Skip to content

Closing: The long walk

The closing chapter walks the whole route once more, through a year in the Lantern Registry with you as its leader. You plan it, run it, take it to the community and the field, and decide what AI may do with it, and you finish with the report a board member and a sponsor would read. Most of the work draws on the exercises in the chapters; where you skipped one, do it now.

Work from the Lantern Registry files, which are synthetic, the export of 15 March 2044 in particular, and keep your answers in the workbook pages named at each step.

Stage one: Before anyone enrolls

Assume the Epilepsy Syndrome X community wants a registry that could one day support a treatment trial, and that the export you have is what the registry's first version produced.

  1. Name the users of the Lantern Registry's data and rank them. Record what the most important user needs from the data, and how you know. (Table 1 test)
  2. Name the first product the community wants, a cohort a sponsor could recruit from, and list what it requires that the export holds, holds at too low a standard, or lacks. (First product and analysis plan)
  3. List the fields that user would need and, for each, a better source and schedule than the export used. (Data source plan)
  4. Draft the protocol synopsis for the registry's next version, including the case definition, the verification tiers, the global unique identifier, and the place for unsolicited information: the question "Is there anything else you would like to share with us today that we missed?" and how its answers are recorded. (Protocol synopsis)
  5. Write data dictionary entries for every field you will report in Stage four.

Stage two: Running the registry

  1. Classify every record by tier and report the enrolled, confirmed, and tier counts together. (Verification tier form)
  2. Log every correction or query you need for the fields you will report, without changing a raw value. (Curation and definitions log)
  3. Reconcile the sample manifest against the export and record each problem. (Sample manifest and linkage log)
  4. Build the agenda for the weekly meeting held on the day of the export, with an owner and a date for every follow-up. (Weekly meeting agenda and minutes)
  5. Profile what is missing, write a hypothesis for each cluster, and draft the reporting statement. (Missing data plan)
  6. Plan the next year of contact, starting with the families whose follow-up is overdue or missing. (Outreach calendar)

Stage three: The community and the field

  1. Describe one interaction with the Epilepsy Syndrome X community, real in kind though invented in detail, that would change a decision in your protocol, and record what it changed. (Community engagement record)
  2. Prepare the briefing pack for a consortium that advises the FDA, including whether you could defend the seizure count field. (Briefing pack and meeting minutes)

Stage four: AI on top of all this

  1. Read the AI-written summary in Chapter 17 and correct every sentence that needs it.
  2. Fill in the AI use register for the Lantern Registry: which tasks AI may do, who reviews each one, and how the output is checked. (AI use register)

The report

Write a short report on the Lantern Registry as of the export of 15 March 2044, for the registry's board and for a sponsor considering the registry for a trial. It should give:

  • The enrolled, confirmed, and evaluable counts, each with its definition, and the tier breakdown.
  • Seizure frequency and developmental delay, each with its denominator and how it was computed.
  • Follow-up status, and what is known about why some follow-up is missing.
  • The state of the biorepository and its links to the registry.
  • What the registry would need for its data to be fit for the purpose the sponsor has in mind.
  • How AI was used in preparing the report, and who checked it.
Check your work: an example opening for the report

As of the export of 15 March 2044, the Lantern Registry holds 32 enrollment records for 31 children; one record, LR-012, is a duplicate enrollment of LR-005. Of the 31 children, 25 have a genetic report on file showing a pathogenic or likely pathogenic variant in GENEX. One has a variant of uncertain significance (Tier 2), one has a clinician letter only (Tier 1), and two rest on self-report or a pending result (Tier 0). One has a pathogenic variant in a different gene and is under review, and one does not meet the case definition. The genetic reports have not yet been reviewed by a named, qualified person, so no diagnosis is confirmed at Tier 3.

Seizure frequency at the most recent follow-up is evaluable for 15 of the 25 participants with a report on file: those with a seizure count free of placeholders and open queries, and a valid follow-up date after a valid enrollment date. Among them, the mean is 14.5 seizures a month and the median is 12. The seizure count field accepted free text, and its placeholder values were set aside and queried rather than averaged. Among the 25 participants with a report on file, all 24 with a known answer have developmental delay, and one answer is unknown.

Follow-up is overdue or missing for six participants. Four of them, all three participants in India and one of the two in Brazil, have no follow-up recorded at all, and no other country has a blank follow-up. That pattern points to how follow-up was offered in those countries rather than to chance, and the coordinators are investigating it before any analysis works around the gap.

Figure Value
Records, and children 32 records for 31 children
Tier 3, reviewed by a qualified person None yet
Report on file, awaiting review 25
Tier 2, Tier 1, Tier 0 1, 1, 2
Conflicting evidence, not a case 1, 1
Evaluable for monthly seizure frequency 15 (mean 14.5, median 12)
Seizure frequency read naively Mean 47.3, median 8, because placeholders are averaged in
Developmental delay, among the 25 with a report on file 24 of 24 with a known answer; 1 unknown
Overdue or missing follow-up LR-010, LR-011, LR-021, LR-022, LR-023, LR-024
Sample manifest entries with a problem S-0106, S-0107, S-0109 to S-0115, S-0118

Questions for reflection

  • Which of the problems in the export would the habits in Part One have prevented, and which would the weekly meeting have caught?
  • Which of those problems could an AI tool have found on its own, and which needed a person who knew what the data should look like?
  • If you took over the Lantern Registry, what would you change first, and who would you need in the room?

Up next: a closing letter. Continue to the key takeaways.