When we see that a child is showing high levels of mental health symptoms, doctors can step in right away. Acting early makes a big difference. It can lead to:
- Better health now and in the future for children, teens, and adults
- Improved quality of life
- Faster access to the right services
- Less illness later on
Mental Health Crisis: Our Primary Focus
Mental health disorders remain one of the biggest challenges in healthcare today. In the U.S., more than one in five adults—roughly 60 million people—experience some form of mental illness each year. For young people, the picture is even harder to take in: suicide has become the second leading cause of death among those aged 10 to 34.
In late 2024, we decided it was time to shift our attention. Instead of focusing only on anxiety, we turned toward the broader problem of mental health crises. The aim was simple but urgent: to catch kids and teens before they end up in the emergency department or hospitalized when a crisis might have been prevented.
To do this, we built the Mental Health Crisis model. It draws on three streams of information: what’s already in the electronic health record (EHR), details about a child’s environment things like greenspace or traffic and social factors such as poverty, housing stability, and the makeup of their neighborhood. Taken together, these pieces give us a sharper picture of when a child might face their first crisis, or when another could follow. At this point, the model can estimate risk at intervals of 1, 2, 3, 6, and 12 months.
By the end of 2025, we’ll be putting this model into clinical practice at Cincinnati Children’s. Right now, it predicts crises with about 82% accuracy at six months and 85% at a year.
Anxiety: Our Initial Focus
In our work we also focus on clinical anxiety (with its clinical significant symptoms) and the corresponding DSM-5 anxiety disorder diagnoses.
For this research we developed a clinical anxiety phenotype in partnership with the clinicians at Cincinnati Children’s who treat patients with anxiety. We identified key instruments, indicators and characteristics considered when diagnosing a patient with anxiety.
The resulting set of anxiety indicators contains data from all aspects of a patient’s life, including data from the electronic health record (EHR), such as:
- Demographic data
- Diagnosis information
- Instruments/survey results
- Keywords or phrases documented in the notes
- Medications
- Patient history
Other information used to diagnose and identify anxiety are data from outside the patient’s clinical history, including:
- Absenteeism
- Academic performance
- Avoidance
- Behavior issues
- Community resilience
- Environmental exposures
- Housing conditions
- Material deprivation
|
- Noise
- Patient mannerisms
- Poverty
- Violent crime
- Voice inflections
- Weather
- Worrying
|
To design a mental health trajectory, we performed three separate chart review efforts with three main goals to:
- Identify where key information is captured for patients in structured and unstructured data areas
- Determine if there are criteria that can be used to create clean case/control cohorts
- Understand the processes that clinicians use to make these assessments
In June 2022, we performed our first concept design experiment. We trained a machine learning model on clinical and environmental data from 3,000 anxiety and non-anxiety patients. Then we analyzed what the model would say about individual known-anxiety and known-non-anxiety patients at different points in their medical history. This exercise allowed us to see what similar trajectory-style models may look like.
Foundational Data Lake
The team has created a foundational data lake of clinical structured and unstructured data transformed to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM).
Additionally, we developed geospatial linkage methods to allow for the linkage of environmental and population-level information to determine the impact certain events may have on clinical outcomes.
Our team has incorporated environmental and population data into the foundational data lake.
Our environmental datasets include:
- Daily ambient fine particulate matter
- National Walkability Index
- Neighborhood-level measures of environmental exposures related to lead paint, diesel combustion, traffic, water discharges and superfund sites
- Proximity and length of major roadways and traffic
- Satellite-based high-resolution land usage (greenspaces, urban imperviousness and development)
- Satellite-derived greenness
- Weather
Our socioeconomic datasets include:
- Child Opportunity Index
- Community Deprivation Index
- Community Resilience
- Mental Health Professional Shortage Areas
- Neighborhood Atlas
- US Department of Agriculture (USDA) Food Access data
Program Goals
The Mental Health Trajectory Program includes several scientific and administrative goals.