DESCRIPTION (provided by applicant): Anxiety and mood disorders (internalizing disorders [IDs]) are common and debilitating conditions. The proposed Career Development Award will provide the candidate with the necessary skills to develop an independent research program focused on identifying how psychological/personality, environmental, and biomarker vulnerabilities predict the development and course of IDs using prospective observational survey designs. The applicant's prior research training involved studying the influence of psychological/personality vulnerabilities on ID severity and course in patient samples. In order to expand his research program, the applicant is seeking advanced training in two key areas (psychiatric epidemiology; machine learning methods) and supplemental training in two additional areas (self-assessed biomarkers; environmental vulnerabilities). Collectively, this training will provide the applicant with the skills (1) to study ID onset and course using prospective survey designs, and (2) to use machine learning methods to develop clinically-useful risk algorithms of ID onset and course using multiple vulnerability domains (psychological/ personality; environmental; biomarker). These skills will be developed through a combination of didactic training, guided readings, and mentored research projects. Training will occur in the Department of Health Care Policy at Harvard Medical School under the mentoring of Dr. Ronald Kessler (an expert in psychiatric epidemiology) and Dr. Sherri Rose (an expert in machine learning). The proposed research plan involves two phases. In Phase 1, machine learning methods will be used to analyze cross-sectional ID data from the World Mental Health (WMH) Surveys in order to develop risk algorithms that predict the onset of major depression, bipolar disorder, and generalized anxiety disorder (Specific Aim 1). Machine learning methods will also be applied to WMH Survey data to develop subtypes of bipolar disorder, generalized anxiety disorder, and posttraumatic stress disorder that maximize the prediction of their long-term course (Specific Aim 2). The proposed Phase 1 studies are highly innovative; in contrast to other areas of medicine, virtually no studies have attempted to develop ID risk algorithms or subtypes using machine learning methods. However, Phase 1 studies are also preliminary; replication and expansion of the algorithms is needed in prospective samples. Accordingly, Phase 2 of the research plan (Specific Aim 3) aims to test the feasibility of conducting a large prospective web-based survey study of psychological/personality (assessed via self-report and behavior), environmental (assessed via self-report), and biomarker (assessed via self-administered salivary assays) vulnerabilities of ID onset and course. Participants will be recruited online to complete a baseline survey and those determined to have an ID or be at high-risk of developing an ID will be asked to complete bimonthly follow-ups over the course of one year. The Phase 2 study is also innovative. Despite being called "the future" of epidemiological research, web-based survey studies of the IDs are rare. In addition, existing prospective epidemiological studies coarsely (i.e., dichotomously) assess IDs and their vulnerabilities, and have long gaps (1-4 years) between follow-ups. The use of bimonthly online surveys of the severity of IDs and their vulnerabilities will address these limitations. Phase 2 exploratory analyses will also be conduct in which latent ID trajectories are predicted using machine learning methods. Overall, the outlined training activities and mentored research projects will be used to develop an R01 application for a prospective epidemiological study that uses machine learning methods to develop clinically useful algorithms of ID onset and course using a broad range of psychological/personality, environmental, and biomarker vulnerabilities factors.

PUBLIC HEALTH RELEVANCE: The proposed research aims to use cutting-edge statistical methods to develop clinically-useful (1) risk algorithms to identify who is at greatest risk of developing different internalizing disorders, and (2) subtypes of internalizing disorders that identify who is most likely to experience a persistence and chronic long-term disorder course. This research has important clinical and public health implications. With this research, individuals at risk of developing internalizing disorders can be targeted using early detection and prevention efforts, while those at risk of experiencing a persistent and chronic long-term course can be targeted using tailored/intensive internalizing disorder interventions.