Chapter 11: Hold the weekly meeting

In this chapter you will learn to:

• Run a weekly meeting that includes everyone who touches the data.
• Build an agenda from the registry itself.
• Reconcile data paths after a change to the study.

Case study: When the Data Stopped Flowing in One Direction

I analyze data for a large multimodal registry with ongoing collection. Participant survey data leave the collection platform along two paths. One goes to an internal database that holds identifiers, which the study coordinators use to look people up and to review incoming data. The other goes to a cloud data warehouse, and from there downstream to the secure environment where external collaborators work under a data use agreement. My analyses drew on the second path, the same substrate the external collaborators use.

In one extract, the data stopped abruptly after a particular month, well over a year before the extract date, even though the internal resources showed that participants had continued to enroll and respond. I raised it at the weekly team meeting, in passing, on the assumption that it was a known refresh delay. It was not known. The team had recently added a sub-study, an additional survey for a particular population, and in building it the connection that sent data to the warehouse had broken. The internal database was complete, so everything the coordinators reviewed looked correct. Only the downstream path, the one every external collaborator depends on, had stopped receiving new data.

The discrepancy was visible only from the downstream end, and it was resolved because the person working at that end and the people who maintained the upstream system sat in the same meeting. The exchange ran in both directions. In the same conversation the team explained the history of the study, why some participants have a given set of data and others do not, and documentation details that are not apparent from the files themselves. An AI coding agent working on the extract would have analyzed the data as delivered; it had no basis for knowing that more data should exist, and no way to ask the people who would know. I would have caught the gap with or without AI assistance. Diagnosing it and understanding its cause took a conversation.

A coffee shop holds a staff meeting for the same reason. The register can show a busy morning while the stockroom runs out of milk, and the problem comes to light in time only when the person at the register and the person who orders supplies compare notes.

Each end of a pipeline sees only its own view; the gap was found because the person at the downstream end and the people who maintain the upstream system sat in the same meeting.

Lessons for registry leaders

A pipeline can be complete at one end and broken at the other, and the people at each end see only their own view. A standing meeting that includes everyone who touches the data is where those views meet, and where the context behind the data passes from the people who know it to the people who analyze it.

The rules

1. Meet weekly, as a team, on a fixed day. This is the habit the others depend on.
2. Include everyone who touches the data, not only the coordinators: the analysts, the data manager, and whoever prepares data for external collaborators. A problem in a data pipeline may be visible from only one end, and the people at the two ends often do not cross paths otherwise.
3. Review every new enrollment since the last meeting, and then walk the cohort: who is overdue for follow-up, who has large gaps, whose diagnosis is still unconfirmed, whose contact details failed, which questions people are skipping.
4. Put everything else on the same agenda: IRB items, vendor problems, website upgrades, new questions, sub-studies, and any challenge a coordinator is having. The meeting is where problems get named before they become losses.
5. Raise anything that looks wrong, even when you assume the team already knows about it. A brief mention in passing is often how a broken connection or a missing extract comes to light.
6. When the study changes (a new sub-study, survey, form, vendor, or platform upgrade), confirm at the next meeting that the new data reach every destination: the internal database, the warehouse, and the environment external collaborators use. Checking only the view the team reviews daily leaves the others unverified.
7. Reconcile the data paths on a schedule. Compare enrollment and response counts by month in the internal system against what appears in the collaborator-facing data, and treat any gap as an agenda item.
8. Discuss identifiable data only with study personnel the IRB has approved, and bring the medical advisor when a diagnosis or a reclassification is on the list.
9. Assign every follow-up to a named person with a date, and start each meeting by reviewing the previous list.
10. Keep minutes that record decisions, not conversation, in shared documentation and shared file storage the whole team can find.
11. Report a missing-data picture at each meeting, by field and by participant, and look for patterns (a country, a language, a form).
12. Use the meeting to pass on context as well as problems: the history of the study, why some participants have data that others lack, and documentation details that are not apparent from the files. Analysts and AI tools both work from the files; the context lives with the team.

What AI can do, and what only a person can do

What AI can do

Produce the overdue list and the missingness summary from the export; draft the agenda; run a count comparison across data paths once someone has defined the check.

What only a person can do

Notice that data which should exist are absent; explain why the study's data look the way they do; decide what to do about each person; make the calls that follow.

Take-home

A cohort that is reviewed one participant at a time has data at the end of the year, and a team that brings everyone who touches the data into one meeting finds the problems that no single view of the data shows.

Practice

Exercise

The export is dated 15 March 2044, and the weekly meeting falls on the same day. Using the synthetic Lantern Registry export, build the agenda in the weekly meeting template. Treat any participant whose last follow-up is more than six months old, or missing, as overdue.

For the reconciliation item, suppose the internal system the coordinators use shows 34 consented participants, while the export that external collaborators receive shows 32 records.

Check your work

• Enrollment: no new enrollments since the last meeting. LR-029 has an enrollment date that cannot exist (month 13) and needs a query.
• Overdue follow-up: LR-011 (last follow-up June 2042), LR-021, LR-022, and LR-024 in India and LR-023 in Brazil (no follow-up recorded), and LR-010, whose follow-up date falls before enrollment.
• Diagnoses still unconfirmed: the 25 reports on file awaiting review by a qualified person, plus LR-003 and LR-011 (Tier 0), LR-027 (clinician letter), LR-006 (variant of uncertain significance), and LR-017 (conflicting gene).
• Records to resolve: the duplicate LR-012 and LR-005; LR-026 (not a case).
• Questions people are skipping or answering oddly: placeholders in the seizure count (999, -1, "daily"), mixed units in age at onset.
• Reconciliation: two consented participants are missing from the external export. Check whether a recent change, such as a new form or sub-study, broke the path to the collaborators' data, and compare counts by month on both paths.
• Follow-ups assigned: each item gets a named person and a date, reviewed first at the next meeting.

Your workbook page

The Weekly meeting agenda and minutes page of the Slow Registry Workbook applies this chapter to your own registry. Fill it in online or print it and work by hand. The AI prompt library has a prompt for this chapter's topic, paired with the step only a person can do.

Check your understanding

Why include the analyst who prepares data for external collaborators in the weekly meeting?

Because a problem in a data pipeline can be visible from only one end. The coordinators' view can look complete while the data collaborators receive have stopped.

What should happen after a new sub-study or form is added?

At the next meeting, confirm that the new data reach every destination: the internal database, the warehouse, and the environment external collaborators use.

Up next: The review meeting will show where the data are missing; the next habit is going to find out why. Continue to Chapter 12.