Developing a Prediction Model for Suicidality Among COVID-19 Patients in Korea Using Timely Data From the National Center for Disaster and Trauma

Abstract

Background: To date, no study has developed a predictive model based on the timely data (diagnosed within a month) of patients with coronavirus disease 2019 (COVID-19). This study aimed to develop screening tools and a model for predicting suicidality based on data from patients diagnosed with COVID-19 within the past month. Methods: We analyzed data from 96,694 COVID-19 patients collected by the Korean National Center for Disaster and Trauma. Using classification and regression tree (CART) and random forest models, we predicted suicidality based on 39 features, including demographic information, COVID-19-related factors, and psychological symptoms such as depression, anxiety, somatic, and post-traumatic stress symptoms. Based on the final model with the highest performance index, we presented the results with a scorecard that includes multiple cut-off scores. Results: The area under curve was 0.71 for the CART model and 0.85 for the random forest model, with the sensitivity and specificity of the random forest model being 73.4% and 83.9%, respectively. The suicidality of COVID-19 patients was most strongly associated with suicidal ideation, fatigue, anticipatory anxiety, uncontrollable worrying, depressive mood, generalized worry, restlessness, changes in appetite, concentration problems, and feeling nervous. Conclusion: We developed a well-performing prediction model for suicidality among Korean COVID-19 patients and developed a scorecard to expand the feasibility of the model in public mental health practice.