Pancreas Cohorts

A federated family of pancreatic study cohorts sharing one case-selection schema, one IHC database, and one review interface across roughly ten sub-studies.

A federated family of pancreatic study cohorts sharing one case-selection schema, one IHC database, and one review interface across roughly ten sub-studies.

What it is

Not a single dataset but a deliberately shared structure, which is the interesting part.

  • Case selection schema — organ = pancreas, with subclassifications held in a common YAML, so sub-studies select from the same frame rather than each defining its own.
  • Images — a de-identified image cohort on the Memorial WSI Share, explicitly not in git.
  • IHC database — shared and treated as the single source of truth. The documented rule is that IHC interpretations must not be re-run independently per sub-study.
  • Review interfacemorphology-research-pancreas-interface, used across the studies.
  • Per-study protocols — every sub-study repo carries its own protocol.md.

The sub-studies span PDAC (adrenal invasion, benign versus malignant, cytology, LN shell, large-duct variant), variant tumours (adenosquamous, solid-pseudopapillary), and ampulla/duodenum/ectopia (ampullary adenocarcinoma, ectopic pancreas, duodenal tubular adenoma).

The design decision worth noting: a case spanning two sub-studies is tagged in both but analysed once in its primary study, to avoid double-counting. That is a well-chosen rule — shared cohorts otherwise create silent dependence between “independent” published results, and few groups handle it explicitly.

Cohort sizes, date ranges, and scanner details are not recorded. [unverified]

Why it matters for my work

This is the group’s most mature example of shared research infrastructure, and it works: seven sub-studies are at recorded stages from pilot through to submitted manuscript, all off one curation effort. It is the model the other clusters could copy.

How it connects

Pancreas Morphology — the umbrella project page with per-study status.

Memorial WSI Share — where the image cohort lives.

Memorial LIS Extracts — the clinical data source behind case selection.

Survival Analysis Workflow — the analysis route for the outcome-based sub-studies.

Omentum — a downstream connection: pancreatic primaries account for many omental metastases, so the two cohorts overlap clinically.

Open questions

  • How many cases per sub-study? Not recorded. [unverified]
  • Is the shared IHC database versioned? If a re-read changes an interpretation, analyses run before and after would silently disagree.
  • The double-counting rule is sound, but is the case-to-primary-study assignment recorded anywhere auditable?