Looking into the Future : How to Use Advanced Statistical Methods for Predicting Psychotherapy Outcomes in Routine Care

Abstract: Psychotherapy research has shifted from mainly focusing on the average effects of different treatments to concentrating more on questions related to the individual patient. When research attention shifts, it can give rise to the implementation of new statistical methods that, in turn, can illuminate new challenges that must be addressed.The aim of the thesis was to study how traditional methods for predicting certain psychotherapy outcomes have been conducted in the past, and how more advanced statistical methods might be used to enhance knowledge of how to predict these outcomes today.Three studies were performed: Paper I focused on how Multi Level Modeling (MLM) can be used to study certain aspects of the relationship between working alliance and treatment outcome. In Paper II, Latent Profile Analysis (LPA) and item-level analysis were used to give nuance to the relationship between psychological distress at baseline and change rate during treatment. Finally, in Paper III, Machine Learning (ML) was used to detect dropout patients in the early phase of treatment by exploring complex patterns of symptom distress during the early phase of treatment.The thesis showed how different goals of scientific exploration can be studied in the context of routine care with the use of these statistical frameworks and discussed some of the challenges and opportunities worth noting when entering this line of research.