The development of psychiatric disorders and adverse behaviors : from context to prediction

Abstract

Psychiatric disorders by definition cause significant impairment in an individual’s daily functioning. Certain disorders, such as borderline personality disorder (BPD) and eating disorders, have worse prognosis and high mortality rates compared to other psychiatric disorders. Similarly, adverse behaviors such as self-harm, suicide, and crime are often present in individuals with psychiatric disorders. It is of interest to further understand the etiology and associations of BPD and eating disorders to uncover potential avenues and opportunities for intervention. Moreover, prediction modeling has recently come of interest to psychiatric epidemiologists with the rise of large data sets. Prediction modeling may provide valuable information about the nature of risk factors and eventually aid clinical diagnostics and prognostics. Thus, the studies included in this thesis seek to examine the etiology, associations, and prediction approaches of psychiatric disorders and adverse behaviors.

Study I examined the individual and familial association between type 1 diabetes (T1D) and eating disorder diagnoses. We used national health care records from Denmark (n = 1,825,920) and Sweden (n = 2,517,277) to calculate the association within individuals, full siblings, half siblings, full cousins, and half cousins. Individuals with T1D had twice the hazard rate ratio of being diagnosed with an eating disorder compared to the general population. There was conflicting evidence for the risk of an eating disorder in full siblings of T1D patients. However, there was no evidence to support a further familial relationship between the two conditions.

Study II aimed to illuminate the nature of the correlates for BPD across time, sex, and for their full siblings. We examined 87 variables across psychiatric disorders, somatic illnesses, trauma, and adverse behaviors (such as self-harm). In a sample of 1,969,839 Swedes with 12,175 individuals diagnosed with BPD, we found that BPD was associated with nearly all of the examined variables. The associations were largely consistent across time and between the sexes. Finally, we found that having a sibling diagnosed with BPD was associated with psychiatric disorders, trauma, and adverse behaviors but not somatic illnesses.

Study III created a prediction model that could predict who would have high or low psychiatric symptoms at age 15 based on data from parental reports and national health care registers collected at age 9 or 12. Additionally, we compared multiple types of machine learning algorithms to assess predictive performance. The sample included 7,638 twins from the Child and Adolescent Twin Study in Sweden (CATSS). Our model was able to predict the outcome with reasonable performance but is not suitable for use in clinics. Each model performed similarly indicating that researchers with similar data and research questions do not need to forgo standard logistic regression.

Study IV aimed to determine if an individual will exhibit suicidal behaviour (self-harm or suicidal thoughts), aggressive behaviour, both, or neither before adulthood with prediction modeling. Through variable importance scores we examined the usefulness of genetic variables within the model. A total of 5,974 participants from CATSS and 2,702 participants from the Netherlands Twin Register (NTR) were included in the study. The model had adequate performance in both the CATSS and NTR datasets for all classes except for the suicidal behaviors class in the NTR, which did not perform better than chance. The included genetic data had higher variable importance scores than questionnaire data completed at age 9 or 12, indicating that genetic biomarkers can be useful when combined with other data types.

In conclusion, the development of psychiatric disorders and symptoms are associated with many factors across somatic illnesses, other psychiatric disorders, trauma, and harmful behaviors. The results of this thesis demonstrates the limitations of prediction modeling in psychiatric clinics but highlights their use in research and on the path forward towards personalized medicine.