Abstract

Artificial intelligence (AI)-based techniques have been widely applied in depression research and treatment. Nevertheless, no specific predictor model for depression has been developed yet in Chile using specific Chilean characteristics (variables). The present study used data from 11525 participants of the National Health Survey (NHS) to create a model to predict risk of depression (PDRM). This model was contrasted with data from 280 outpatients diagnosed with depression. To develop the PDRM we employed classification algorithms models and fusion of data about depression from two waves of the NHS (2009-2010 and 2016-17). Validation of the model of 19 variables (questions) was done applying machine learning algorithms. Based on 2009-10 data, Recall for Naive Bayes (NB) yielded 0.92, LOGIT was 0.86 and SVM=0.84. After setting up the PDRM this predictor was contrasted with the data from patients (P) diagnosed Bipolar Disorder (140 P); Major Depressive Disorder (MDD, 140 P); and Adjustment Disorder (80 P, control). Fusion of patient's data from anamnesis and consultations was used to determine the presence or absence of each 16 variables per patient. Prediction of depression using the PDRM model detected 122 cases of depression out of the total 140 depressive cases, with a mean area under the receiver operating characteristic curve of 0.8 and a recall 0.74 (with NB). This predictive model may contribute to decision-making processes in a fast, simple and economical way.