Machine learning predicts psychosocial resilience among healthcare workers

A team of researchers trained three machine learning models-Logistic Regression, Random Forest, and Support Vector Machine (SVM)-to classify healthcare workers according to their level of psychosocial resilience, using information collected during the COVID-19 pandemic. The Logistic Regression model produced the most accurate results, with an accuracy of 75.6% and an area under the ROC curve of 0.816, outperforming Random Forest (72.6%) and Support Vector Machine (70.8%). The study was published in the journal Discover Artificial Intelligence.

Unlike most studies on healthcare workers' mental health, which focus on burnout and other negative outcomes, this research sought to predict a person's ability to actively cope with stress.

To train the models, the team used the "How Right Now Mental Health & Coping" dataset, collected by NORC at the University of Chicago between 2021 and 2022, comprising 2,055 respondents in the United States. The researchers constructed a resilience index based on four psychometric variables-resilience, ability to bounce back, control, and confidence-and divided it into two levels, high and low, using the median as the cutoff point.

The results show that stress, with a weight of 0.182 in the model, depression (0.160), and anxiety (0.145) are the strongest predictors of low resilience, followed by hopelessness and changes in sleep. Coping strategies-seeking social support, engaging in hobbies, prayer, and meditation-had a lower weight in the model but were associated with greater resilience. Their effect appeared to be cumulative; in other words, respondents who reported using several strategies simultaneously had, on average, higher scores than those who used only one.

The authors note that the data represent a single point in time, are based on participants' self-reports, and come from a U.S. population. These factors limit the extent to which the conclusions can be applied to other healthcare systems, including those in Latin America. They therefore propose that the next step should be to validate the model with Latin American healthcare workers and incorporate other measures, such as sleep data or physiological indicators.

The authors also suggest that models of this kind, based on existing data, could be used by occupational health departments to identify people at risk at an early stage, without relying on complex artificial intelligence systems.