For decades, the “Diagnostic and Statistical Manual of Mental
Disorders” (DSM) has been the bible for psychiatric
diagnosis. But it’s controversial, with many mental health
experts questioning the scientific validity of the data on which
it’s based. Continual revisions have not fixed the problem;
the DSM-5, published in 2013, is the most controversial yet. Now, a
team of researchers, including Rachael Hageman Blair, an assistant professor
of biostatistics in UB’s School of Public Health and Health
Professions, has received an NSF grant to investigate whether big
data can be used to develop a more rigorous approach to classifying
mood disorders—one that, hopefully, everyone can agree
There is no biological marker. The diagnosis is based on a semi-structured interview, which consists of various modules related to different disorders. There are other tools like a “Mood Disorder Diagnostic Questionnaire,” which is based on patient self-reporting.
The National Institute of Mental Health noted
recently that treatment for mood disorders is effective in less
than 25 percent of patients.
The symptoms overlap considerably. And they’re very
elusive, arising from different biological, psychosocial and
We’re taking first steps to cluster patients with respect
to some of the features known to drive mood disorders. It’s
challenging, because there’s uncertainty everywhere.
There’s uncertainty in the data, especially self-reporting
data. It’s also not clear what features contribute to the
disorder, so there’s a variable selection problem. And the
labels of the disorders themselves—such as unipolar and
bipolar—are loaded with uncertainty that comes from doctor
bias and the ambiguity of the DSM. We’re developing new
methods to overcome some of these challenges.
We have a tremendous amount of data that ranges from MRI images
to genetic profiles, clinical measures, information about
childhood, finances, diet and many self-reporting-style questions
Our long-term goal is to develop interactive tools that can be
used by a clinician to help categorize patients, and identify those
who would benefit from certain drugs.
There are lots of things, because existing methods simply “break” under the load. More work needs to be done at a fundamental level to fully realize what big data has to offer.