Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare
Problem
Sepsis is a leading cause of hospital death and hard to predict early because its signs resemble less critical conditions.
Why it matters
Early sepsis detection is critical to reducing hospital mortality.
What the AI did
The SERA algorithm combines structured data and unstructured clinical notes to predict and diagnose sepsis before onset, and was tested on independent clinical notes.
Results
AUC 0.94, sensitivity 0.87, specificity 0.87 at 12 hours before onset; up to 32% more early detection and 17% fewer false positives versus physician predictions.
Impact
Shows potential to improve early sepsis detection over physician predictions; clinical deployment not stated.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Kim Huat Goh, Le Wang, Adrian Yong Kwang Yeow, Hermione Mei Niang Poh, Ke Li, Joannas Jie Lin Yeow, Gamaliel Yu Heng Tan |
| Citations | 412 (OpenAlex, at import time) |
| References | 41 |
| Model or method | SERA algorithm |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-021-20910-4 |
| Open access | gold |
| Keywords | sepsis prediction, clinical notes, artificial intelligence, early diagnosis, AUC |
Importance rating
major — millions of lives, an industry or a field
Abstract states sepsis is a leading cause of hospital death and early prediction is critical to reducing mortality; algorithm beats physician predictions with quantified gains (32% more early detection).