Health care treatments that are fine-tuned to individuals and their unique physicality—like using one’s own immune system to combat cancer—are changing the medical field for the better.
Precision Psychiatry in the Palm of Your Hand
Personalized medicine made possible with the power of data science
Published: April 19, 2022
Precision psychiatry, however, has not made as much headway. Konrad Kording, Nathan Mossell University Professor at Penn, combines brain science and data science to identify new tools that can help monitor and individualize care in the field of mental health.
One such tool is at our fingertips: cell phones. In particular, Kording and collaborators found that the language sentiments we express in text messages can indicate the severity of depression.
Potential for precision
These findings, a part of the new field of digital phenotyping, present opportunities for precision medicine in psychiatry.
Smartphone sensors, for instance, can track user movements. Location variance and activity levels indicate behaviors associated with low or high mood, and can therefore point to mental health patients’ risk for worsening or improving symptoms of depression.
“There is a surprisingly high comorbidity of loneliness and depression,” explains Kording. “We see depressed people getting out of the house less, as measured by cell phone sensor data.”
But together with colleague Lyle Ungar, Professor of Computer and Information Science, Kording’s recent study takes things one step further.
Drawing on the resources and collaborations available to him as a PIK Professor with dual appointments in the Perelman School of Medicine and the School of Engineering and Applied Science, Kording’s recent study evaluated text message language to measure cognitive and social-emotional data.
According to Ungar, “depressed people talk more about themselves and how badly they are feeling, both physically and emotionally. They express more sadness, anger, and anxiety, question the world more (‘why…?’), and complain more about poor sleep. We see depressed people not only using words like ‘alone’ more, but also sending and receiving fewer texts and phone calls.”
Language sentiments can be captured and tracked in text messages, Kording and Ungar found. What’s more, a combined analysis of language and movement data can provide an even better read on patient wellbeing. When applied together, these tools can be used to make predictions for depression severity.
Wrangling big data
These new insights rooted in brain science are made possible by Kording’s dual expertise in data science.
The PIK Professor conducted his analysis using machine learning, or automated computer models that find patterns and draw correlations across massive amounts of data. They found preliminary indicators that participants’ texting language mirrored their self-reported depressive state.
“Modern research in this area is basically only possible because of machine learning,” says Kording. “In text messages, there are lots of pieces to the puzzle.
“No human could read all of the texts to produce an accurate image – that’s what machine learning does for us. Arguably, much of modern research requires machine learning, from microscopes, via discovery of patterns, to precision medicine. In the age of big data, machine learning is needed to enable the combination of pieces of information.”
The models comb the data and draw correlations, identify trends, and provide insights that psychiatrists could potentially use to inform their evaluations and better diagnose mental health patients.
Improving psychiatry for good
Once explored further, this initial work could bring much-needed improvements to practice.
For example, in his 2017 Journal of the American Medical Association opinion piece, Thomas R. Insel, MD, suggests that the field of psychiatry has moved away from its former strength evaluating behavior. Short appointments focus on medication management rather than a full assessment of overall biological, behavioral, and social factors.
“The study with Ungar showed how text messages contain surprisingly high-quality information,” Kording explains. “It is quite possible that text messages allow for truly actionable insights.”
In many communities, texting is relatively ubiquitous, and this approach could be easily adopted. However, information about one’s cognitive and affective state is also incredibly personal. Kording is keenly aware, therefore, that public policy must ensure anonymity if this tool is brought to practice.
Double up on expertise
The greatest value of Penn’s PIK Professors is found in the connections they forge across disciplinary boundaries. Like Kording, PIK Professors build knowledge that might not have been possible without bringing two disparate fields together.
The resources and collaborations made available to Kording through the PIK Professorship, and through dual appointments in medicine and engineering, have uncovered a potential new tool to deliver more individualized care.
“Precision psychiatry could allow better treatments for patients. But the proof is in the pudding,” Kording says. “Why has precision psychiatry, despite a long history, not made much impact on health care? We should ask these kinds of questions quantitatively and obtain some clarity about the directions that are most promising.”