Martino Belvederi Murri1, Luca Cattelani2, Federico Chesani3, Pierpaolo Palumbo4, Federico Triolo5, George S Alexopoulos6. 1. Department of Neuroscience and Rehabilitation, Institute of Psychiatry, University of Ferrara (MBM), Ferrara, Italy. 2. Department of Computer Science and Engineering, University of Bologna (LC, FC), Bologna, Italy; Faculty of Medicine and Health Technologies, Tampere University (LC), Tampere, Finland; Institute of Biomedicine, University of Eastern Finland (LC), Kuopio, Finland. 3. Department of Computer Science and Engineering, University of Bologna (LC, FC), Bologna, Italy. 4. Department of Electrical, Electronic and Information Engineering "Guglielmo Marconi", University of Bologna (PP), Bologna, Italy. 5. Aging Research Center, Department of Neurobiology, Care Sciences and Society, Karolinska Institutet (FT), Stockholm, Sweden. 6. Weill Cornell Institute of Geriatric Psychiatry, Weill Cornell Medicine (GA), White Plains, NY. Electronic address: gsalexop@med.cornell.edu.
Abstract
OBJECTIVE: To develop streamlined Risk Prediction Models (Manto RPMs) for late-life depression. DESIGN: Prospective study. SETTING: The Survey of Health, Ageing and Retirement in Europe (SHARE) study. PARTICIPANTS: Participants were community residing adults aged 55 years or older. MEASUREMENTS: The outcome was presence of depression at a 2-year follow up evaluation. Risk factors were identified after a literature review of longitudinal studies. Separate RPMs were developed in the 29,116 participants who were not depressed at baseline and in the combined sample of 39,439 of non-depressed and depressed subjects. Models derived from the combined sample were used to develop a web-based risk calculator. RESULTS: The authors identified 129 predictors of late-life depression after reviewing 227 studies. In non-depressed participants at baseline, the RPMs based on regression and Least Absolute Shrinkage and Selection Operator (LASSO) penalty (34 and 58 predictors, respectively) and the RPM based on Artificial Neural Networks (124 predictors) had a similar performance (AUC: 0.730-0.743). In the combined depressed and non-depressed participants at baseline, the RPM based on neural networks (35 predictors; AUC: 0.807; 95% CI: 0.80-0.82) and the model based on linear regression and LASSO penalty (32 predictors; AUC: 0.81; 95% CI: 0.79-0.82) had satisfactory accuracy. CONCLUSIONS: The Manto RPMs can identify community-dwelling older individuals at risk for developing depression over 2 years. A web-based calculator based on the streamlined Manto model is freely available at https://manto.unife.it/ for use by individuals, clinicians, and policy makers and may be used to target prevention interventions at the individual and the population levels.
OBJECTIVE: To develop streamlined Risk Prediction Models (Manto RPMs) for late-life depression. DESIGN: Prospective study. SETTING: The Survey of Health, Ageing and Retirement in Europe (SHARE) study. PARTICIPANTS: Participants were community residing adults aged 55 years or older. MEASUREMENTS: The outcome was presence of depression at a 2-year follow up evaluation. Risk factors were identified after a literature review of longitudinal studies. Separate RPMs were developed in the 29,116 participants who were not depressed at baseline and in the combined sample of 39,439 of non-depressed and depressed subjects. Models derived from the combined sample were used to develop a web-based risk calculator. RESULTS: The authors identified 129 predictors of late-life depression after reviewing 227 studies. In non-depressed participants at baseline, the RPMs based on regression and Least Absolute Shrinkage and Selection Operator (LASSO) penalty (34 and 58 predictors, respectively) and the RPM based on Artificial Neural Networks (124 predictors) had a similar performance (AUC: 0.730-0.743). In the combined depressed and non-depressed participants at baseline, the RPM based on neural networks (35 predictors; AUC: 0.807; 95% CI: 0.80-0.82) and the model based on linear regression and LASSO penalty (32 predictors; AUC: 0.81; 95% CI: 0.79-0.82) had satisfactory accuracy. CONCLUSIONS: The Manto RPMs can identify community-dwelling older individuals at risk for developing depression over 2 years. A web-based calculator based on the streamlined Manto model is freely available at https://manto.unife.it/ for use by individuals, clinicians, and policy makers and may be used to target prevention interventions at the individual and the population levels.