search

BorgwardtLab / mgp-tcn

Sepsis Prediction on MIMIC
BSD 3-Clause "New" or "Revised" License
66 stars 25 forks source link

Question: Is there any interest to extend this work to a bigger dataset like eICU? #3

Open harshblue opened 4 years ago

harshblue commented 4 years ago

BorgwardtLab,

https://eicu-crd.mit.edu/ is a much larger multicenter dataset. I am wondering how the models transfer to different centers across eicu

Given that MIMIC data is only from one center Beth Israel Deaconess Medical Center in Boston, establishing generalizability of the models would be fantastic.

Is there any interest to do above research? I ask to cheer from sidelines if it is ongoing, if not I would like to collaborate if their interest from your lab.

There is a recent Nature paper which validated their models on both Mimic and eICU [https://www.nature.com/articles/s41598-019-42637-5]. If there is interest we can reach out to get the dataset for reproducibility and comparison.

Thanks, Sree

mi92 commented 4 years ago

Dear Sree, I am very interested in extending the analysis to e-ICU. Apparently, it is not super straightforward to implement Sepsis-3 on eICU which is why in the Nature paper, they had to slightly modify it. Therefore, and not to 'modify' the definition several times in an uncomparable manner, I would like to work with their dataset implementation. However, their eICU code is currently still pending. For this type of validation analysis, I am very interested in collaboration.

harshblue commented 4 years ago

@mi92 thank you for the reply. The Natures paper authors have put their code include how they worked with eICU here https://github.com/instigatorofawe/shockalert-documented

Is this a good starting point?

Thanks, Sree