Aug 19, 2020
Philip Resnik, PhD, joins host Lorenzo Norris, MD, to discuss
the use of AI and natural language processing to help clinicians
identify patterns in the behaviors of patients with mental
illness.
Dr. Resnik is
a professor in the department of linguistics at the University of
Maryland, College Park. He also has a joint appointment with the
university’s Institute for Advanced Computer Studies.
Dr. Resnik has disclosed being an adviser for Converseon, a
social media analysis firm; FiscalNote, a government relationship
management platform; and SoloSegment, which specializes in
enterprise website optimization. Some of the work Dr. Resnik
discusses has been supported by an Amazon AWS Machine Learning
Research Award.
Dr.
Norris disclosed having no conflicts of interest.
And don’t miss the “Dr. RK” segment, with Renee Kohanski, MD.
Take-home
points
- Artificial intelligence (AI) refers to the effort to get
computers to develop capabilities that humans would consider
intelligent when people do them. For example, a “smart” thermostat
learns patterns of behaviors and changes the temperature
accordingly.
- Natural language processing (NLP), an AI approach, focuses on
the content of language from the words used and looks for cues
within the content. NLP technology allows computers to do things
more intelligently with human language, and NLP has generated
technologies such as Siri, Alexa, and Google Translate.
- Much of clinical work is focused on language, and clinicians
look for cues within the content. Dr. Resnik is a technologist who
believes that NLP can help facilitate clinical progress, especially
in the face of a shortage of mental health clinicians and the
limited amount of time that clinicians are able to spend with their
patients.
- Research aimed at using machine learning and NLP to analyze
social media and other types of online presence to evaluate for
suicide risk and the presence of mood disorders is underway.
- Dr. Resnik imagines an ecosystem in which computers and humans
balance their efforts, with each “brain” doing what they are best
at; he believes in technology’s ability to save us time so we can
prioritize our efforts.
Summary
- A common example of NLP is automatic dictation and
transcription software embedded in medical records. Dr. Resnik
thinks of technology as an enabler and augmentation strategy.
- Resnik and his wife, Rebecca
Resnik, PsyD, completed a study using NLP to automatically
detect clusters of language in the writing samples of college
students. NLP software evaluated the natural patterns of language
that might correlate with vegetative and somatic symptoms of
depression and social isolation. His team was able to home in on
language themes specific to college students that suggest specific
symptoms of depression.
- Another example of NLP in mental health is using predictive
modeling, taking in data, and then making a prediction about a
pertinent variable to understand mental health outcomes. For
example, Glen Coppersmith, PhD, and associates evaluated social
media posts with NLP software and concluded that analysis of
language in social media posts can accurately identify individuals
at risk of suicide and facilitate earlier interventions.
- Resnik imagines a future in which speech and language samples
are used to give a point-of-care evaluation of a patient’s mood and
suicide risk.
- “Clinical white space” is all the “space” (for example, the
time between clinical encounters) and this is where decompensation
occurs. Resnik suggests that NLP software could be used to fill
this white space by using apps to collect text samples from
patients. Software would analyze the samples and warn of patients
who are at risk of decompensation or suicide.
- Barriers to using this technology include engaging the
technologists and clinicians, and accessing data samples because of
privacy concerns, especially because HIPPA was written before the
emergence of mega data.
References
Coppersmith G et al. Natural Language Processing of Social Media
as Screening for Suicide Risk. Biomed Inform Insights. 2018 Aug 27.
doi:
10.1177/1178222618792860.
Zirikly A et al. CLPsych 2019 Shared
Task: Predicting the Degree of Suicide Risk in Reddit
Posts. In Proceedings of the Sixth Workshop on
Computational Linguistics and Clinical Psychology. 2019 Jun 6.
24-33.
Lynn V et al. CLPsych 2018 Shared
Task: Predicting Current and Future Psychological Health from
Childhood Essays. In Proceedings of the Fifth Workshop on
Computational Linguistics and Clinical Psychology: From Keyboard to
Clinic. 2018. 37-46.
Selanikio J. The big-data revolution in health care.
TEDx talk.
Graham S et al. Artificial Intelligence for Mental Health and
Mental Illnesses: An Overview. Curr Psychiatry Rep. 2019 Nov
7;21(11):116. doi: 10.1007/s11920-019-1094-0.
Show notes by Jacqueline Posada, MD, who is associate producer
of the Psychcast and consultation-liaison psychiatry fellow with
the Inova Fairfax Hospital/George Washington University program in
Falls Church, Va. Dr. Posada has no conflicts of
interest.
For more MDedge Podcasts, go
to mdedge.com/podcasts
Email the show: podcasts@mdedge.com