Simplicity and equity in depression risk prediction among Brazilian adults: a temporal machine learning study
1
Health and Kinesiology, University of Illinois Urbana-Champaign, Champaign, United States
2
School of Public Health, University of Sao Paulo, Sao Paulo, Brazil
Popul. Med. 2026;8(Supplement Supplement 1):
ABSTRACT
INTRODUCTION:
Public health systems need depression risk predictions that are simple to deploy, stable over time, and equitable across populations. However, many existing models fail to achieve these goals simultaneously, limiting their usefulness for action at scale.
METHODS:
We aimed to develop, simplify, and evaluate the equity of a depression risk model for Brazilian adults using easy-to-collect, self-reported health data. The 2013 Brazilian National Health Survey (Pesquisa Nacional de Saúde, N = 60195) were used for model development, and the 2019 cohort (N = 88525) for temporal validation. Nine machine learning algorithms were implemented, including logistic regression, random forest, and CatBoost. To prioritize simplicity, each algorithm was evaluated under three feature conditions: the full set (85 variables), the Boruta reduced set (31 variables), and a principal component set (69 components). Equity performance was assessed across 19 sociodemographic strata including age, sex and income.
RESULTS:
The Boruta reduced model achieved comparable discrimination to the full feature model, with an area under the curve of 0.83 (95% CI 0.82 to 0.83) while reducing the input features by 64%. Key predictors included chronic diseases, sociodemographic, and lifestyle behaviors. In equity analyses of the full model, discrimination was generally stable across strata, although error rates varied, with higher sensitivity in males (0.81) and older adults (0.83) compared with females (0.63) and younger adults (0.75). Boruta reduced models preserved equity patterns similar to those of the full model. Temporal calibration remained stable across models.
CONCLUSIONS:
A simple model using 31 easy-to-collect variables in public health initiatives maintains accuracy and equity under temporal shift. These results support the use of this machine learning model for an equitable depression detection in the Brazilian health system. The findings suggest that our model can move forward with a prospective effectiveness evaluation, intending future implementation with continuous equity monitoring.
Public health systems need depression risk predictions that are simple to deploy, stable over time, and equitable across populations. However, many existing models fail to achieve these goals simultaneously, limiting their usefulness for action at scale.
METHODS:
We aimed to develop, simplify, and evaluate the equity of a depression risk model for Brazilian adults using easy-to-collect, self-reported health data. The 2013 Brazilian National Health Survey (Pesquisa Nacional de Saúde, N = 60195) were used for model development, and the 2019 cohort (N = 88525) for temporal validation. Nine machine learning algorithms were implemented, including logistic regression, random forest, and CatBoost. To prioritize simplicity, each algorithm was evaluated under three feature conditions: the full set (85 variables), the Boruta reduced set (31 variables), and a principal component set (69 components). Equity performance was assessed across 19 sociodemographic strata including age, sex and income.
RESULTS:
The Boruta reduced model achieved comparable discrimination to the full feature model, with an area under the curve of 0.83 (95% CI 0.82 to 0.83) while reducing the input features by 64%. Key predictors included chronic diseases, sociodemographic, and lifestyle behaviors. In equity analyses of the full model, discrimination was generally stable across strata, although error rates varied, with higher sensitivity in males (0.81) and older adults (0.83) compared with females (0.63) and younger adults (0.75). Boruta reduced models preserved equity patterns similar to those of the full model. Temporal calibration remained stable across models.
CONCLUSIONS:
A simple model using 31 easy-to-collect variables in public health initiatives maintains accuracy and equity under temporal shift. These results support the use of this machine learning model for an equitable depression detection in the Brazilian health system. The findings suggest that our model can move forward with a prospective effectiveness evaluation, intending future implementation with continuous equity monitoring.
| eISSN: | 2654-1459 |
We process personal data collected when visiting the website. The function of obtaining information about users and their behavior is carried out by voluntarily entered information in forms and saving cookies in end devices. Data, including cookies, are used to provide services, improve the user experience and to analyze the traffic in accordance with the Privacy policy. Data are also collected and processed by Google Analytics tool (more).
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
