The THRIVE Project
This project, entitled “Technology & Health Related Information to improVe wEllness” or THRIVE, is a collaboration with clinical researchers and practitioners at Northwell Health in New York. It aims to examine how machine learning and computational techniques may be applied on patient-contributed social media and other online data to eventually inform clinical decision-making and positively influence treatment in the wild. We are focusing on early and first-episode psychosis patients seeking treatment at Northwell for these investigations.
Schizophrenia is associated with significant impairment. While earlier intervention improves outcomes, relapse rates remain high, and limit the established benefit of available services. Although the majority of patients initially achieve clinical remission of psychotic symptoms, up to 80% experience a relapse within the first five years. Early, precise, and continuous monitoring of symptomatic exacerbation is critical to facilitate the initiation of proactive relapse prevention strategies. The rise in social media activity can be an opportunity to inform early relapse identification strategies, especially for youth who are both the highest utilizers and at the greatest risk for the emergence of a psychotic disorder. Accordingly, this project is leveraging patient-generated and patient-contributed Facebook archives to build and evaluate machine learning methods that predict patients’ relapse hospitalizations. A complementary investigation is assessing if the same data can help identify early warning signs of the onset of the illness in patients, with a goal of eventually reducing the duration of untreated psychosis.
SocWeB Collaborators (Current): Sindhu Ernala, Domino Weir, Munmun De Choudhury
Collaborating Institutions: Northwell, Cornell Tech
Funding: NIH; Northwell
The REDUCE Project
This project, entitled “Real-time Ensemble Data for Understanding Suicide Epidemiology” or REDUCE, aims to develop improved public health surveillance algorithms for real-time estimation of rates of suicide nationally. Unfortunately, suicide rates have risen nearly 30% nationally over the past 15 years. Planning for suicide prevention activities in this emerging epidemic is limited by a lack of real-time trend data on the suicide burden. Consequently, public health organizations like the CDC are investigating the accuracy and applicability of novel approaches that rely on using machine learning models to combine data from multiple time series data sources to more accurately estimate the real-time burden of suicide and self-directed violence nationally.
The developed algorithms will leverage multiple, already-existing large-scale social media (Twitter, Reddit, Google Trends, YouTube Search trends, etc) and web data sources as an input, combined with data from clinical and public health sources (Emergency Department syndromic surveillance data, vital statistics data, crisis text and call line data, etc) for robust and scalable prediction and forecasting of nationwide rates of suicide. These outcomes will be evaluated statistically as well as via comparison with actual suicide surveillance data available to CDC.
SocWeB Collaborators (Current): Daejin Choi, Jordan Taylor, Chaitanya Konjeti, Munmun De Choudhury
Collaborating Institutions: CDC
Funding: CDC
The Tesserae Project
In our networked information age, as information work is evolving and as boundaries of work and personal life blur, we need to rethink traditional definitions of how workers perform, adapt to, and engage in work. Information work is the dominant driver of our modern economy and is essential towards the secure protection of our national interests. Characteristics of information work include the ability to be flexible while working in a high stakes, high risk, and often stressful work environment. Although there is rich literature that explores individual measures that relate to job performance, few studies have examined how the various measures interact in different contexts to impact job performance over time. Complicating matters, individual measures may be constrained or amplified by aspects such as context, environment, organizational dynamics, and social interactions (online and offline) affecting generalizability and broad applicability.
Our innovation lies in the joint modeling of the mental states, behaviors, interactions, and context of diverse cohorts of individuals from five major sites in workplace and home contexts across a 12-month timeframe. We do this with personal, social, contextual, and specialized sensors that will unobtrusively collect and model physiological, psychological, behavioral, and physical states of individuals; their offline and online conversations and interactions; computer, phone, and social media activity and workplace routines; and health and well-being at work and at home.
SocWeB Collaborators (Current): Koustuv Saha, Vedant Das Swain, Manikanta Reddy, Hemang Rajvanshy, Tahirah Ahmad, Munmun De Choudhury
Collaborating Institutions: University of Notre Dame, University of California at Irvine, Dartmouth College, University of Washington, University of Colorado Boulder, Ohio State University, University of Texas at Austin
Funding: IARPA
The CampusLife Project
Mental health concerns among young adults attending universities are a significant challenge to the overall well-being of the academy. Among US college students, approximately 50% experience some form of mental disorder annually. In particular, stress among college students has been shown to be associated with cognitive deficits, decreased academic and life satisfaction, poor health behaviors, and even suicide. Despite growing recognition of this challenge, few university students seek help related to psychological stress and its comorbid mental health challenges – only 18% of students with a past-year mental disorder are known to seek treatment. It has been posited that introduction of timely interventions, coping strategies and mitigation programs might decrease the negative effects of mental disorders. This, however, necessitates that there are adequate ways to detect and understand mental health challenges and its correlates accurately, continuously, inexpensively and with little intrusion.
This project examines how student contributed data gathered via active and passive sensing techniques, including that from smartphones, wearable devices, and social media, as well as through large-scale physical and digital campus infrastructures, can be computationally modeled to assess risk factors of a variety of mental illnesses in college students, such as depression, stress, and anxiety.
