology and led clinical programs prior to founding CDR. "Based on that information, you are able to do the most advanced decision support."
Developed by John Lyons of the University of Kentucky, CANS measures a range of mental health conditions and precursors from ADHD to family difficulties, and it includes measurements of children's strength to determine how resilient they are to their problems. BHRS and FBMH treat different types of symptoms, and they have some differences in their methods.
"There can be many vagaries in the system as to which program a child gets, and there is debate often if one child should be in one program or the other," Warner said. "The most notable part is the fact that we are using math to help with mental health, which is a very new field that hasn't really been touched upon," said Zhang, who is now pursuing a Ph.D. in management and science operations at the London Business School. "What we're doing is using this model to assign children with behavioral problems to one of two treatments. One is family-based, and the other is more independent."
Using historical data, Yi and Zhang built a mathematical model to determine which program served which symptoms best.
"If one treatment did not work well with a patient with a certain set of symptoms, then in the future, a similar patient should be recommended the other treatment," Zhang said.
Yi and Zhang assigned scores to each parameter identified in the CANS assessment depending on which service was more effective to address its situation based on historical data. Once the framework was complete, Yi and Zhang could input an individual's CANS results into an optimization formula that would suggest the most beneficial treatment.
"Data is very helpful to provide suggestions, guidance and insight to the people who are making the decisions," said Yi, who joined the master's of science in information technology - business intelligence and data analytics program at CMU's Heinz College of Information Systems and Public Policy. "I hope to use my skills and knowledge to provide them more precise information that they can use to make decisions."
Through research like this, Carnegie Mellon is committed to community-university partnerships to help solve social impact problems. Offner said that Yi and Zhang collaborated well with their counterparts.
"It took a lot of persistence to see this project through from inception to completion," Offner said. "We're very grateful to Dan and the CDR team for sharing their data with us and also for a lot of guidance and expertise they lent to the project."
Yi said that she hopes the collaboration between mental health researchers and mathematicians can continue to progress.
"It's very interesting to have that conversation with psychologists to get to know what this is usually associated like based on their experience," Yi said.