Benchmark results over standard EHR tasks like diagnosis prediction

[avani17101]

Was there any benchmarking done over EHR tasks like next disease, medication predictions tasks or any tasks described in this paper: A Comprehensive EHR Timeseries Pre-training Benchmark @mmcdermott, @bnestor ?

[mmcdermott]

So far, we haven't, no; the focus to date has just been on the software/API development and not on specific model architectures. It is a great idea, though, and we're very open to contributions on that front! It's also on our roadmap going forward.

On Fri, Jul 21, 2023, 5:12 PM Avani Gupta @.***> wrote:

Was there any benchmarking done over EHR tasks like next disease, medication predictions tasks or any tasks described in this paper: A Comprehensive EHR Timeseries Pre-training Benchmark https://www.researchgate.net/publication/350736320_A_comprehensive_EHR_timeseries_pre-training_benchmark @mmcdermott https://github.com/mmcdermott, @bnestor https://github.com/bnestor ?

— Reply to this email directly, view it on GitHub https://github.com/mmcdermott/EventStreamGPT/issues/41, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADS5X36KHSPLDIZP5A5N4DXRLWC5ANCNFSM6AAAAAA2TKI7RA . You are receiving this because you were mentioned.Message ID: @.***>

[mmcdermott]

Worth noting, the paper you reference also uses MIMIC-III, not MIMIC-IV. Doesn't refute the value of getting results on standard benchmarks, but just wanted to clarify there is a dataset difference here.

[avani17101]

@mmcdermott Thanks for the response! will be looking forward to its performance on such downstream evaluation tasks!

Also, truly noted the benchmark paper is on MIMIC-III, but there are many papers evaluating on a subset of those (length of stay, diagnosis levels, readmission, mortality, etc) over MIMIC-IV. For instance UniHPF.