Comprehensive Summary
This research explores using personality testing scores to identify victims of childhood abuse with machine learning models. This study includes a sample of 733 adolescents ages 11-17 receiving treatment in an inpatient psychiatric facility who had self-reported a history of childhood trauma. De-identified data was taken from these adolescents’ medical records, and consent was given to use this information in research upon admission. Self-reports of trauma were measured using the Childhood Trauma Questionnaire (CTQ), while personality traits and psychopathology were measured using the Minnesota Multiphasic Personality Inventory, Adolescent Version (MMPI-A) Statistical analyses were done to look at correlates between MMPI-A scales and abuse subtypes (emotional, physical, or sexual abuse). Using this data, the research team built machine learning models to potentially identify abuse type experienced based on MMPI-A traits. This paper found that each type of abuse has distinct MMPI-A scale correlates, meaning that victims of specific types of childhood abuse tend to develop specific personality traits that can be measured by this scale. Further, the research team built a variety of machine learning models to predict whether a patient has experienced abuse based on personality traits and demographics. The models tested include Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), Gaussian Naive Bayes (GNB), and Multi-layer Perceptron. The models work by looking at which MMPI-A scales are clinically elevated and determining which type of abuse is most likely based on those elevations. SVM and RF were found to have the best specificity and sensitivity. The article compares the prediction accuracy of these models to that of rapid urine drug screenings. This research finds that there are “abuse fingerprints” in personality data, elevations in clinical scales that are specific to types of abuse experienced. AI models can learn to detect these “fingerprints,” which can provide information about individuals who may have a high probability of having experienced a certain type of abuse.
Outcomes and Implications
It is difficult to identify victims of childhood trauma and/or abuse, often due to the fact that these topics may be difficult to talk about and self-reports about these issues may be distorted. This technology can be utilized to provide clinicians information about the probability of an individual having a history of abuse without asking direct questions about childhood trauma that may be difficult for patients to answer, allowing for potentially more accurate diagnosis. Additionally, this detection technology can help to make the diagnostic process more efficient, which may be necessary when providers need to make quick clinical decisions in settings such as in the ED. In the future, such technologies may be used to link those identified as high-risk to evidence-based interventions that have been shown to prevent or treat mental health conditions in those who have experienced trauma. However, the findings in this paper will need to be replicated before this technology can be used for real-world clinical applications.