Developing Artificial Intelligent Tools to Improve Mental Health Diagnosis for Telehealth Behavioral Health Services during COVID-19

Rural populations are highly vulnerable to behavioral health problems during the COVID-19 pandemic. Rural children have higher rates of behavioral disorders and more severe symptoms than urban youth, and rural families experience higher rates of poverty, health problems, and addictions, which are further challenged by poor access to care and scarce community resources. Community telehealth behavioral services delivered to the home via videoconferencing systems have become the most available and safest option for child behavioral treatment during the COVID-19 pandemic. However, behavioral health providers are poorly equipped to engage families with home telehealth services. Caregiver engagement is negatively impacted by technological constraints and pandemic social isolation conditions. Providers have limited ability to assess non-verbal cues (e.g. they only see the caregiver’s face), and typical engagement behaviors are different (e.g. “eye contact” by staring at a camera). Caregivers, in turn, may also have difficulty engaging with providers because of the new health care delivery system (e.g. virtual-only contact) as well as their personal experience of social isolation during the pandemic (e.g. difficulty re-connecting to discuss family issues). Caregiver engagement is a priority in child behavioral health systems because it is fundamental to optimize child retention in care and equally important to improve child outcomes. Caregivers are typically the primary target for child behavioral interventions (e.g. parenting skills training), and positive caregiver engagement is associated with improved child behavioral outcomes. Unfortunately, most youth in community clinics fail to complete evidence-based therapy treatments because their families drop out of care.

To address this, GAMMA is collaborating with the child behavioral health experts at the University of Maryland Baltimore for an UMCP-UMB MPower COVID-19 project. Artificial Intelligence (AI) strategies assess engagement as an emotional state. We propose a new semi-supervised AI learning paradigm using Dual Attention Networks (DANs), which will jointly leverage visual and vocal/text attention mechanisms from our caregivers to capture the fine-grained interplay between visual cues and language. DANs attend to specific regions in images and words in text/voice through multiple steps and gather essential information from both modalities to measure affective states, i.e. engagement. This multimodal fused analysis will be used to derive an automated engagement score. From a health care perspective, an automated engagement score can be used in future studies to give providers real-time, quantitative feedback during or immediately post-session. Relevant to artificial intelligence science, this research can be used to model pandemic-related contextual factors (e.g. social isolation), and will influence the discordance between trained observer ratings and self-report on an emotional state.

We are also working with rural child behavioral health providers (social workers, counselors, nurses, psychologists) and caregivers enrolled in treatment at the Community Behavioral Health or Maple Shade Youth and Families services clinics in Maryland Mid-Shore and Eastern Shore counties (10 clinics, 6 rural counties). These clinics are overseen by the Medical Director, Dr. Sushma Jani.

Whether mental health is measured using automated computational techniques or through more traditional measures (e.g. self-report questionnaires), it is not known how the extended social isolation experienced by families during the pandemic may impact caregiver engagement and related behaviors. This information is important for both AI learning paradigms (e.g. identifying engagement behaviors) and engagement techniques (e.g. responding to non-verbal cues). Van Bavel et al. reviewed the wide range impact of the pandemic on social perceptions and behaviors, including changes in social norms (e.g. no handshakes), the perceived threat associated with social interactions (e.g. COVID-19 transmission risk), reduced social support for coping (e.g. social isolation) and mistrust of authority figures (e.g. controversies about COVID-19 science and patient care). Social role functioning can also be impaired as typical activities (e.g. attending church, going to work) may be on hold during the pandemic. It is unclear if social isolation and impaired social role functioning may impact either the affective experience of engagement or the presence of typical engagement behaviors.