Use and Performance of Language-based Artificial Intelligence (AI) Models to Screen for Depressive Disorders.

purpose of reviewThis review summarised and assessed the current evidence on the use of language-based Artificial Intelligence (AI) models for the screening of depressive disorders in adults. RECENT

findingsMost of the studies assessing the use of language-based AI models for the screening of depression were conducted in high-income countries, primarily in the US, and used heterogeneous datasets and populations, limiting generalizability. Most of the evidence was based on text data analysed using transformer models, followed by speech data through convolutional or recurrent neural networks (CNN and RNN models respectively). The evidence combining both modalities (text and speech) and models is limited. Considering the performance for the screening of depressive. disorders, the models tested achieved a performance comparable to standard instruments and cut-off scores, such as the 9-item version of the Patient Health Questionnaire (PHQ-9) with a cut-off of 10 or higher. A better understanding of the performance of language-based AI models in real-world settings is required. However, the evidence identified shows that they can be a relevant resource for this screening of depressive disorders and, consequently, for preventing them and reduce their prevalence, burden, and impact at all levels.