deep-review
deep-review copied to clipboard
greenelab
Section Status: Electronic health records
Thanks for offering to help @spiros. It's unclear how active the manuscript update is right now. I added some thoughts in https://github.com/greenelab/deep-review/issues/810#issuecomment-560817165, and we'll see what the other project maintainers think.
I am excited to get this project rebooted and going. With #986 I'm adding infrastructure to handle authors from multiple versions. Could you describe what portions you think should be cut, updated, or expanded?
As the review is quite long, we probably won't want to expand sections without at least some cut in other areas. Fortunately, there may be recent reviews that we could cite if there are topics better dealt with by a more specialized review but that should still be covered.
I think we could restructure the sections based on three main use cases
- Risk prediction - how can ML/AI used with EHR to enable risk prediction of adverse outcomes
- Phenotyping - high-throughput phenotyping approaches using semi-supervised learning to tackle the labelling problem
- Subtype discovery - identify, validate latent disease types (static or longitudinal)
We can then weave in some cross-cutting things like embeddings (which can be used for all three use cases etc). I would cut out any NLP-specific tasks (like extracting info from radiology reports or similar) and move them in their own section.
It might be a good idea to add a paragraph on challenges such as interpretability, noise, biases in EHR data etc.
@spiros : could you weigh in over at #1005 about positioning for this section? We're considering a reorganization and this is one of the trickier ones. Does it go with genetics, where many immediate applications are, or does it belong with text/images?