Predicting rehospitalization within 2 years of initial patient admission for a major depressive episode: a multimodal machine learning approach

Autor: Henning Teismann, Walter Heindel, Theresa Winter, Klaus Berger, Udo Dannlowski, Anbupalam Thalamuthu, Bernhard T. Baune, Scott R. Clark, Micah Cearns, Claas Kähler, Heike Minnerup, Nils Opel
Rok vydání: 2019
Předmět:
Male
Declaration
Logistic regression
computer.software_genre
Helsinki declaration
Machine Learning
Prognostic markers
Major depressive episode
education.field_of_study
Brain
Middle Aged
Magnetic Resonance Imaging
Antidepressive Agents
Psychiatry and Mental health
Treatment Outcome
Area Under Curve
Predictive value of tests
Biomarker (medicine)
Major depressive disorder
Female
medicine.symptom
Adult
medicine.medical_specialty
Population
Machine learning
Patient Readmission
Clinical decision support system
Article
lcsh:RC321-571
Cellular and Molecular Neuroscience
Neuroimaging
Predictive Value of Tests
Human behaviour
medicine
Humans
education
lcsh:Neurosciences. Biological psychiatry. Neuropsychiatry
Biological Psychiatry
Aged
Depressive Disorder
Major
Modalities
business.industry
medicine.disease
Clinical research
Artificial intelligence
Outcomes research
business
computer
Biomarkers
Follow-Up Studies
Zdroj: Translational Psychiatry, Vol 9, Iss 1, Pp 1-9 (2019)
Translational Psychiatry
ISSN: 2158-3188
Popis: Background: Currently, clinicians have no validated decision support systems to predict rehospitalization risk in major depressive disorder upon initial patient hospital admission. Methods: The BiDirect study is an ongoing 12-year study, integrating three cohorts, patients with depression, population controls and patients with an acute coronary event. 380 patients aged ≥ 35 and < 66 years, with a current or recent hospitalisation for a major depressive episode (Rehospitalized: yes = 102, no = 278) who completed structural imaging and genetic testing were included in the study. Machine learning models trained with baseline data from July 2010 to June 2013 were used to predict rehospitalization within two years of an initial inpatient episode. Predictor modalities used included clinical, blood-biomarker, structural imaging, genetic (polygenic risk scores), and electrocardiography predictors. Findings: Optimal performance was achieved with a multimodal panel containing structural imaging, blood-biomarker, clinical, medication type and sleep quality predictors, attaining a test AUC of 67.74 (p = 9.99-05) in an outer 10-fold cross validation loop. After controlling for covariates in a non-penalized logistic regression model, baseline right hippocampal volume and the number of previous inpatient admissions were significantly associated with rehospitalization. Interpretation: In this analysis, our multimodal model outperformed models based on clinical variables alone, combined biomarkers and individual data modality prognostication for rehospitalization. This finding points to the potential of predictive models that combine multimodal clinical and biomarker data in the development of clinical decision support systems. Funding: The BiDirect Study is funded by the German Federal Ministry of Education and Research (BMBF, (01ER0816 and 01ER1506, and 01ER1205). Additional funding for the analysis was provided by the German Research Foundation (DFG, grant FOR2107 DA1151/5-1 and DA1151/5-2 to UD; SB-TRR58, Projects C09 and Z02), the Interdisciplinary Center for Clinical Research (IZKF) of the medical faculty of Munster (grant Dan3/012/17 to UD, SEED 11/18) and the Deanery of the Medical Faculty of the University of Munster. In addition, this work was supported with supercomputing resources provided by the Phoenix HPC service at the University of Adelaide. Declaration of Interest: None. Ethical Approval: The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Databáze: OpenAIRE
Externí odkaz: