The project goes beyond the state of the art in several ways, with expected results tied to each.
First, regarding establishing whether someone is depressed, the current state of the art in the field is to either conduct one clinical interview per person, or one questionnaire that queries depression symptoms. Regarding interviews (that are not feasible for WARN-D in the first place), inter-rater reliability for depression has consistently been shown to be among the lowest of all mental disorders, meaning that clinicians more often disagree than agree with each other whether a person should be assigned a diagnosis of depression. Regarding the questionnaires that are being used, they are short and only contain a very limited amount of information about how a person is doing, and are usually given only once. In WARN-D, we have therefore collected multidimensional data from various sources, all of which have shown to play a role in deciding whether someone should be considered depressed. Our battery not only integrates wellbeing, impairment, depression and anxiety symptoms, as well as questions on recently obtained diagnoses, therapy status or taking medication, but also information on whether increases in problems could be expected based on e.g. questionnaires on severe or adverse life events. Further, we have continuous data (compared to the one-time assessments usually done), meaning we can utilize data collected every single day to make a more informed decision whether we should consider someone to be depressed. We are currently working on a pipeline to make the optimal decision based on all data sources, which goes considerably beyond the state of the art and will lead to important output regarding the conceptualization, operationalization and measurement of depression.
Second, there is growing agreement that mental health problems like depression are biopsychosocial systems out of which mental illness emerges. However, what this system is and how it operates is largely unknown. One of the PhD projects will map out this mood system in detail, utilizing the enormous amount of relevant data with high temporal resolution from different data sources we have collected in the project. The statistical focus of this project is on utilizing statistical network models that can integrate various data sources into one system, considerably extending the literature on the human mood system.
A third milestone, based on the perspective that mental health problems operate as systems, is to see whether depression onset can be anticipated using early warning signals (EWS). Other fields such as ecology have shown that when e.g. lakes tip from a healthy to a disease state, this transition can be anticipated using statistical features of the system that can be measured and modeled. EWS thus can serve as signals acquired from data that may indicate a potential transition to an alternative system state, such as developing depression, and may serve as a point for preemptive interventions. One PhD project is currently concerned with the question of what particular EWS can be identified in our data, and how well they do in prediction depression onset. As part of this effort, we are testing existing EWS, but also testing novel EWS we developed specifically for psychological questions in the team.
Finally, these findings will be integrated together into a prospective prediction model, which will use state of the art machine learning to create models that can forecast depression onset. These models will be both data-driven, utilizing all information we collect on participants in the project, but also theory-driven, leveraging the impressive EWS literature in other areas where many EWS have been established successfully. We hope to utilize these models to develop a real-time app that can integrate wearables and daily diaries to deliver smartphone-based warning signals.
