Artificial intelligence to help identify depression

Depression is a common disease that affects approximately 4% of the world population, being one of the main causes of disability. Despite the fact that there are effective treatments, a high percentage of patients can suffer recurrent episodes and many of them are resistant to treatment.

The heterogeneity of depression symptoms makes it difficult to identify the pathophysiological mechanisms of this disease. A promising conceptual framework for understanding its pathophysiological mechanisms is the study of inflammatory dysregulation, recent theories pointing to an association between immunometabolic alterations and specific subtypes of depression.

Taking this knowledge base into account, and using different machine learning algorithms (a modality of artificial intelligence), a team led from the Autonomous University of Madrid (UAM) in Spain carried out a study on the classification of patients with major depressive disorder from of immunometabolic variables (for example C-reactive protein, tumor necrosis factor, high-density lipoprotein cholesterol, triglycerides, blood sugar level, blood pressure, and waist measurement) and oxidative stress (lipid peroxidation and glutathione levels ), as well as variables related to lifestyle (for example, smoking habits, alcohol consumption and physical exercise).

Machine learning methods based on making explicit assumptions about patient subtypes and the subsequent adjustment of the data to these assumptions, allow the development of models that can contribute to creating more homogeneous groups of patients. In addition, it does not require prior knowledge of the possible relationships between variables, which is why it is being used more and more in the investigation of mental disorders.

In this study, a total of 171 participants were evaluated, of which 91 were patients with depression and 80 healthy people. Taking into account the aforementioned variables, it was possible to optimally classify patients versus healthy ones and patients according to their symptoms and their response to treatment. In addition, it was possible to analyze the relative importance of each of the variables in the classification.

Artificial intelligence’s help in identifying depression could prove decisive in predicting early enough when someone is planning suicide. (Illustration: Amazings/NCYT)

“These results confirm the importance of inflammatory and metabolic variables in depression. The inflammatory alterations presented by patients with major depressive disorder could be a consequence of hyperactivation of the hypothalamic-pituitary-adrenal axis caused by chronic stress, impacting on the production of proinflammatory proteins and oxidative agents through deregulation of glucocorticoid production. “, details Dr. Pilar López García, co-author of the study.

Therefore, according to the authors of the study, the increase in inflammatory and oxidative stress proteins such as glutathione, tumor necrosis factor or C-reactive protein in patients with major depressive disorder may serve to differentiate between healthy subjects and patients with depressive disorder”.

On the other hand, regarding the metabolic state, the risk of depression is approximately 50% in the presence of abdominal obesity. Abdominal adiposity is characterized by the accumulation of visceral fat, more related to metabolic deregulations and with a greater effect on inflammation.

Regarding lifestyle, the study has shown that both alcohol consumption and physical exercise are important when classifying subjects with a diagnosis of depression, and are also determinant in classifying depressive subtypes.

These results have important implications in clinical settings, as the identification of unhealthy lifestyles and immunometabolic disorders may guide and assist in the management of clinical depression.

Finally, the work demonstrates the good potential of using machine learning algorithms to address depression and, in particular, what kind of algorithms produce the best performance in doing so.

“It is important to highlight that the possible improvement of these techniques can help mental health professionals in the future to redefine mental disorders objectively, being able to identify patients and their prognosis based on risk factors determined as variables. predictive and, in turn, personalize the treatments according to the patients”, concludes Yolanda Sánchez-Carro, co-author of the study.

The work was carried out within the MARIDE project (Study of Inflammatory MARcators in DEpression), in which the UAM and Hospital Universitario de la Princesa in Madrid and Hospital del Mar and Hospital de Bellvitge in Barcelona collaborated.

The study is titled “Importance of immunometabolic markers for the classification of patients with major depressive disorder using machine learning. And it has been published in the academic journal Progress in Neuro-Psychopharmacology and Biological Psychiatry. (Source: UAM)