# Machine learning for identifying caregiving adversities associated with greatest risk for mental health problems in children ***Authors***: Anna Vannucci, Andrea Fields, Charlotte Heleniak, Paul A. Bloom, Chelsea Harmon, Aki Nikolaidis, Ian J. Douglas, Lisa Gibson, Nicholas L. Camacho, Tricia Choy, Syntia S. Hadis, Mary Dozier, Michael P. Milham* & Nim Tottenham* (*These authors contributed equally to this work) ## Contents of repository This repository contains supplementary material, scripts, model outputs, and plots for this manuscript. There are 2 primary analysis components within each folder: 1. `Confirmatory factor analysis`: extract participant p-factor scores 2. `Conditional random forest`: examine the importance of crEAs for obtaining replicable, generalizable links to risk for psychopathology symptoms The file naming structure is: `sample_ivs_dv` - **Sample**: - `full` full sample (*N* = 306) of children with and without a history of caregiving-related early adversity (crEA) - `creas` subsample (*N* = 201) of children exposed to caregiving-related early adversity (crEA) - **IVs (independent variables - caregiving experiences)**: - `predcreayn` dichotomous presence or absence of exposure to caregiving-related early adversities (crEAs); *(examined in full sample only)* - `predcrea`distinct crEA experiences, including caregiving switches, domestic violence exposure, emotional abuse, emotional neglect, physical abuse, physical neglect, sexual abuse, supervisory neglect - `predall`distinct crEA experiences + concurrent caregiving strategies (consistent, monitoring, positive attention) - `predwindow`: total duration (age of latest crEA occurrence - age of crEA onset in months), infancy (0-17mo), toddlerhood (18-35mo), preschool-age (36-71mo), early school-age (72-95mo), late school-age (96-155mo); *(examined in crEA-exposed subsample only)* - **DVs (dependent variables - mental health)** - `pfactor` p-factor score - `tot` CBCL total behavior problems T-score - `int` CBCL internalizing problems T-score (aka internalizing behaviors) - `ext` CBCL externalizing problems T-score (aka externalizing behaviors) - `in` CBCL attention problems T-score (aka inattentive behaviors) - **Sociodemographic confounds** - `age` in years (6-12 years-old) - `sex` parent-reported sex at birth (female, male) - `race` American Indian or Alaska Native, Asian American, Black or African American, White, Multiple Races, Other/Unknown Race - `income-to-needs ratio` index of socioeconomic status that is calculated by dividing a family's annual household income by the federal guideline for poverty by family size in the year of data collection ## Data availability statement Given the sensitivity of the early adversity information, raw sociodemographic and experiential data cannot be shared publicly to protect the privacy of the participants. The data are shared on the NIMH Data Archive Repository (NDAR; Data Collection ID = 2803; https://nda.nih.gov/). ## Abstract Developmental and experiential heterogeneity associated with caregiving-related early adversities (crEAs) poses a major challenge to identifying replicable, generalizable findings. Here conditional random forests evaluated the importance of unique crEA experiences for estimating risks to mental health in 306, 6-12 year-old children with heterogeneous crEA experiences (different forms of caregiver-involved abuse and/or neglect, permanent/significant parent-child separation). The better that crEAs improve the accuracy of symptom estimates in held-out, never-before-seen children, the more important/generalizable they were considered. We show that earlier timing and longer duration of crEAs is especially important for elevated general psychopathology (p-factor scores). The mere presence (vs. absence) of crEAs is more valuable for estimating symptom risk than were specific adversities in a broad sample. Specific adversities become more important when only looking within the crEA-exposed subsample, with adversities of an interpersonal-affective nature being the most likely to increase transdiagnostic symptom risk. Concurrent consistent caregiving also have high importance, motivating consideration of later-occurring environmental experiences in future studies of early adversity. ## Acknowledgments We express our deep appreciation to the families participating in this study. The authors gratefully acknowledge current and former lab members who assisted with data collection and coding. This work was supported by the *National Institute of Mental Health*: 5R01MH091864 (co-PIs: Tottenham, Milham) and the *NIH Blueprint and BRAIN Initiative Diversity Specialized Predoctoral to Postdoctoral Advancement in Neuroscience* (D-SPAN) Award: F99NS134207 (PI: Vannucci).