A predictive model developed with machine learning techniques that included a small number of clinical variables was effective in distinguishing patients with vs without depression as well as between cases of incident and chronic depression. This is according to an analysis of the Brazilian Longitudinal Study of Adult Health (ELSA-Brazil) study published in Psychological Medicine.
In this analysis of the ELSA-Brazil study, which enrolled civil servants from 6 public institutions across major cities in Brazil, researchers examined data between 2008 and 2010 (n=15,105) and between 2012 and 2014 (n=13,922).
A predictive model for depression cases, incidence, and chronicity in the occupational cohort was created and included variables such as sociodemographic variables, obesity, smoking status, mental disorders, psychotropic use, and negative life events.
In the machine learning analysis, the investigators initiated an elastic net regularization analysis with a 10-fold cross-validation procedure with the socioeconomic and clinical variables as predictors to differentiate patients with depression vs without depression, patients with incident depression vs those who did not develop depression, and patients with chronic (ie, persistent or recurrent) depression vs patients without depression at follow up.
A total of 1085 patients presented with a history of depression at both the baseline and follow-up time points, whereas 12,387 patients had not experienced any depressive episode. In this analysis classifying patients with vs without depression, the elastic net model featured an area under the curve (AUC) of 0.79 (95% CI, 0.76–0.82). A total of 499 patients reported a new depressive episode at follow up.
For the prediction of incident depression, the model produced an AUC of 0.71 (95% CI, 0.66–0.77), with past or present history of smoking and educational level discarded by the model. In the model distinguishing between chronic depression and patients without depression, the AUC was 0.90 (95% CI, 0.86–0.95), with self-reported race discarded by the model.
A limitation of the study was the inclusion of only civil servant workers, which limited the ability to include unemployment, a risk factor for depression, as a predictor.
Knowing which patients may experience “a depressive episode, and within these, which will have a more chronic and debilitating course, could help improve how we assess patients in clinical settings, shifting our focus from treating acute episodes to preventing them.”
Librenza-Garcia D, Passos IC, Feiten JG, et al. Prediction of depression cases, incidence, and chronicity in a large occupational cohort using machine learning techniques: an analysis of the ELSA-Brasil study [published online June 4, 2020]. Psychol Med. doi: 10.1017/S0033291720001579