Chapter 10: Track every sample and every link¶
In this chapter you will learn to:
- Keep a crosswalk from the registry identifier to every other identifier a participant acquires.
- Reconcile a sample manifest against the registry.
- Explain why a person reviews every return, even when code does the cataloging.
Case study: Every Vial Comes Back With a New Number¶
I work with a large multimodal registry, and much of its data go out and come back. Blood goes out in several vials, and each vial is tagged with a number. Even when we attach the participant identifier that corresponds to our master identifier, the lab often assigns its own, and different processing steps assign others. The proteomics and the metabolomics go to different labs, each with its own scheme. In the other direction, when a family sends us a PDF, it does not arrive with their identifier on it; someone has to tag it. Electronic health records do not arrive with our study identifier either, and a person seen in more than one health system or clinic may come back under several record identifiers that all have to be recognized as one person. Every one of these returns has to be logged and matched to the master identifier before it is any use.
Yes, there are ways to do this automatically, and we write code and use systems that catalog and track and merge. There is still no substitute for a manual quality-control pass with a careful editing eye, because so much can go wrong between the vial leaving and the result coming back. There is also no excuse for skipping it. A study with two or three thousand participants and a few hundred active at any time is small enough that a person can review every return and make sure nothing was lost along the way.
Lessons for registry leaders¶
Each lab, processing step, vendor, and health system assigns identifiers of its own. Code can catalog and merge them; a person reviewing each return is what confirms that nothing was lost or mismatched, and in a registry of a few thousand participants that review is achievable.
The rules¶
- Assign a master registry identifier that never changes and never encodes anything about the person, and keep a crosswalk from it to every other identifier the person acquires: lab identifiers, processing identifiers, vendor identifiers, and each health system's record identifier.
- Tag everything that arrives on the day it arrives. A PDF from a family, a record extract, and a lab result all come in without your identifier on them, and the person who receives them attaches it before the item goes anywhere else.
- If you collect biological samples, keep a chain of custody: who collected, when, where each vial is stored, under which consent, which lab it went to, which identifier that lab used, and which master identifier it maps to.
- Expect one person to have several record identifiers, from more than one health system or clinic, and resolve them to one person by a written rule, not by eye alone.
- Before merging or linking with another cohort, decide the matching rule in writing and test it on known cases.
- Record every link you make and every one you reject, with the reason.
- Review every return by hand. Code and tracking systems do the cataloging; a person does the quality-control pass on what came back, and a registry of a few thousand participants is small enough that there is no excuse not to.
- Check the links on a schedule; a sample with no participant and a participant with no sample are both errors to go and find.
- Say in every analysis whether sample data are linked, and how.
What AI can do, and what only a person can do¶
What AI can do¶
Propose candidate matches for review; check identifier formats; draft a sample manifest template.
What only a person can do¶
Confirm a match with the family or the lab; sign off on a linkage that will reach a publication or a sponsor.
Take-home¶
Comprehensive data means a person can be followed from form to sample to cohort, and the following is done by someone keeping a list.
Practice¶
Exercise¶
Open the sample manifest next to the synthetic Lantern Registry export. Every sample should trace back to one consented, enrolled participant. Find every sample that does not, and every entry that would make the trail unreliable. Record what you find in the sample manifest and linkage log.
Check your work
| Sample | Problem |
|---|---|
| S-0106 | Registered to LR-012, a duplicate enrollment of LR-005, so that child's samples are split across two identifiers |
| S-0107 | Registered to LR-033, which does not exist in the registry |
| S-0109 | Registered to LR-017, whose diagnosis is in conflict; the genomics result may help resolve it and should go to the reviewer |
| S-0110 | Identifier written as LR-19, which does not match the registry format |
| S-0111 | Lab B identifier PB-0533 is already assigned to sample S-0108 |
| S-0112 | Received by the lab before the recorded collection date |
| S-0113 | Registered to LR-026, who is not a case; decide whether the sample stays in the biorepository and how it is labeled |
| S-0114 | No consent version recorded |
| S-0115 | No result returned more than nine months after the lab received it |
| S-0118 | Collected before the participant's enrollment date, and no result returned |
Your workbook page¶
The Sample manifest and linkage log 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¶
A lab returns results under its own sample numbers. What lets you attach them to the right participant?
A crosswalk kept by the registry from each lab's sample identifier to the master registry identifier, checked by a person each time results come back.
Why review every return by hand when code can do the matching?
So much can go wrong between a vial leaving and a result coming back that a person's quality-control pass is what confirms nothing was lost or mismatched. A registry of a few thousand participants is small enough to make that review achievable.
Up next: Lists like these are only kept current if someone sits down and reviews them regularly. Continue to Chapter 11.