To describe a cohort in plain language, state who belongs in it, who does not, which data you need for each patient, over what time period, and which linked data types must come from the same patient. Then mark each element as required or preferred. A description built this way can be searched across institutions and answered with a count instead of a conversation.

This guide covers requests for clinical data and linked imaging. For physical material such as FFPE blocks and unstained slides, see what to include in a biospecimen request.

What is a cohort definition?

A cohort definition is the set of rules that decides which patients are included in a study population. It turns a research question into criteria a system can check against records.

“Patients with colorectal cancer” is a topic. “Adults diagnosed with stage III colorectal adenocarcinoma between 2015 and 2022, with documented treatment and at least two years of follow-up” is a cohort definition. The second can be counted.

How do inclusion and exclusion criteria differ?

Inclusion criteria are the characteristics a patient must have to enter the cohort. Exclusion criteria remove patients who otherwise qualify.

Write each criterion as one testable statement. Name the diagnosis, the stage or grade, age limits, and any molecular marker that defines eligibility. Then list exclusions, such as prior treatment, a second primary cancer, or missing consent for research use. Staging should name its system, for example the AJCC staging system, because stage definitions change between editions.

Which fields are required, and which are preferred?

Required fields are the data without which a record cannot be used. Preferred fields add value but do not exclude a patient when missing.

Separate them explicitly. If treatment history is essential but smoking status is useful, say so. A network can then report how many patients meet the required set, and how many of those also carry the preferred fields. This keeps a strict requirement from silently shrinking a cohort to zero.

Name the coding you expect where it matters. Clinical networks often map data to shared models such as OMOP or CDISC. Saying “diagnosis as ICD-10 code” is clearer than “diagnosis.”

Why does the time window matter?

The time window is the period during which events must occur for a patient to count. It shapes both the size and the quality of a cohort.

Specify three things: the diagnosis period, the minimum follow-up after diagnosis, and when key measurements must fall relative to an index date. The index date is the anchor event for each patient, such as the date of diagnosis or of surgery. A request for “outcomes” without a follow-up window can return patients diagnosed last month.

How do you request linked data types?

Linked data means records of different types that belong to the same patient: clinical data, pathology reports, whole-slide images, molecular results, and outcomes. A whole-slide image is a high-resolution digital scan of a glass pathology slide.

State which types must be linked and which can stand alone. A model-development project may need every image paired with a diagnosis and outcome. A descriptive study may accept clinical data alone. Our guide to linking FFPE blocks, slides, and clinical data explains how those records are matched.

What does a complete description look like?

The following description is illustrative:

Stage III colorectal cancer: FFPE blocks with matched whole-slide images and clinical data.

That sentence is a reasonable starting point. Expanded into criteria, it might read:

  • Include: adults with stage III colorectal adenocarcinoma (AJCC 8th edition), diagnosed 2015 to 2022.
  • Exclude: neoadjuvant treatment before surgery; a second primary cancer.
  • Required, linked per patient: FFPE block from the resection, an H&E whole-slide image of that block, diagnosis, stage, and treatment.
  • Preferred: microsatellite instability status, recurrence, and overall survival with at least three years of follow-up.
  • Quantity: the number of patients, not slides or files.

Each line can be checked against a record. Each line tells a supplier what matters most.

What happens after you submit a description?

The next step is feasibility. For this patient-based example, a feasibility count is the number of patients at each participating site who meet the criteria, returned before any data changes hands. Our article on what a feasibility count is covers how to read one.

At Biobase, a researcher describes the cohort in plain English or as precise criteria. Agents check partner sites and return counts and attributes without any data changing hands. Governance, review by an institutional review board (IRB) where the study requires one, and de-identification are handled in the ordering workflow, described in from request to delivered cohort.

Write the question before the criteria

A clear research question produces clear criteria. If the criteria feel hard to write, the question may need narrowing first. Start with the decision the data must support, then work back to the patients and fields that decision depends on.