Dimensions of early life adversity links to trimodal brain age in youth
Poster No:
1232
Submission Type:
Abstract Submission
Authors:
Dani Beck1, Lucy Whitmore2, Niamh MacSweeney1, Alexis Brieant3, Ann-Marie de Lange4, Lars Westlye5, Kathryn Mills2, Christian Tamnes1
Institutions:
1University of Oslo, Oslo, Norway, 2University of Oregon, Eugene, OR, 3University of Vermont, Vermont, VT, 4University of Lausanne, Oslo, Norway, 5Norwegian Centre for Mental Disorders Research (NORMENT), Oslo University Hospital, Oslo, Norway
First Author:
Co-Author(s):
Lars Westlye
Norwegian Centre for Mental Disorders Research (NORMENT), Oslo University Hospital
Oslo, Norway
Norwegian Centre for Mental Disorders Research (NORMENT), Oslo University Hospital
Oslo, Norway
Introduction:
Exposure to early-life adversity (ELA) impacts brain development1,2. Research suggests a discordant relationship between ELA factors related to threat and deprivation on brain maturation. Threat factors such as physical abuse are linked with smaller brain volumes and increased functional activity, while this is not observed with deprivation1. With increasing evidence of different features of ELA associated with unique brain outcomes, data-driven efforts that characterise the dimensionality of ELA have been developed3.
The ABCD Study (N=11,900, age mean=10.75) overcomes existing limitations of small samples and cross-sectional designs. Brain age prediction offers an individualised marker for assessing brain maturity. Looking at the deviation from age-expected patterns (i.e., brain age gap (BAG)) and how deviation changes over time provides an indication of whether ELA factors are associated with accelerated or delayed brain maturation.
We aimed to investigate cross-sectional and longitudinal associations between ten dimensions of ELA and BAGs derived from three brain age models trained on T1, DTI, and rsfMRI data. We expected threat-related dimensions of ELA to be linked with higher BAGs, and deprivation-related dimensions to be linked with lower BAGs. We also expected links to exacerbate over time.
The ABCD Study (N=11,900, age mean=10.75) overcomes existing limitations of small samples and cross-sectional designs. Brain age prediction offers an individualised marker for assessing brain maturity. Looking at the deviation from age-expected patterns (i.e., brain age gap (BAG)) and how deviation changes over time provides an indication of whether ELA factors are associated with accelerated or delayed brain maturation.
We aimed to investigate cross-sectional and longitudinal associations between ten dimensions of ELA and BAGs derived from three brain age models trained on T1, DTI, and rsfMRI data. We expected threat-related dimensions of ELA to be linked with higher BAGs, and deprivation-related dimensions to be linked with lower BAGs. We also expected links to exacerbate over time.
Methods:
Age prediction was carried out using XGBoost regression4. Here, 50% of the ABCD Study sample (baseline and two-year follow-up; N=~9000) was used as the test set and 50% was used for model training (N=~7200) and ten-fold cross-validation (N=~1800). R2, RMSE, and MAE assessed prediction accuracy; age bias was statistically corrected.
Bayesian multilevel modelling tested the associations between dimensions of ELA and T1, DTI, and rsfMRI BAG. BAG was entered as the dependent variable with each ELA dimension separately entered as the independent fixed effect along with sex, with subject ID as the random effect. Longitudinal ELA effects were assessed with interaction effects of time-point.
Bayesian multilevel modelling tested the associations between dimensions of ELA and T1, DTI, and rsfMRI BAG. BAG was entered as the dependent variable with each ELA dimension separately entered as the independent fixed effect along with sex, with subject ID as the random effect. Longitudinal ELA effects were assessed with interaction effects of time-point.
Results:
The DTI brain age model was the most accurate (r=.66, p<.01, MAE=.71), followed by T1 (r=.59, p<.01, MAE=.79) and rs-fMRI (r=.41, p<.01, MAE=.89).
