# Validation of a software application using electronic health records for automatic detection of community onset sepsis

## TL;DR

Researchers developed and validated a software tool that automatically detects community-onset sepsis in hospital admissions using electronic health records, achieving high accuracy.

## Contribution

A novel software application for automatic sepsis detection using EHR data, validated against physician reviews with high performance metrics.

## Key findings

- The software detected 7,027 sepsis cases among 60,213 hospital admissions with high sensitivity and specificity.
- Validation showed 95% sensitivity, 99% specificity, and 92% positive predictive value compared to physician reviews.
- Lower respiratory tract was the most common infection site identified by the tool.

## Abstract

Our aim was to design and validate a software application, based on the Sepsis-3 criteria, capable of retrospectively identifying community-onset sepsis among emergency department patients requiring hospital admission.The application was developed using QlikView (Qlik, King of Prussia, PA, USA) software, and accessed data from the electronic health records TakeCare (CompuGroup Medical, Koblenz, Germany), and CliniSoft (CliniSoft, Kuopio, Finland). The application utilized indicators such as blood culture data, antibiotic administration, and Sequential Organ Failure Assessment scores to detect sepsis cases according to Sepsis-3 criteria. The application was tested retrospectively against a cohort from a large city hospital in Stockholm over a 2-year period, and its performance was compared to physician record reviews in a subset of cases identified by stratified random sampling. The results showed that among 229,195 emergency department visits leading to 60,213 hospital admissions, the application detected 7027 cases of sepsis. Validation using physician record review of a random selection of 426 cases demonstrated a sensitivity, specificity, positive predictive value, and negative predictive value of 95%, 99%, 92%, and 99%, respectively. The lower respiratory tract was the most common site of infection. This software application effectively identified community-onset sepsis patients using electronic health record data with high performance. It has the potential to improve sepsis identification as it operates independently of diagnostic codes and may, therefore, facilitate research in many areas of sepsis. Furthermore, it can be used as a tool within the healthcare system to enhance sepsis surveillance and evaluate quality improvement interventions.