# Clinical Predictive Models for COVID-19: Systematic Study

**Authors:** Patrick Schwab, August DuMont Schütte, Benedikt Dietz, Stefan Bauer  
**arXiv:** [2005.08302](https://arxiv.org/abs/2005.08302)  
**Date:** 2020-05-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2005.08302v2/

## Key points

- Study explores use of predictive algorithms to alleviate strain on healthcare systems caused by COVID-19
- Authors focus on identifying patients most likely to test positive for SARS-CoV-2, require hospitalization, or need intensive care
- Machine learning techniques and clinical data used to develop clinical predictive models
- Retrospective evaluation conducted using data from 5,644 patients
- Predictive models can identify SARS-CoV-2 positive patients with 75% sensitivity and 49% specificity
- Models can predict hospitalization with AUC value of 0.92 and critical care with AUC value of 0.98
- Clinical features that are most predictive for each outcome determined
- Predictive models can help inform care decisions and prioritize resources in managing COVID-19 cases
- Potential applications in predicting clinical pathways for COVID-19 and improving patient outcomes while reducing strain on healthcare system

## Summary

In their study titled "Clinical Predictive Models for COVID-19: Systematic Study," authors Patrick Schwab, August DuMont Schütte, Benedikt Dietz, and Stefan Bauer explore the use of predictive algorithms to alleviate the strain on healthcare systems caused by the rapid transmission of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), which causes Coronavirus Disease 2019 (COVID-19). The authors focus on identifying patients who are most likely to test positive for SARS-CoV-2, require hospitalization, or need intensive care. Using machine learning techniques and routinely collected clinical data, the authors develop clinical predictive models to estimate these outcomes. To evaluate the performance of their models, they conduct a retrospective evaluation using clinical and blood analysis data from a cohort of 5,644 patients. The experimental results indicate that their predictive models can identify patients who test positive for SARS-CoV-2 with a sensitivity of 75% and a specificity of 49%. Additionally, the models can predict which SARS-CoV-2 positive patients will require hospitalization with an area under the curve (AUC) value of 0.92 and which ones will require critical care with an AUC value of 0.98. Furthermore, the authors determine which clinical features are most predictive for each of these outcomes. This information can help inform care decisions and prioritize resources in managing COVID-19 cases. Overall, this study demonstrates that predictive models trained on routinely collected clinical data have potential applications in predicting clinical pathways for COVID-19. By identifying high risk individuals early on, healthcare systems can better allocate resources and provide timely interventions to those who need them most; thus improving patient outcomes while reducing strain on medical personnel and facilities during times of crisis such as pandemics like COVID 19.