MGP-AttTCN: An interpretable machine learning model for the prediction of sepsis
Problem
Sepsis kills about 5.4 million people worldwide each year and costs over 16 billion dollars in the USA alone, and early detection in hospitals is difficult.
Why it matters
Sepsis is a leading cause of hospital mortality with millions of deaths and billions in costs.
What the AI did
MGP-AttTCN, combining multitask Gaussian Process and attention-based deep learning, early-predicts sepsis onset in an interpretable manner; the authors also released an open-access labelling of the MIMIC-III dataset.
Results
Claims to outperform current state-of-the-art; concrete metrics not stated in source
Impact
Provides an open-access labelled MIMIC-III resource and a model claimed to beat state-of-the-art for early sepsis detection.
Record
| Venue | PLoS ONE |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Margherita Rosnati, Vincent Fortuin |
| Citations | 56 (OpenAlex, at import time) |
| References | 82 |
| Model or method | MGP-AttTCN (multitask Gaussian Process + attention-based deep learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1371/journal.pone.0251248 |
| Open access | gold |
| Keywords | sepsis, early prediction, deep learning, Gaussian process, MIMIC-III, interpretability |
Importance rating
major — millions of lives, an industry or a field
Abstract states sepsis kills 5.4 million people yearly with over $16 billion US healthcare cost, a leading cause of hospital mortality; early detection is difficult and the work adds an open-access labelled MIMIC-III resource.