# The Detection of COVID-19-Related Multivariate Biomarker Immune Response in Pediatric Patients: Statistical Aspects

## TL;DR

This paper presents a statistical approach to detect a severe form of COVID-19 in children using immune biomarkers and compares it to machine learning models.

## Contribution

A novel two-stage statistical method combining biomarker selection and dimension reduction for predicting MIS-C in pediatric patients.

## Key findings

- A bivariate analysis identified significant immune biomarkers associated with MIS-C in children.
- Dimension reduction via principal components improved predictive accuracy for MIS-C detection.
- A logistic regression model using principal components outperformed an artificial neural network in predicting MIS-C.

## Abstract

The development of new point-of-care diagnostic testing tools for the detection of infectious diseases such as COVID-19 are a key aspect of clinical care and research. Accurate predictive classification methods are required to correctly identify and treat patients. Here, the onset of multisystem inflammatory syndrome in children (MIS-C), a more serious form of COVID-19, was predicted in a pediatric population using a set of multivariate immunological biomarker expression values. A first-stage bivariate detection of statistically significant biomarkers was obtained from a chosen set of standard cytokines and chemokine biomarkers considered relevant to COVID-19-related infection and disease. To incorporate the observed correlation structure among the resulting set of significant biomarkers, dimension reduction was then applied in the form of principal components. A second-stage logistic regression model using a small number of the principal component variables provided a highly predictive classification model for MIS-C. The resulting model was shown to compare favorably with an artificial neural network-based predictive model.