Although there is scientific evidence of the presence of immunometabolic alterations in major depression, not all patients present them. Recent studies point to the association between an inflammatory phenotype and certain clinical symptoms in patients with depression. The objective of our study was to classify major depression disorder patients using supervised learning algorithms or machine learning, based on immunometabolic and oxidative stress biomarkers and lifestyle habits. Taking into account ...
Although there is scientific evidence of the presence of immunometabolic alterations in major depression, not all patients present them. Recent studies point to the association between an inflammatory phenotype and certain clinical symptoms in patients with depression. The objective of our study was to classify major depression disorder patients using supervised learning algorithms or machine learning, based on immunometabolic and oxidative stress biomarkers and lifestyle habits. Taking into account a series of inflammatory and oxidative stress biomarkers (C-reactive protein (CRP), tumor necrosis factor (TNF), 4-hydroxynonenal (HNE) and glutathione), metabolic risk markers (blood pressure, waist circumference and glucose, triglyceride and cholesterol levels) and lifestyle habits of the participants (physical activity, smoking and alcohol consumption), a study was carried out using machine learning in a sample of 171 participants, 91 patients with depression (71.42% women, mean age = 50.64) and 80 healthy subjects (67.50% women, mean age = 49.12). The algorithm used was the support vector machine, performing cross validation, by which the subdivision of the sample in training (70%) and test (30%) was carried out in order to estimate the precision of the model. The prediction of belonging to the patient group (MDD patients versus control subjects), melancholic type (melancholic versus non-melancholic patients) or resistant depression group (treatment-resistant versus non-treatment-resistant) was based on the importance of each of the immunometabolic and lifestyle variables. With the application of the algorithm, controls versus patients, such as patients with melancholic symptoms versus non-melancholic symptoms, and resistant versus non-resistant symptoms in the test phase were optimally classified. The variables that showed greater importance, according to the results of the area under the ROC curve, for the discrimination between healthy subjects and patients with depression were current alcohol consumption (AUC = 0.62), TNF-α levels (AUC = 0.61), glutathione redox status (AUC = 0.60) and the performance of both moderate (AUC = 0.59) and vigorous physical exercise (AUC = 0.58). On the other hand, the most important variables for classifying melancholic patients in relation to lifestyle habits were past (AUC = 0.65) and current (AUC = 0.60) tobacco habit, as well as walking routinely (AUC = 0.59) and in relation to immunometabolic markers were the levels of CRP (AUC = 0.62) and glucose (AUC = 0.58). In the analysis of the importance of the variables for the classification of treatment-resistant patients versus non-resistant patients, the systolic blood pressure (SBP) variable was shown to be the most relevant (AUC = 0.67). Other immunometabolic variables were also among the most important such as TNF-α (AUC = 0.65) and waist circumference (AUC = 0.64). In this case, sex (AUC = 0.59) was also relevant along with alcohol (AUC = 0.58) and tobacco (AUC = 0.56) consumption. The results obtained in our study show that it is possible to predict the diagnosis of depression and its clinical typology from immunometabolic markers and lifestyle habits, using machine learning techniques. The use of this type of methodology could facilitate the identification of patients at risk of presenting depression and could be very useful for managing clinical heterogeneity.
+
