Clinical AI & Diagnostic Bias · Capstone brief
CLIN-09 · Consent and secondary use
A public clinical dataset exists because patients were treated, not because they volunteered for your project. Write the consent and data-use analysis for training on it as if your health system were the steward deciding whether to release it — and say what you would refuse to release.
The question
A public clinical dataset exists because patients were treated, not because they volunteered for your project. Write the consent and data-use analysis for training on it as if your health system were the steward deciding whether to release it — and say what you would refuse to release.
System / materials
One public clinical dataset with a documented provenance and data-use agreement; the DUA and any published de-identification method are the primary reading, not the CSV. Frameworks to read for vocabulary: HHS guidance on de-identification of protected health information — Safe Harbor and Expert Determination (https://www.hhs.gov/hipaa/for-professionals/special-topics/de-identification/index.html); the Common Rule on human-subjects research and secondary use (https://www.hhs.gov/ohrp/regulations-and-policy/regulations/45-cfr-46/index.html). No student signs a data-use agreement.
Expected failure modes
Treating "de-identified" as "consented" — they are different questions and the brief fails if they are merged. Summarizing the regulation instead of applying it to this dataset. Ignoring re-identification risk from combinations of quasi-identifiers. Writing a policy with no refusal in it, which is a brochure.
Done looks like
A steward's memo: how these records were collected and under what authority; what de-identification was applied and what residual re-identification risk remains, with the combining fields named; who benefits and who bears the risk from release; conditions the steward would attach; and an explicit would-not-release list with reasons.
Five C's
CT: separating legality, de-identification, and consent as three distinct questions. CR: setting release conditions that are enforceable rather than aspirational. CO: a peer argues the researcher's side against the steward's. CM: a memo a compliance officer could read in one sitting. CZ: which patients bear re-identification risk and have no voice in the decision.
Mentor role
A compliance officer, IRB coordinator, health-information manager, or clinical informaticist reviews the memo. Standing instruction: reject any memo without a would-not-release list. School-supervised.
Rubric calibration
R1: one dataset, one steward posture. R2: every claim traced to the DUA or published guidance. R3: comparator is the actual release terms of the dataset. R4: re-identification pathways named concretely. R5: memo is one sitting. R6: the refusal list is specific.
Two ways this goes wrong
(a) A regulation summary with the dataset mentioned once. (b) "It's de-identified, so consent doesn't apply" — the exact conflation the brief exists to break.
Credit lane fit
Lane A immediately (health-science capstone, health-informatics or law-and-ethics cross-listing). Strong non-modeling option. No verified credit claim.