We found a positive association between ELA dimension F2, representing factor loadings from measures of socioeconomic disadvantage and neighbourhood safety, and T1 BAG, indicating that this dimension was associated with older-looking brains. Further, we found a negative association between ELA dimensions F3 (secondary caregiver lack of support) and F4 (primary caregiver lack of support) and T1 BAG, indicating that dimensions related to neglect are associated with younger-looking brains. In terms of interaction effects, we found negative associations between F10 (lack of supervision) and T1 BAG, indicating that unsupervised youth diverge more from normative age patterns over time.
In line with the T1 BAG findings, we found a positive association between F2 and DTI BAG. In terms of interaction effects, we found a positive association between ELA dimension F1 (caregiver psychopathology) and DTI BAG.
For fMRI, we found a positive association between F2 and fMRI BAG, aligning with findings with T1 and DTI BAG. Positive associations were also found for F1 and fMRI BAG, aligning with DTI findings, and F6 (caregiver substance abuse and separation from biological parents), F8 (family aggression), and F9 (trauma exposure) and fMRI BAG. These positive associations indicate that dimensions related to threat are related to older-looking brains. A negative association was found between F4 and fMRI BAG, in line with findings from T1 BAG. In terms of interaction effects, we found a negative association between F3 and F5 (family conflict) and fMRI BAG, and a positive association between F6 and fMRI BAG.
We found a positive association between ELA dimension F2, representing factor loadings from measures of socioeconomic disadvantage and neighbourhood safety, and T1 BAG, indicating that this dimension was associated with older-looking brains. Further, we found a negative association between ELA dimensions F3 (secondary caregiver lack of support) and F4 (primary caregiver lack of support) and T1 BAG, indicating that dimensions related to neglect are associated with younger-looking brains. In terms of interaction effects, we found negative associations between F10 (lack of supervision) and T1 BAG, indicating that unsupervised youth diverge more from normative age patterns over time.
In line with the T1 BAG findings, we found a positive association between F2 and DTI BAG. In terms of interaction effects, we found a positive association between ELA dimension F1 (caregiver psychopathology) and DTI BAG.
For fMRI, we found a positive association between F2 and fMRI BAG, aligning with findings with T1 and DTI BAG. Positive associations were also found for F1 and fMRI BAG, aligning with DTI findings, and F6 (caregiver substance abuse and separation from biological parents), F8 (family aggression), and F9 (trauma exposure) and fMRI BAG. These positive associations indicate that dimensions related to threat are related to older-looking brains. A negative association was found between F4 and fMRI BAG, in line with findings from T1 BAG. In terms of interaction effects, we found a negative association between F3 and F5 (family conflict) and fMRI BAG, and a positive association between F6 and fMRI BAG.
·Figure 1. Trimodal brain age prediction. Performance of T1, DTI, and rsfMRI brain age models.
·Figure 2. ELA Dimensions. Factor loadings from the ten early-life adversity dimensions. * indicates that the variable was reverse-scored. Y = Youth Report; CG = Caregiver Report.
Conclusions:
Different ELA features uniquely impact brain outcomes in youth. Deprivation-related features suggest delayed maturation, while threat-related features indicate accelerated maturation.
Lifespan Development:
Early life, Adolescence, Aging 1
Novel Imaging Acquisition Methods:
Anatomical MRI
Diffusion MRI
Multi-Modal Imaging 2
Non-BOLD fMRI
Keywords:
Development
FUNCTIONAL MRI
Machine Learning
MRI
STRUCTURAL MRI
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
Other - Early life adversity (ELA)
1|2Indicates the priority used for review
Provide references using author date format
2. McLaughlin, K. A. & Sheridan, M. A. Beyond Cumulative Risk: A Dimensional Approach to Childhood Adversity. Curr. Dir. Psychol. Sci. 25, 239–245 (2016).
3. Brieant, A. et al. Characterizing the dimensional structure of early-life adversity in the Adolescent Brain Cognitive Development (ABCD) Study. Dev. Cogn. Neurosci. 61, 101256 (2023).
4. Chen, T. & Guestrin, C. XGBoost: A Scalable Tree Boosting System. in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 785–794 (ACM, 2016). doi:10.1145/2939672.2939785.
5. Bürkner, P.-C. brms : An R Package for Bayesian Multilevel Models Using Stan. J. Stat. Softw. 80, (2017).