How to report participant flow in an observational study

A concrete task with RigorMed

1. Turn selection records into a flow-diagram drawing brief

   These are suggested task inputs and outputs to request, not a new execution, a customer outcome or a guarantee of completion.

   Materials to provide: The study design, prespecified eligibility rules, verified aggregate stage counts and exclusion reasons.

   Requested outputs / delivery goal: A flow-node and count checking table, drawing instructions and a caption draft.

2. An example request

   Organise flow nodes, denominators and exclusion reasons in the actual screening order. Check whether stage counts reconcile and draft drawing instructions and a caption. Without actual counts, retain placeholders rather than inventing numbers.

3. What still needs review

   The diagram's numbers must come from your verified outputs. The synthetic example on this page is neither actual patients nor a model-analysis result.

Four records to prepare before drawing

1. Define the source, study unit and starting population

   Record the data version, study period and initially assessable population. One row may represent a person, visit or measurement; raw row counts do not necessarily count participants. Decide how repeated records contribute according to the protocol, not by arbitrary deduplication for the figure.

2. Separate eligibility, file matching and participation

   Define inclusion criteria, then distinguish component eligibility, unmatched joins and actual non-participation. An unmeasured cross-sectional component, refusal, a missing field and interrupted cohort follow-up are different events. Do not label all of them lost to follow-up; retain unknown reasons when evidence is absent.

3. Fix the screening order and retain overlaps

   For each rule, record entering, newly excluded and remaining counts, plus its definition and execution basis. Sequential exclusions operate on those still present; separately retain overlap information for all reasons. Changing the order can change per-reason incremental counts, so disclose the order rather than choosing attractive numbers.

4. Distinguish enrolled and analysis samples

   Label study inclusion, the descriptive table, the main model and relevant subgroup or sensitivity samples separately. Report cases and controls separately in case-control studies; report follow-up stages where relevant in cohorts. Branches may have different denominators, but explain their source and reconcile methods, tables and text.

A synthetic flow with reconciled counts

Assume the protocol has justified the following selection rules and complete-case analysis. This demonstrates counting, not a recommendation to exclude missing observations automatically. Start with 1000 unique participants, not arbitrary rows.

Check the main path: 1000 − 80 − 120 − 40 − 45 = 715. Each incremental exclusion must be counted within the preceding remaining population. Real reporting also needs the definition and basis of each rule. The 120 here reflect measurement eligibility, not loss to follow-up.

Checks before adding boxes and arrows

Verify the numbers before styling:

• Are the starting population, data version, study unit and period clear?
• Can entering, incremental exclusion and remaining counts be reproduced for each rule?
• Are overlaps deduplicated or explicitly labelled, and is the order disclosed?
• Are study inclusion, Table 1, models and branches each assigned a defined denominator?
• Are ineligibility, unmatched records, missingness, refusal and loss to follow-up distinguished?
• Are only aggregate counts public, without claiming the figure removes selection bias?

Put remaining counts on the main path, exclusion boxes beside their stage and named analysis samples on branches. Attrition alone does not establish bias; reconciled counts do not prove its absence. Researchers must still assess selection mechanisms, missingness assumptions and generalisability.

Frequently asked questions