SocWeB Collaborators (Current): Vedant Das Swain, Koustuv Saha, Jayant Jain, Munmun De Choudhury
Collaborating Institutions: University of Texas at Austin, University of Washington, Stanford University, Penn State University
Technologies to Support the Clinical Workflow in Mental Health Treatment
Digital traces of people, such as what they write on Facebook, can reveal a variety of valuable aspects of mental health of individuals, such as affect or depression.The goal of this project is to design a visualization interface that presents various computational analyses of patients’ social media data, in order to support structured interviews administered by mental health clinicians. We believe that the inferred mental health status attributes based on analyses of social media usage of patients can be useful collateral information which mental health clinicians and patients could utilize in their treatment decision-making processes.
SocWeB Collaborators (Current): Dong Whi Yoo, Sindhu Ernala, Munmun De Choudhury
Collaborating Institutions: Northwell
Funding: NIH, Northwell
Technology-Mediated Self-Reflection for Mental Health
In recent years, machine learning algorithms have begun to support tracking and intervention technologies in many health domains. However, most such tools enable only simplistic mechanisms to review one’s data and require high compliance from individuals actively volunteering relevant information. These limitations prevent such tools from effectively supporting reflection, i.e., conscious re-examination of prior experiences to form new understanding. Reflection is a key to improved health and health maintenance, and recent work has shown the promise of data-driven health reflection. Effectively supporting reflection, however, requires more sophistication than simply showing a patient their data, especially in the context of mental health. Motivated by these observations, in this project, we are developing tools to support reflection for eating disorders (ED) by combining voluntarily shared and unobtrusively gathered social media data with strategic presentation of machine learning analysis to cater to multi-stakeholder needs. We believe this research will result in novel mechanisms to support the treatment of devastating mental illnesses, going beyond existing personal informatics tools by being sensitive to the complex psychological struggles of ED.
SocWeB Collaborators (Current): Kelsie Belan, Munmun De Choudhury
Collaborating Institutions: Lehigh University
Funding: NSF
Human-Centered Technologies for Teen Online Safety
This project aims to examine, build, evaluate, and bring to market state-of-the-art technologies to detect adolescent (ages 13-17) online risk behaviors (e.g., cyberbullying, sexual solicitations and grooming, exposure to explicit content, information breaches, non-suicidal self-injury, suicidal ideation, and other imminent risks).
Teens are the future of our networked society; therefore, protecting and empowering teens to mitigate the harm caused by online risks is a timely and critical problem that must be addressed to ensure the well-being and safety of our youth. Networked technology is an ever-present force in the lives of nearly all teens in the United States; according to Pew Research, 92% of teens (ages 13-17) in the U.S. access the internet on a daily basis, 71% have at least one social media account, and 73% have access to mobile smartphones. While internet-enabled technologies afford a number of benefits and opportunities to our youth, they also expose teens to a myriad of online risks. Through the contributions of this project, we suggest a significant societal pivot, where adolescent online safety becomes a shared responsibility for all, especially the online platforms in which teens encounter risks, thereby serving teens, their families, and society as a whole.
SocWeB Collaborators (Current): Munmun De Choudhury
Collaborating Institutions: University of Central Florida, University of South Florida, Mozilla
Funding: NSF
The WorkWell Project
The WorkWell project is a collaboration with with researchers in organizational psychology and organizational management related to the topic of WORKplace WELLbeing. We examine and understand how individuals adapt and perceive psychological, cognitive, and environmental demands of workplace. We model attributes of workplace wellbeing and functioning by leveraging social media and other sources of employee-contributed online data (eg., Glassdoor). These platforms enable individuals to not only express, disclose, and share their day-to-day and long-term perceptions of work-life, but also to self-present and self-promote their individual portfolio (eg., on LinkedIn). We demonstrate that these unobtrusive, large-scale, and longitudinal sources of data can help us gauge newer insights on job satisfaction, workplace functioning, organizational wellness, and organizational performance. The findings are to be evaluated against gold-standard metrics and survey-based measures to establish the feasibility of our study in the context of these online data sources. One of the primary implications of our work is to build and design complementary online data driven technologies with empirically guided richer information of employee and workplace wellbeing at scale.
SocWeB Collaborators (Current): Koustuv Saha, Vedant Das Swain, Asra Yousuf, Munmun De Choudhury
Collaborating Institutions: Purdue University, Ohio State University
Ethics and Privacy
Algorithmic inferences derived from social media data of individuals hold great potential in supporting early detection and treatment of mental disorders and in the design of interventions. At the same time, the outcomes of this research can pose great risks to individuals, such as issues of incorrect, opaque algorithmic predictions, involvement of bad or unaccountable actors, and potential biases from intentional or inadvertent misuse of insights. Amplifying these tensions, there are also divergent and sometimes inconsistent methodological gaps and under-explored ethics and privacy dimensions. Noting these challenges, this project seeks to identify issues in algorithmic prediction of mental health status on social media data, such as the gap between ethics committees and participants in such research, on what can be sensitive and sometimes stigmatizing data; tensions in methods, such as construct validity and bias, interpretability of algorithmic output, and privacy; or the manner in which humans and stakeholders are conceptualized within the research process in this emergent field.
SocWeB Collaborators (Current): Stevie Chancellor, Sindhu Ernala, Tahirah Ahmad, Munmun De Choudhury
Collaborating Institutions: Northwell, CDC, Lehigh University, Columbia University
Funding: NIH