Using AI to Predict COVID-19 Severity

Sample size: 421 publication 10 minutes Evidence: moderate

Author Information

Author(s): Yuri Oliveira, Iêda Rios, Paula Araújo, Alinne Macambira, Marcos Guimarães, Lúcia Sales, Marcos Rosa Júnior, André Nicola, Mauro Nakayama, Hermeto Paschoalick, Francisco Nascimento, Carlos Castillo-Salgado, Vania Moraes Ferreira, Hervaldo Carvalho

Primary Institution: University of Brasilia

Hypothesis

Can machine learning algorithms effectively predict clinical outcomes of COVID-19 patients based solely on hospital admission data?

Conclusion

The study found that machine learning algorithms can accurately predict severe outcomes in COVID-19 patients using only data from hospital admission.

Supporting Evidence

The system achieved an accuracy of 80%.
The Area Under Receiver Operating Characteristic Curve (AUC) was 91%.
Positive Predictive Value was 87% and Negative Predictive Value was 82%.
Data was collected from multiple hospitals to enhance generalizability.
Machine learning algorithms were trained on clinical data from hospital admission.

Takeaway

This study shows that computers can help doctors figure out which COVID-19 patients might get really sick just by looking at their information when they first arrive at the hospital.

Methodology

The study used a prospective multicenter cohort design, collecting clinical data from hospitalized COVID-19 patients to train machine learning algorithms.