Individual Differences in The Intergenerational Transmission of Attention-Deficit/Hyperactivity Disorder: Genetic and Environmental Factors

This repository contains the R analysis, visualization, and sensitivity refit scripts for the study:

Pak M, Lee J, Lee E, Kim B-G, Kim G, Posner J, Joo YY, Cha J.
Individual Differences in The Intergenerational Transmission of Attention-Deficit/Hyperactivity Disorder: Genetic and Environmental Factors.
Translational Psychiatry (under revision; preprint available at Research Square: https://doi.org/10.21203/rs.3.rs-7405635/v1).

---

Overview

The scripts in this repository implement generalized random forest (GRF; Athey, Tibshirani, & Wager, 2019; R package grf) models in the Adolescent Brain Cognitive Development (ABCD) Study (Release 4.0) to:

1. Estimate the covariate-adjusted average association (average treatment effect, ATE) between parental ADHD symptoms (asr_scr_adhd_t >= 65) and child ADHD symptoms (cbcl_scr_dsm5_adhd_t).
2. Test for omnibus heterogeneity in the conditional association (test_calibration slope on out-of-bag causal forest predictions) across four nested covariate configurations (environmental, genetic [30 polygenic scores], and structural brain volume covariates) at baseline and 1-, 2-, and 3-year follow-ups.
3. Identify candidate covariates along which the fitted conditional association varies using a 500-iteration permutation-based variable importance test (after Gong et al., 2021) and characterize their partial dependence curves.
4. Execute comprehensive revision sensitivity analyses (revision_r1/), including:
   • Measured baseline child BMI sensitivity refits (Table S14: substituting measured child BMI for BMI PGS, and including both simultaneously).
   • Longitudinal causal forest refits adjusting for baseline child ADHD symptoms at 1-, 2-, and 3-year follow-ups (Table S15).
   • Non-oversampled European-ancestry sensitivity refits under both sex-matched ($N = 726$) and full unexposed ($N = 6,106$) comparison groups (Table S16).
   • Shared-informant exclusion sensitivity refits (excluding parental psychopathology indicators, child sleep problems, or all parent-reported covariates).

---

Repository Structure

Path	Description
Analysis2.R	Complete primary pipeline: ABCD Release 4.0 table loading (baseline_year_1_arm_1 and longitudinal waves), variable construction, near-zero-variance filtering (caret v6.0-93), $k$-nearest-neighbor imputation (impute v1.72.3, $k = 10$), sex-matched control downsampling (set.seed(1)), European SMOTE oversampling (DMwR v0.4.1), 25,000-tree causal forest fitting (5 seeds × 5,000 trees, clustered on site_id_l), permutation importance testing, partial dependence computation, and regional brain volume correlation tests (Tables S4–S11).
Visualization.R	Script generating Figure 1 (ATE, calibration test, and longitudinal trajectory plots) and Figure 2 (top 20 variable importance bar plots and partial dependence curves).
Code.R	Standalone template script for fitting the 25,000-tree causal forest, 500-permutation variable importance threshold, and partial dependence curves on a single prepared analytic data frame (data_multi_env_gen).
revision_r1/extract_summary_diagnostics.R	Reproducible extraction of raw/analytic CBCL $t$-score standard deviations, out-of-bag individual CATE 95% confidence intervals, partial dependence quartile contrasts (Q1–Q3, Q0–Q4), PGS discovery sample size rank correlations, and sample verification.
revision_r1/run_refit_grf_models.R	Reproducible causal forest sensitivity refits for non-oversampled European samples (Table S16), baseline-adjusted longitudinal outcomes (Table S15), and shared-informant covariate exclusions.
revision_r1/run_bmi_sensitivity_refit.R	Reproducible causal forest sensitivity refits evaluating measured baseline child BMI alongside or in place of the BMI polygenic score (Table S14).
revision_r1/results/	Aggregate machine-readable JSON summaries and Markdown tables (summary_diagnostics.json, sensitivity_refits.json, table_s14_bmi.md, table_s16_european.md).

---

Data Availability

The data analyzed in this study are from the Adolescent Brain Cognitive Development (ABCD) Study Data Release 4.0 (doi:10.15154/1523041), available to qualified researchers with an approved Data Use Certification (DUC) through the NIMH Data Archive (https://nda.nih.gov).

In strict compliance with the ABCD Data Use Certification, no individual-level participant data, subject identifiers (subjectkey), or derived participant-level .RData workspaces are distributed in this repository. To reproduce the analyses, obtain ABCD Release 4.0 tabular files from the NIMH Data Archive and place them in the working directory expected by Analysis2.R.

---

Software Environment

As documented in the manuscript and Supplementary Material:

• R: v4.2.1 (RStudio 2024.4.2.764; revision sensitivity refits verified under R 4.4)
• grf: v2.3.2 (generalized random forests; Athey et al., 2019)
• caret: v6.0-93 (near-zero-variance filtering of structural brain features)
• impute: v1.72.3 ($k$-nearest-neighbor missing data imputation, impute.knn)
• DMwR: v0.4.1 (synthetic minority oversampling in the secondary European analysis; archived on CRAN on 2021-03-16 and installable from https://cran.r-project.org/src/contrib/Archive/DMwR/DMwR_0.4.1.tar.gz)

Note on grf::test_calibration: grf::test_calibration returns one-sided $p$-values (pnorm(t.value, lower.tail = FALSE)). Throughout the revised manuscript and Supplementary Material, all calibration test $p$-values are reported as two-sided (2 * pnorm(-abs(t.value))).

---

License

This code is released under the MIT License.

Contact

Corresponding author: Jiook Cha, PhD, Department of Psychology, Seoul National University (connectome@snu.ac.kr).