Reporting Cohort Flow and Sample Availability

Why Cohort Flow Is Part of the Analysis

A validation result is interpretable only if readers can determine which participants contributed to it and how that set was formed. A report that gives a final sample size alone may conceal substantial exclusions, unavailable assays, or differences between the discovery and validation analysis sets. In multi-omics studies, a participant can have clinical data and one assay but lack another; therefore, “sample size” may refer to several different counts.

A cohort-flow summary is a transparent accounting of participant counts from identification through the analysis population. It is descriptive, not an inferential model: its arithmetic should reconcile, and each count should have a defined unit and denominator. The earlier material on Assessing Cohort Eligibility and Comparability addresses which participants are appropriate for a validation claim; here the emphasis is on showing the resulting flow and availability clearly.

The unit should ordinarily be the participant, not the specimen, assay file, aliquot, or measurement. If one participant contributes multiple specimens, report participant counts for the flow and explain separately how specimens were selected or aggregated. Decide and state what “available” means: for example, a quality-controlled assay result that passed the prespecified processing rules, rather than merely a specimen having been collected.

Structure and Accounting Rules

A useful flow has sequential stages: records or participants identified, duplicates resolved, eligibility assessed, eligible people invited or approached, enrollment or inclusion, and the final analysis sets. At each stage, specify the starting count, the number and reasons for removal, and the resulting count. Reasons should be mutually exclusive within a partition: each participant removed at that stage is assigned once, using a stated priority rule if more than one reason applies.

Here \(E_j\) is the count assigned to exclusion reason \(j\) at that particular stage. This identity is appropriate only when those reason categories are disjoint and exhaustive for the people removed at that stage. A later-stage count must not be added to an earlier-stage count as though they were separate groups: they represent participants who passed through one stage and then another, not distinct people in a single partition.

For each modality, report both its count and the denominator used to calculate availability. A common descriptive rate is the proportion of the enrolled cohort with usable data for modality \(m\):

where \(n_m\) is the number with usable data for modality \(m\). If availability is instead calculated among an eligible subset, state that denominator and why it is the relevant target. Marginal availability rates do not determine the number with all required modalities. That joint count must be tabulated or computed from participant-level availability patterns.

Before interpreting any count, check the cohort definition, participant identifier, deduplication rule, eligibility date, assay usability rule, and data-freeze date. Counts can differ legitimately when they use different rules or cutoffs, but the rules must be stated. A flow diagram, a table, or both can be used; the format is less important than clear stages, reconciled arithmetic, and analysis-specific denominators.

Worked Cohort Flows

Joint Availability and Analysis Denominators

A modality-by-modality table is useful for marginal counts, but it cannot reveal all joint patterns. If three modalities are required together, counts for each modality can be high while the complete-case set is appreciably smaller. A pattern table makes that reduction visible and helps identify which missingness combinations are responsible.

Derivation, Diagnostics, and Reporting Choices

The key accounting principle follows from a partition. If every participant in a stage is assigned to exactly one of several disjoint outcomes, the total count equals the sum of those outcome counts. For example, when the outcomes are “excluded for reason 1,” “excluded for reason 2,” or “remaining,” every participant appears once, so summing those categories recovers the stage total. If a person can be counted under two reasons, the categories overlap and direct addition overcounts; resolve this by defining a primary reason or report overlapping reasons separately without presenting them as a partition.

Likewise, joint availability cannot generally be recovered by adding marginal counts. For two modalities \(A\) and \(B\), the participant-level count with both is governed by inclusion-exclusion:

The union count is needed, so knowing only \(n_A\) and \(n_B\) is insufficient. For three or more modalities, the availability pattern table or direct joint count is usually clearest. State whether the final analysis set requires complete data on all listed variables, permits a specified missing-data method, or uses another prespecified rule; the denominator should reflect the actual procedure.

Several checks catch common failures before publication:

• Recalculate each sequential stage independently; a stage total should equal its mutually exclusive removals plus the remaining count.
• Check that unique participant identifiers, rather than specimen counts, underlie participant-flow totals.
• Verify that each assay count uses the stated quality-control and data-freeze rules.
• Confirm that joint counts cannot exceed any required marginal count, and that analysis counts do not exceed their eligible or available-data parent sets.
• Compare discovery and validation denominators only after confirming that the count definitions and reporting rules are aligned; otherwise, label the difference rather than implying a like-for-like comparison.

A flow diagram is effective for sequential screening and exclusions. A modality matrix or pattern table is better for showing overlapping availability, and an analysis-set table is useful when different models require different combinations of data. In practice, a concise flow diagram paired with a table of modality counts and analysis denominators often communicates the full structure. The limitation is that counts alone do not explain why data are unavailable or establish that missingness is harmless. Report reasons and timing where known; do not interpret a lower availability rate as evidence of a biological difference without an appropriate analysis.

A frequent mistake is to add a later-stage count to the exclusion counts at an earlier stage. For example, an enrolled count is already the remainder after eligibility exclusions and refusals; adding it again to a count of people eligible at an intermediate stage mixes sequential stages and double-counts participants. The correction is to show each conservation equation separately and label the stage to which it applies. A second common error is to report assay counts as if they were analysis sample sizes. Instead, name the exact variables and rules required for every analysis denominator.