Depressive symptoms of the PHQ-9 questionnaire associated with suicidal ideation using machine learning algorithms in the peruvian population

Abstract

Introduction: suicidal ideation is the thought of self-elimination that is not always reported by patients tested for depression. The objective was to identify and analyze depressive symptoms from the Patient Health Questionnaire-9 associated with suicidal ideation in the Peruvian population. Material and methods: observational, analytical and cross-sectional study based on data from 32,062 participants of the national family health survey using the patient health questionnaire-9. The Chi- square test, Poisson regression with robust variance, multilayer perceptron and decision tree were used. Results: in women, the decision tree algorithm correctly classified 91.10 % of cases of suicidal ideation. In men, it was 94.70 %. Using multilayer perceptron, in women, the percentage of incorrect predictions was 8.90 %. The variables being included: feeling bad, feeling depressed, speaking or moving slowly, problems concentrating and sleeping problems. In men it was 8.12 %, including the variables: feeling bad, feeling depressed, speaking or moving slowly, sleep problems and little or a lot of appetite. Conclusions: supervised learning algorithms are viable and efficient to identify depressive symptoms from the Health Questionnaire-9 associated with suicidal ideation in the Peruvian population, with somatic symptoms predominating in women and affective-cognitive symptoms in men. The use of supervised learning algorithms can be a complement for mental health professionals.