Fast identification of depressive symptomatology using probabilistic machine learning
Depression is often overlooked and this tends to lead to the provision of sub-optimal symptom treatment or a lack of treatment altogether. This can generate an irreversible social and economic impact on society. Therefore, there is an urgent necessity to elaborate faster, scalable, and low-cost depression identification tools.
To identify people with depressive symptomatology (DS) in a faster way, we aim to apply 3 approaches using probabilistic Machine learning techniques:
Using socio-demographic and health individual data to develop a tool to target the most likely people with DS. This tool will enable the specialists to save time while screening people to detect depressive symptoms, as it will provide a ranking list with the most probable people at the top. The tool will also evaluate the benefits of using the machine learning model, through the trade-off between how many people with DS they expect to get (sensitivity rate), versus how much screening time is reduced.
To detect people with DS, often the Patient Health Questionnaire with 9 questions (PHQ-9) is applied through a screening process. Usually, this questionnaire follows an order from 1 to 9. The aim here is to look for different orders and a minimum number of questions to be asked. Therefore, reducing the time and effort of the specialists.
There are other questionnaires used to detect DS that are shorter than the PHQ-9. For this approach we aim to apply the methodology in 2) to different questionnaires to analyse which one gives the faster response but still detects DS accurately.