Predictive models of mental health before, during and after COVID-19 pandemic trough machine learning (ML)

• Grant number: 20/11214-5
• Support type: Scholarships in Brazil - Post-Doctorate
• Effective date (Start): October 01, 2020
• Effective date (End): May 31, 2022
• Field of knowledge: Humanities - Psychology - Psychological Treatment and Prevention
• Principal researcher: Isabela Judith Martins Bensenor
• Grantee: Marina Lopes Moreno
• Home Institution: Hospital Universitário (HU). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated research grant: 20/05441-9 - Mental health impact on the novel Coronavirus (COVID-19) in São Paulo State participants of the Brazilian longitudinal health study (ELSA-Brasil), AP.R

Abstract

Since the end of 2019, the world has witnessed the spread of the new coronavirus COVID-19 or Sars-CoV-2. In previous pandemic situations that counted with quarantine periods such as SARS, Ebola, H1N1, MERS and equine influenza, the prevalence of symptoms such as acute stress, post-traumatic stress disorder (PTSD), anxiety, depression, insomnia, increased substance use and the presence of feelings of confusion, anger, frustration and fear of contagion increased [1]. Side effects of the pandemic, such as economic crisis, social isolation, decrease in social support groups and decreased access to psychological assistance, may increase mental health impacts such as the increase in suicide rates [2]. Thus, it is expected that there will be an increase in psychological demand during this period, requiring actions to expand the network of mental health care and correct allocation of resources targeting higher risk publics and interventions with greater cost-benefit in the face of emergency needs. Strategies based on machine learning (ML) techniques have been applied in the field of mental health in order to assist in diagnosis, treatment and planning of interventions [3] [4], and may increase the accuracy of resource application in situations such as the COVID-19 pandemic.ML is a technique within the field of artificial intelligence (AI) and is achieved by building algorithms that "learn" from the data that are imputed to it, without an a priori hypothesis being defined [5]. This approach is especially useful for exploratory studies based on a large database or big data, since it derives predictive models without the need for initial hypotheses on the sample studied. Moreover, there is the possibility of predicting outcomes influenced by many factors, even if the relationship between them is unknown or not sufficiently defined [6]. The use of the technique consists of 4 stages: the treatment of the data, the learning process, the modeling, and the evaluation of the predictive power of the model [7]. The advantage of this type of analysis is that the predictive model is dynamic and "learns" more with each entry of new data, making the predictive model progressively more robust and accurate, with more learning about that group of data. These characteristics make ML an important tool for analysis and prediction in view of the complexity and multifactoriality of mental health outcomes such as variation in clinical symptoms, number of hospital admissions or number of suicides.Thus, this study proposes to analyze, through ML techniques, the behavior of mental health outcomes and the impact of the OVID-19 pandemic on them, performing the identification of groups and periods of alert. It is also intended, based on the results found and data survey in the literature after pandemic period of COVID-19, to propose possible measures to mitigate these effects in similar future scenarios.