Integrative data analysis (IDA)

• Methodological framework that allows for the simultaneous analysis of data from multiple studies
  • Not a specific analysis approach
• Involves several steps, including
  • Combining datasets
  • Coding study characteristics
  • Harmonizing measures
  • Conducting the statistical analysis

Why do IDA?

• Data synthesis
• Larger sample
  • Increase statistical power
  • Increase sample heterogeneity
  • Increase frequency of low base-rate behaviors
• Improved replicability

TIDAL project

• The ADHD Teen Integrative Data Analysis Longitudinal dataset
  • R03 from National Institute of Mental Health (NIMH)
  • Coxe and Sibley: joint co-PIs
  • Combined dataset is available at NIMH Data Archive: https://nda.nih.gov/
  • Six articles published
    • 1 protocol, 2 harmonization, 3 applied

The studies

• Four RCTs of psychosocial interventions for adolescents with ADHD
  • 854 total participants (128, 325, 123, 278)
  • Three measurements: baseline, post-treatment, follow-up
  • Five different treatment conditions: STAND, STAND-G, STP-A, usual care, no treatment
• Varied: time of year, setting, clinician type, treatment duration, some specific measures

Data harmonization

• Create common measures using moderated nonlinear factor analysis (MNLFA; Bauer, 2017)
  • Invariance testing / differential item functioning (DIF) testing
• Two measures harmonized across studies in this project
  • Parent depression: Different measures
  • Adolescent ADHD symptoms: Different versions

Attention-deficit / hyperactivity disorder (ADHD)

• Neuro-developmental disorder characterized by
  • Inattention problems
  • Hyperactivity (especially in children)
• Treatments
  • Medication: Stimulants like Adderall and Ritalin
  • Behavioral

Heterogeneity

• Adolescents with ADHD display heterogeneity
  • Symptoms / presentation
    • 6 of 18 symptoms across 2 broad domains (inattention & hyperactivity)
  • Response to treatment
    • Both behavioral and medication
  • Problems related to and less related to their diagnosis
    • Discipline, comorbidities

Why does that matter?

• Screening and diagnosis biases
  • Less likely to be diagnosed:
    • Girls and young women
    • African American children and adolescents
    • Older adolescents
    • Other sources of disadvantage: SES, divorced parents, non-English
• Treatment matching
  • Modular treatments

Explore heterogeneity when seeking treatment

• What do the adolescents look like when they show up (baseline) seeking treatment for ADHD?
  • Are there groups of people that are similar to one another but distinct from each other?
  • Are there demographic differences between these groups?
• Latent profile analysis (LPA)

Latent profile analysis (LPA)

• Mixture modeling: Identify latent (unobserved) groups of individuals that are relatively homogenous, based on a set of continuous and/or categorical indicators
• Similar to clustering methods
  • Probabilistic assignment, classification uncertainty
• Use in the combined TIDAL sample (n = 854)
  • LPA requires a large sample (200+ but more is better!)
    • Individual samples ranged from 123 to 325

Indicator variables: Externalizing comorbidity

• Oppositional defiant disorder (42.5%) / conduct disorder diagnosis (5.7%)
  • Disruptive behavior disorder
• ADHD combined subtype
  • Both inattention and hyperactivity (53%)
  • Versus inattention only (46.6%) or hyperactivity only (0.4%)

Indicator variables: Internalizing comorbidity

• Internalizing disorder diagnosis
  • Depression
  • Anxiety

Indicator variables: Academic problems

• Organization, time management, planning (OTP)
  • Targeted in behavioral treatments in adolescents
• Disruptive classroom behavior / discipline problems

Indicator variables: Cognitive and academic

• Grade point average (GPA)
  • Objective measure of academic performance
  • Extremely meaningful to parents and schools
• IQ
  • Normed to have mean = 100, SD = 15
  • All participants had normal range IQ (\(\ge\) 70)

Analysis

• Mplus 7.2
  • Structural equation modeling (SEM) software
• Mixture model with 8 indicators
• Fixed effects IDA
  • 3 dummy codes indicating study
  • Regress categorical latent variable on dummy codes

Model selection

• Select number of profiles based on
  • (Sample-size adjusted) BIC
  • Vuong-Lo-Mendell-Rubin LR test
  • Theory: Do the profiles make sense?
• 3 profiles
  • Entropy = 0.80
  • Correct classification probability: 0.80, 0.95, 0.87

Three profiles

Three profiles

• ADHD simplex (n = 544)
  • Majority of the sample, no notable comorbidities
• ADHD + internalizing (n = 97)
  • Very high probability of comorbid depression and anxiety
• Disruptive / disorganized (n = 213)
  • Very high probability of ADHD combined sub-type (both inattention and hyperactivity)
  • More organization problems and discipline problems

Demographic variables

• Age
• Sex
• Family adversity
  • Single parent, < HS education, non-English speaking, 3+ children
• Race / ethnicity
  • Non-Hispanic White, African American, Hispanic any race

Demographic differences: Sex

Demographic differences: Adversity

Demographic differences: African American

Summary of demographic results

• ADHD with internalizing disorders
  • More girls and young women
• ADHD with disruptive (ODD, CD) / disorganized behaviors
  • More boys and young men
  • More adversity (single parent, <HS, non-English, 3+ children)
  • More African American adolescents
• ADHD without comorbidities
  • Less adversity (than disruptive)

Conclusions: Statistical

• This analysis could only happen because of IDA
  • Larger sample required for LPA
  • Increased heterogeneity
• Open science
  • Data sharing
  • Secondary data analysis

Conclusions: Clinical treatment matching

• Sex differences in screening and diagnosis
  • Girls less often diagnosed with ADHD, diagnosed with internalizing
• Academic problems due to depression vs organizational problems
  • Is there a secondary disorder to treat?
• Organizational problems and ODD/CD respond poorly to medication
  • Behavioral treatments (especially involving parents) more effective
• Adolescents experiencing additional forms of adversity
  • Require additional social supports beyond treatment

Thank you

Demographic differences: Age

Demographic differences: White

Demographic differences: Hispanic

Moderated nonlinear factor analysis

References