Alcohol use effects on adolescent brain development revealed by simultaneously removing confounding factors, identifying morphometric patterns, and classifying individuals

Sang Hyun Park, Yong Zhang, Dongjin Kwon, Qingyu Zhao, Natalie M. Zahr, Adolf Pfefferbaum, Edith V. Sullivan, Kilian M. PohlScientific 2018 · DOI

Abstract

Group analysis of brain magnetic resonance imaging (MRI) metrics frequently employs generalized additive models (GAM) to remove contributions of confounding factors before identifying cohort-specific characteristics. However, considering confounding factors before group analysis may remove information relevant for distinguishing specific cohorts. This study introduces Joi-GAM-Class, a machine-learning model that identifies neuromorphometric patterns by simultaneously training a GAM and a generic classifier. Using a baseline dataset from the National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA, N = 705), the joint model identified a unique pattern of eight brain regions in adolescents who misuse alcohol. This approach proved more accurate and impartial to confounding factors than traditional sequential methods, aligning with established alcohol literature.

Introduction

Studying the adolescent brain (ages 12–21) presents a unique challenge: the brain is a moving target. Structural changes—such as gray matter thinning and increased white matter organization—occur rapidly. When investigating the effects of alcohol misuse, researchers must distinguish between alcohol-induced changes and natural developmental variations or socioeconomic factors.

Traditional statistical approaches use a sequential two-step process: first, a Generalized Additive Model (GAM) "cleans" the data of confounds like age; second, researchers look for patterns in the residuals. This paper argues that this "pre-cleaning" can inadvertently strip away subtle signals necessary for accurate classification.

Methodology: The Joi-GAM-Class Model

The researchers proposed a joint optimization framework, Joi-GAM-Class (Joint GAM Classification). Unlike sequential models, this approach embeds the GAM directly into the classification's energy function.

Conceptual comparison of data harmonization: Sequential vs. Joint optimization. (a) Synthetic data with age as a confound. (b) Raw scores. (c) Sequential approach removes signal. (d) Joint approach preserves group separation.

Key features of the model include:

  • Joint Optimization: Simultaneously determines parameters for confound removal and weights for a sparse logistic classifier, ensuring harmonization is informed by the classification goal.
  • Sparsity Constraints: Forces the model to identify a small, interpretable subset of the most informative brain regions.
  • Multimodal Integration: Analyzes 144 metrics across macrostructural MRI (volumes) and microstructural Diffusion Tensor Imaging (DTI).

Results

The study utilized the NCANDA baseline dataset, comparing 671 "minimal drinkers" to 34 "regular drinkers."

Classification Performance

Joi-GAM-Class achieved a normalized accuracy of 75.9%, significantly outperforming sequential GAM approaches. Crucially, the model passed an "age-test"—it could not distinguish between older and younger non-drinkers, proving it was truly age-blind and not simply using age as a proxy for drinking status.

The Eight-Region Signature

The model identified a stable neuromorphometric pattern involving eight specific regions: the lateral ventricles, mid-posterior and central corpus callosum, centrum semiovale, anterior corona radiata, posterior thalamic radiation, fornix, and cingulate gyrus.

Normalized frequency of patterns identified by the Joi-GAM-Class model, demonstrating the stability of the sparse signature.

Regular drinkers exhibited smaller volumes in the mid-posterior corpus callosum and centrum semiovale, alongside higher axial diffusivity in the fornix. These findings reinforce existing literature regarding alcohol's impact on white matter integrity during adolescence.

Discussion and Conclusion

The core insight of this research is that data harmonization should not be task-agnostic. By allowing the classification task to influence how confounds are removed, Joi-GAM-Class preserved critical group differences that sequential methods discarded as noise.

While the small sample size of the drinking cohort (N=34) and the cross-sectional nature of the baseline data are limitations, the study provides a robust framework for future longitudinal analysis. The findings suggest that joint optimization is a superior method for uncovering the subtle structural signatures of substance use in the developing brain.

Summary updated

Abstract

Group analysis of brain magnetic resonance imaging (MRI) metrics frequently employs generalized additive models (GAM) to remove contributions of confounding factors before identifying cohort specific characteristics. For example, the National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA) used such an approach to identify effects of alcohol misuse on the developing brain. Here, we hypothesized that considering confounding factors before group analysis removes information relevant for distinguishing adolescents with drinking history from those without. To test this hypothesis, we introduce a machine-learning model that identifies cohort-specific, neuromorphometric patterns by simultaneously training a GAM and generic classifier on macrostructural MRI and microstructural diffusion tensor imaging (DTI) metrics and compare it to more traditional group analysis and machine-learning approaches. Using a baseline NCANDA MR dataset (N = 705), the proposed machine learning approach identified a pattern of eight brain regions unique to adolescents who misuse alcohol. Classifying high-drinking adolescents was more accurate with that pattern than using regions identified with alternative approaches. The findings of the joint model approach thus were (1) impartial to confounding factors; (2) relevant to drinking behaviors; and (3) in concurrence with the alcohol literature.
BibTeX
@inproceedings{park2018alcohol,
  author = {Sang Hyun Park and Yong Zhang and Dongjin Kwon and Qingyu Zhao and Natalie M. Zahr and Adolf Pfefferbaum and Edith V. Sullivan and Kilian M. Pohl},
  title = {{Alcohol use effects on adolescent brain development revealed by simultaneously removing confounding factors, identifying morphometric patterns, and classifying individuals}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-018-26627-7},
}
Figure