Independent researcher seeking reference for MIMIC-IV access — happy to connect with fellow researchers

[alwaysmindy]

Hello PhysioNet community,

I am an independent researcher based in South Korea working on a project that combines zero-knowledge proofs (ZKP) with machine learning fairness verification in clinical AI — specifically, verifying that diagnostic models produce fair outcomes across demographic groups (sex, ethnicity) without exposing raw patient data.

I have completed the CITI "Data or Specimens Only Research" course (Score: 92%, completed 23-Mar-2026) and I am ready to apply for MIMIC-IV access.

However, as an independent researcher without a current institutional supervisor, I am looking for a colleague or fellow MIMIC user who would be willing to serve as my reference on PhysioNet.

If anyone in this community is familiar with privacy-preserving ML research or healthcare AI fairness, I would love to connect. I am happy to discuss my research in detail, share my research plan, and of course reciprocate support wherever I can.

Please feel free to reply here or contact me directly.

Thank you very much for your time and support.

Best regards,
Minsun Lee

[amitsaha]

Hi @alwaysmindy i am keen to learn more. What is the best way to contact you directly or learn more about your work?

[alwaysmindy]

### Research Proposal and Request for Professional Reference (MIMIC-IV Access) Dear Amit Saha, Thank you so much for your enthusiastic response! I am thrilled to hear that you are interested in learning more about my work. To answer your question, replying directly to this email is the best way to contact me. As mentioned, I am focusing on a framework to apply zero-knowledge proofs to clinical fairness auditing using the MIMIC-IV dataset. I have attached the PDF version of my current research plan to this email for your review, and you can also track the project repository directly on GitHub via this link: https://github.com/alwaysmindy/zkml-fairness-clinical-ai I would love to hear your thoughts or feedback on the technical pipeline. If you find the research direction meaningful and feel comfortable supporting it, I would be incredibly grateful if I could list you as a professional reference for my "Application for Credentialed Access" on PhysioNet. To fill out the required fields in their application form, I would just need your permission to enter the following basic details: - Reference Category: (e.g., Professional Contact / Mentor) - Name: - Email: - Organization: - Job Title: Please note that according to PhysioNet’s guidelines, it is highly recommended for the reference to be a registered user on PhysioNet and have their ORCID iD linked to their profile, as this significantly expedites their identity verification process. There is absolutely no pressure, so please take your time to look over the research plan first. Thank you again for your time and willingness to help. I look forward to staying in touch! Best regards, Minsun Lee 2026년 5월 17일 (일) 오전 7:43, Amit Saha ***@***.***>님이 작성:

…

Hi @alwaysmindy <https://github.com/alwaysmindy> i am keen to learn more. What is the best way to contact you directly or learn more about your work? — Reply to this email directly, view it on GitHub <#1978 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AWBI5DOZENG5OZLCOZAHQKL43DVJTAVCNFSM6AAAAACW3U4XAGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMOJUGMYDKMI> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[amitsaha]

Hi @alwaysmindy, thank you for sharing the project plan. Since i cannot see your email here, please send me an email to asah0916@uni.sydney.edu.au ?

[alwaysmindy]

Thank you again for reaching out!
I have sent you an email with my contact information.
Please check your inbox.
Looking forward to hearing from you soon.

[fw2274]

Hi Minseon and Amit, great to see the discussion and hear about your project! I'm curious about the how resampling and technical approach enables fairness and ensure black box privacy. The topic seems to be interesting! Can I join the plan discussion if you guys are having one? I have done MIMIC project before and am familiar with mortality data - should be able to be listed as a reference as well. My email is ***@***.***, feel free to contact me! Best, Fan

…

On Tue, May 26, 2026, 2:45 AM Minseon_2 ***@***.***> wrote: Thank you again for reaching out! I have sent you an email with my contact information. Please check your inbox. Looking forward to hearing from you soon. — Reply to this email directly, view it on GitHub <#1978?email_source=notifications&email_token=AHIJ24O4KMQ7E6QIIBKYFG344U4RXA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZQGU4DENRSUZZGKYLTN5XKU43VMJZWG4TJMJSWJJLFOZSW45FMMZXW65DFOJPWG3DJMNVQ#discussioncomment-17058262>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHIJ24N5EEDB7QQYLH734Z344U4RXAVCNFSM6AAAAACW3U4XAGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTOMBVHAZDMMQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.*** com>

[netanelcyber]

Hi all,

This discussion on temporal validity and evaluation design resonates strongly with something we observed while building a reproducible ICU pathogen prediction pipeline on MIMIC-IV.

We found that even when AUROC remains relatively stable, model reliability can change substantially under different validation regimes, especially:

• temporal splits vs random splits
• subgroup-specific calibration shifts
• rare outcome / pathogen class instability
• sensitivity to preprocessing and feature aggregation choices

In practice, small pipeline differences (imputation strategy, windowing, feature collapsing) can lead to non-trivial changes in calibration behavior even when discrimination metrics look similar.

We are currently packaging these observations into a reproducible benchmark-style pipeline focused specifically on:

• calibration drift tracking
• temporal generalization stress tests
• subgroup robustness reporting
• pathogen prediction as a ranking problem rather than binary classification

Repo (for context, not as a claim of completeness):
https://github.com/netanelcyber/PenuX

I would be very interested to hear if others here have tried:

• systematic temporal calibration tracking across MIMIC-IV pipelines
• or observed similar instability under “minor” preprocessing variations

This seems like an area where a shared evaluation template (beyond AUROC reporting) could meaningfully improve reproducibility across studies.

Thanks for the discussion — it’s extremely aligned with ongoing issues in clinical ML evaluation.