# Predicting identity dissociation using childhood maltreatment and genetic variation in the stress-response gene FKBP5: a machine learning analysis

## TL;DR

This study explores how childhood maltreatment and genetic variations in FKBP5 may interact to increase the risk of identity dissociation, using machine learning to predict this condition.

## Contribution

The study introduces a machine learning model that identifies gene-environment interactions between FKBP5 and childhood maltreatment in predicting identity dissociation.

## Key findings

- The model achieved fair discrimination with an AUC of 0.709 and 58.6% sensitivity.
- Decision curve analysis showed net clinical benefit for risk stratification across a wide range of threshold probabilities.
- The negative predictive value was high at 0.91, suggesting strong utility in ruling out identity dissociation.

## Abstract

Identity dissociation is challenging to detect and treat, and its etiology remains incompletely understood. Childhood maltreatment and FKBP5 polymorphisms, which modulate the stress response, may contribute by disrupting the integration of autobiographical experiences essential for identity development. We examined whether gene-environment interactions involving childhood maltreatment and FKBP5 polymorphisms predict clinically significant identity dissociation. In a cohort of N = 377 participants, we assessed childhood maltreatment and identity dissociation using validated questionnaires and genotyped CATT haplotypes within FKBP5 linked to stress reactivity. Identity dissociation was dichotomized using an established clinical threshold. An elastic net regularized logistic regression model incorporating maltreatment subtypes, CATT haplotype count, and their interactions was trained (N = 194) and validated (N = 183). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and Matthews correlation coefficient. Decision curve analysis assessed clinical utility across varying risk thresholds. The model demonstrated fair discrimination (AUC = 0.709) with 58.6% sensitivity and 79.9% specificity. While the positive predictive value was modest (35.0%) due to the low prevalence of identity dissociation (15.9%), decision curve analysis revealed a net clinical benefit across a broad range of threshold probabilities (6–76%), indicating practical utility for risk stratification in clinical settings. The negative predictive value was 0.91. These findings provide initial evidence that gene-environment interactions between childhood maltreatment and FKBP5 variation may contribute to the risk of identity dissociation. While predictive precision remains limited, this study demonstrates the feasibility of applying machine learning approaches to dissociation and highlights the need for further research into their traumatic and biological underpinnings to improve detection, prevention, and treatment.

The online version contains supplementary material available at 10.1038/s41598-026-42512-0.