Two documents, one decision: how Mithras consent works

The signature line is on page nineteen. Before it: eighteen pages of defined terms, nested subclauses, and risk disclosures drafted by one set of lawyers and polished by another. The form will survive an audit. It satisfies every reviewer who will ever read it. Whether the person holding the pen understood what they agreed to is a different question — and in most research consent processes, an unasked one.

Mithras Health asks it first. Our answer is a consent built from two documents and one decision, and this post explains how the pieces fit.

The form nobody reads

Consent forms fail because one document is asked to do two jobs — legal protection and human understanding — and the legal job always wins. The result is well documented: readability studies of research consent forms have found, again and again, that they test at college reading level, while plain-language guidance for patient materials recommends something closer to middle school. The Common Rule itself requires that consent information be presented "in language understandable to the subject," and the Office for Human Research Protections has long pushed investigators in that direction. The forms keep getting longer anyway.

This is not because anyone wants confusion. It is because every clause earns its place defensively, no clause is ever deleted, and the patient's comprehension is the one stakeholder without a lawyer in the drafting room. A signature on a document the signer could not parse is consent in the narrowest procedural sense — and trust-destroying in every other sense.

Two documents, one decision

The Mithras design splits the two jobs apart instead of pretending one document can do both. Every consent request a patient receives consists of a plain-language summary and a formal authorization, presented together, decided together.

The plain-language summary exists to be understood. It says, in short sentences: which study is asking, which parts of your record it wants, what the study is for, what you will be paid, and how to change your mind. It is short enough to actually read and written for the person whose data is at stake — not for the file.

The formal authorization is the legal instrument. It carries the statutory machinery a research authorization needs, in the register regulators and review boards require.

One decision binds both. There is no scenario where you agree to the friendly version while the binding version says something else: the two documents describe the same grant, the summary is the authorization translated rather than abridged, and both are archived together with your decision.

What the formal document carries

The authorization exists because the law is specific about what a valid research permission contains. Under HIPAA's §164.508, a research authorization must describe the information to be used, who may disclose and receive it, the purpose, an expiration, and the right to revoke — we walk through the details in HIPAA and research. Under the GDPR, explicit consent for health data must meet the conditions of Article 7, including the requirement that withdrawing consent be as easy as giving it; the UK's ICO reads UK GDPR the same way.

Mithras builds to all of these at once, because the platform is designed for HIPAA, GDPR, UK GDPR, EHDS, and Washington's My Health My Data Act simultaneously. The formal document is where those obligations live, so the summary doesn't have to carry them — only translate them.

One decision, one study

Mithras consent is per-study and scope-limited: a yes covers the named study, the named data scope, and nothing else. This is a deliberate departure from biobank-style broad consent, where one signature authorizes an open-ended future of unnamed projects — a trade-off we examine in broad consent vs. dynamic consent. When the next study comes, it asks. It arrives as a new pair of documents, a new plain-language summary, a new decision. Nothing is assumed from the last yes, and silence is always a no.

This costs more — more requests, more reading, more chances for a patient to decline. We think the cost runs in the right direction. A platform whose revenue depends on patients continuing to say yes, study after study, has every reason to keep each request honest, specific, and small.

Versioned, archived, revocable

Both documents are versioned, and every decision is recorded against the exact versions presented — on the same append-only, cryptographically anchored audit trail that records everything else on the platform, with six-year retention. The question "what exactly did I sign, and which wording was in front of me?" has a permanent answer, for the patient and for any auditor. We describe how that spine works in append-only, tamper-evident.

Versioning also answers the harder question: what happens when a study changes? Protocols get amended; scopes drift. On Mithras, an amended request is a new version of both documents and a fresh decision — your old yes does not silently stretch to cover the new terms. The previous pair stays archived exactly as you saw it, the new pair is presented exactly as it now reads, and the audit trail records which version governed which use. Consent that cannot be quietly re-scoped is the only kind worth recording.

Revocation rides the same rails. It is a first-class event, not a support ticket: a revocation stops future provisioning of your records immediately, is logged with the same permanence as the consent it cancels, and is honored as fast as the original yes. What it cannot do is reach back and erase aggregate results already published — and the plain-language summary says exactly that, in plain words, before you decide. Honesty about limits is part of the job; the full reasoning is in what revocable consent really means.

Page nineteen was never the problem; the eighteen pages before it were. A consent process earns the word informed only when understanding is engineered into it with the same care as liability — two documents, each doing its own job, one decision a person can actually stand behind.