# MDS-ED: Multimodal Decision Support in the Emergency Department -- a Benchmark Dataset for Diagnoses and Deterioration Prediction in Emergency Medicine

**Authors:** Juan Miguel Lopez Alcaraz, Nils Strodthoff  
**arXiv:** [2407.17856](https://arxiv.org/abs/2407.17856)  
**Date:** 2024-07-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2407.17856v1/

## Key points

- Introduction of a dataset and benchmarking protocol for evaluating multimodal decision support in the emergency department (ED)
- Dataset based on MIMIC-IV with diverse data modalities collected within the first 1.5 hours of patient arrival
- Analysis covers 1443 clinical labels for predicting diagnoses with ICD-10 codes and forecasting patient deterioration
- Multimodal diagnostic model achieves AUROC score over 0.8 for various conditions including cardiac issues like myocardial infarction and non-cardiac conditions such as renal disease and diabetes
- Deterioration model performs well with scores above 0.8 for critical events like cardiac arrest, mechanical ventilation, ICU admission, short- and long-term mortality
- Effectiveness of incorporating raw waveform data into models highlighted
- Potential to revolutionize decision-making in acute and emergency medicine by enhancing acute care through AI models enabling early diagnosis, predicting admissions to different care units, estimating survival rates, and providing cost-effective diagnoses
- Methodology involves linking ECG waveforms from MIMIC-IV to clinical features from MIMIC-IV and MIMIC-IV-ED datasets to predict discharge diagnoses and patient deterioration during an ED visit

## Summary

The researchers have introduced a dataset and benchmarking protocol for evaluating multimodal decision support in the emergency department (ED) to address challenges in benchmarking medical decision support algorithms. The dataset is based on MIMIC-IV and includes diverse data modalities collected within the first 1.5 hours of patient arrival. The analysis covers 1443 clinical labels across two contexts: predicting diagnoses with ICD-10 codes and forecasting patient deterioration. Results show that the multimodal diagnostic model achieves an AUROC score over 0.8 for 357 out of 1428 conditions including cardiac issues like myocardial infarction and non-cardiac conditions such as renal disease and diabetes. The deterioration model performs well with scores above 0.8 for critical events like cardiac arrest, mechanical ventilation, ICU admission, short- and long-term mortality, highlighting the effectiveness of incorporating raw waveform data into the models. This study emphasizes the uniqueness of their dataset in covering a wide range of clinical tasks early in the ED visit and its potential to revolutionize decision-making in acute and emergency medicine. Furthermore, it discusses how AI models can enhance acute care by enabling early diagnosis, predicting admissions to different care units, estimating survival rates, and providing cost-effective diagnoses. While there has been exponential growth in AI-related publications in healthcare, many studies have limitations such as narrow scopes or requiring costly diagnostic tests. The methodology used involves linking ECG waveforms from MIMIC-IV-ECG to clinical features from MIMIC-IV and MIMIC-IV-ED datasets to create a comprehensive dataset for predicting discharge diagnoses and patient deterioration during an ED visit. Overall, this research contributes valuable insights into improving decision-making processes in emergency medicine through advanced AI models leveraging multimodal data early in patient care.