The loops

A funded study is a strange object. It has money, a protocol, ethics approval pending, a principal investigator with a question they've wanted to answer for a decade, and a research assistant who has just been told to build the database. Nobody budgeted for that last part. It's the part that decides when the study finishes.

What happens next is so predictable I've started calling it the loops. Ethics reads the consent form and wants the consent architecture changed, so the forms are rebuilt. Governance asks for a data management plan, and nobody knows what it should say, so a template is found and filled in with things that aren't true yet. Cyber security asks what the posture is for a database holding patient data, and the honest answer is that it's a spreadsheet on a shared drive, so that starts again. Meanwhile the research assistant has built the instruments from something they found on the internet, and for six months it collects the wrong thing. Analysis starts. The dataset can't answer the question the protocol asked. And, years later, someone asks whether the cohort could be linked to hospital outcomes, and it can't, because nobody designed it to be.

Every loop is the same loop: a decision made too late.

Where the loops come from

None of these people is being unreasonable. Ethics is right about consent. Governance is right to want a plan. Cyber is right that a spreadsheet of patient data is a breach waiting for a date. The research assistant did the best anyone could with a template. The problem is that every one of these is a decision, and every one of them was made after something had already been built that the decision would change.

Consent architecture is a decision. What the data flow is, who holds it and where, is a decision. Which validated instruments to use, and what quality rules apply to each field, is a decision. Whether the identifiers and demographics are recorded in a form that a linkage service will accept in 2040 is a decision. Made on day one, each takes an hour. Made on day ninety, each costs a month, because there's now a thing to unbuild.

The same loops, everywhere

I first saw this in research, but it's not a research problem. A data request to a linkage service goes through the same loops when the request was written before anyone asked what the custodians would test for. A data platform goes through them when access tiers are bolted on after the products are built. An AI pilot goes through them when the model is trained before anyone decided which of three versions of the data was true. In every case the organisation experiences it as delay, and blames the approver. The approver is the last person to blame. They're the one who finally asked the question.

Designing the loops out

The fix is not a faster committee. It's moving the decisions to the front. For a study, that means the build starts with the consent architecture, the data flow, the security posture and the linkage-readiness written down, before a single form exists, in the language the approvers use, because you've read enough of their decisions to know it. Then the forms are built to those decisions, once. Submissions arrive complete. Approvals happen the first time. The dataset arrives at analysis able to answer the question. And the cohort can be followed passively for decades without anyone having to re-contact a participant, because the identifiers were right from the start.

For a platform or a data request the pattern is identical, only the vocabulary changes. Find out what the person who says yes is going to test for. Build that in. Then ask.

The approver is the last person to blame. They finally asked the question.

Why nobody does this

Because it's boring, and because it requires someone who has been on both sides. The person building the study rarely knows what the research office is looking for. The research office rarely builds studies. Platform administrators, quite reasonably, don't build projects for their users. So the decisions sit in a gap between people, and the gap is where the loops live.

I've spent six years on the approving side of that gap and a good deal longer on the building side. The whole service, whether it's a study, a request or a platform, is the same sentence: make the decisions on day one, in the approver's language, and there are no loops. It's not clever. It's just early.