ITHS Tools
Login
Need Help? Have a Question?

Contact the Research Navigator

Connect with Us
Visit UW in a new window/tab
 
ITHS › ITHS Impact › Translational Science Benefits Model Profiles › Deep Learning to Identify Adolescent Development Patterns

Deep Learning to Identify Adolescent Development Patterns

TRANSLATIONAL SCIENCE BENEFITS MODEL PROFILE

Benefits

Download the 1-page project profile.

Summary

Mental health issues often begin to show up during the teen years, but the causes can start much earlier, during childhood. These issues often overlap and can happen at the same time, which makes them hard to diagnose using only one label or category. Because of this, we are using a method that doesn’t depend on labels to better understand mental health patterns. Our goal is to find groups of teens who follow different paths in how they think and behave as they grow. To do this, we will use different kinds of data collected over time. We will use deep learning methods (VaDER and CRLI) to group teens based on this information. Then, we will look at how each group connects to things like personal traits, risk factors, and mental health outcomes.

Significance

Mental health problems are becoming more common around the world. When these problems begin in the teen years and are not treated early, the symptoms often continue into adulthood. Some teens show early signs that suggest they may face serious problems later, but these signs can be missed because they don’t match current ways of diagnosing mental illness. The ABCD Study is a large, long-term project that follows nearly 12,000 young people from childhood into early adulthood. It collects many types of health, behavior, and brain data from a diverse group of participants. This rich dataset is unlike any before it, and it allows us to use new methods to study complex health trends. In this project, we will use this data to find common patterns in how teens think and behave as they grow. Finding these patterns can help guide future research, medical care, and treatment planning.

Translational Science Solution

This work will help build clear models that use patient starting information to predict who may be at higher risk for poor outcomes. These models can support care that is based on risk and help patients and doctors make decisions together. This can lead to more personalized care and a stronger focus on preventing problems instead of reacting after they happen.

Benefits

  • Demonstrated benefits are those that have been observed and are verifiable.
  • Potential benefits are those logically expected with moderate to high confidence.

Association testing between subgroups and outcomes to potentially inform diagnostic practices.
Potential.

Clinical Benefits:
Diagnostic Procedures

Subgroup identification could inform future observational cohort study design and collection of specific follow-up measures and assessments.
Potential.

Clinical Benefits:
Investigative Procedures

Identification of risk factors could inform patient care (screening, intervention, etc.) guidelines around mental and behavioral health
Potential.

Clinical Benefits:
Guidelines

Preemptive intervention or screening based on presence of factors associated with trajectories of concern
Potential.

Clinical Benefits:
Therapeutic Procedures

Subgroups enriched for co-occurring diagnoses could inform trans-diagnostic drug discovery or repurposing.
Potential.

Clinical Benefits:
Drugs

Longitudinal models like CRLI and VaDER could be deployed actively in the electronic health record to assess trajectories prospectively as measures are collected over time.
Potential.

Clinical Benefits:
Software Technologies

Screening practices informed by this work could involve community outreach at schools, religious centers, extracurriculars, and other places of gathering for youths.
Potential.

Community Benefits:
Community Health Services

At-home tracking of children’s risk factors could help parents and caregivers to identify concerning trends earlier.
Potential.

Community Benefits:
Consumer Software

This work can inform toolkits and graphics that show the importance of monitoring of multiple aspects of health during adolescence.
Potential.

Community Benefits:
Health Education Resources

Enrichment-depletion analysis of subgroups may help inform priorities for improving health care access or delivery.
Potential.

Community Benefits:
Health Care Accessibility

Subpopulations most vulnerable for self-injurious or suicidal behavior may be identified preemptively and intervened on as a result of this work.
Potential.

Community Benefits:
Life Expectancy & Quality of Life

By reducing the use of higher dose opioids and long-term opioid therapy, the Six Building Blocks program could reduce the substantial economic cost of opioid use disorder on society.10
Potential.

Economic Benefits:
Societal & Financial Cost of Illness

This work can be disseminated at adolescent health, epidemiology, suicidology, and other related venues to educate researchers, healthcare practitioners, and policymakers.
Demonstrated.

Policy Benefits:
Scientific Research Reports

Identification of risk factors could inform patient care (screening, intervention, etc) policies for mental and behavioral health at the government or organizational level.
Potential.

Policy Benefits:
Policies

This research has clinical, community, economic, and policy implications. The framework for these implications was derived from the Translational Science Benefits Model created by the Institute of Clinical & Translational Sciences at Washington University in St. Louis.

Funding

Funding provided by the Institute of Translational Health Sciences under grant TL1 TR002318

References

  1. Duffy KA, Gandhi R, Falke C, et al. Psychiatric Diagnoses and Treatment in Nine- to Ten-Year-Old Participants in the ABCD Study. JAACAP Open. 2023;1(1):36-47. doi:10.1016/j.jaacop.2023.03.001
  2. Brady KT, Killeen TK, Lucerini S. Comorbidity of Psychiatric Disorders and Posttraumatic Stress Disorder. Prim Care Companion CNS Disord. 2000;2(Suppl 2: Editor Choice):3403.
  3. Paus T, Keshavan M, Giedd JN. Why do many psychiatric disorders emerge during adolescence? Nat Rev Neurosci. 2008;9(12):947-957. doi:10.1038/nrn2513
  4. de Jong J, Emon MA, Wu P, et al. Deep learning for clustering of multivariate clinical patient trajectories with missing values. GigaScience. 2019;8(11):giz134. doi:10.1093/gigascience/giz134
  5. Ma Q, Chen C, Li S, Cottrell GW. Learning Representations for Incomplete Time Series Clustering. Proc AAAI Conf Artif Intell. 2021;35(10):8837-8846. doi:10.1609/aaai.v35i10.17070
  6. Dugré JR, Potvin S. Developmental multi-trajectory of irritability, anxiety, and hyperactivity as psychological markers of heterogeneity in childhood aggression. Psychol Med. 2022;52(2):241-250. doi:10.1017/S0033291720001877

Research Team

Bhargav Vemuri, PhD Candidate (University of Washington Biomedical Informatics and Medical Education)

  • Laura Richardson, MD, MPH (Seattle Children’s Research Institute)
  • Molly Adrian, PhD (Seattle Children’s Research Institute)
  • Jennifer Bramen, PhD (Pacific Neuroscience Institute)
  • Christine Schaeffer, PhD, MBA, MPH (Providence Health & Services)
  • Jennifer Hadlock, MD (Institute for Systems Biology)

Learn More About the Project

2024 ABCD Insights & Innovations Meeting (AIIM): poster | recording