# Derivation and validation of the Pediatric Community-Acquired Pneumonia Severity (PedCAPS) score: A prospective cohort study

## TL;DR

This study aims to develop and validate a new severity score for pediatric community-acquired pneumonia using a large, multicenter approach.

## Contribution

The study introduces a novel, large-scale, prospective method for predicting CAP severity in children using biomarkers and machine learning.

## Key findings

- A new severity prediction model for pediatric CAP will be derived and validated across multiple emergency departments.
- Biomarkers like C-reactive protein and procalcitonin will be evaluated for their role in predicting CAP severity.
- The model will be tested in over 6000 children across 14 emergency departments.

## Abstract

Community-acquired pneumonia (CAP) is a frequent and costly cause of pediatric emergency department (ED) visits and hospitalizations. Previous prognostic tools for CAP are limited by small samples, single-center or retrospective designs, lack of generalizability to ED settings, lack of biomarkers, or limited objective data. To overcome these limitations, we will derive and externally validate a prediction rule for pediatric CAP severity in a large, multicenter prospective cohort.

This is a prospective cohort study of children 3 months to 18 years old with CAP who present to EDs within the Pediatric Emergency Care Applied Research Network. Enrollment began 8/2023 and will end 7/2027. We exclude children with recent hospitalizations and chronic conditions (e.g., immunosuppression). A follow-up survey and record review is completed 8–15 days after the visit. Blood and nasal specimens are obtained to evaluate the role of C-reactive protein, procalcitonin, proadrenomedullin, and viral detection in severity prediction. The primary outcome is severity (three-tiered outcome of mild, moderate, or severe CAP) within 7 days of ED presentation. Model derivation will occur in ~4000 children from 7 EDs over 2 years. External validation will occur in a distinct cohort of at least 2000 children from 7 different EDs. Penalized regression, recursive partitioning, and machine learning will be used in model development.

At study completion, we will have a validated CAP severity prediction rule well-positioned for implementation and further evaluation. We will also understand the role of specific biomarkers in predicting outcomes in children with CAP.