Reconcile Study Flow Before Reporting Follow-Up

Start with a participant flow map

Write the sequence of stages your study actually used: screened, excluded before allocation, randomized or enrolled, received intervention, followed, and analyzed for the specified outcome. Each stage needs a definition and a cutoff date. A count without a defined stage cannot reliably be reconciled with another report.

The CONSORT 2025 explanation of losses and exclusions calls for group-specific losses and exclusions with reasons. It also distinguishes intervention discontinuation from outcome follow-up. A participant can stop treatment and still contribute the primary outcome.

Choose a unique identifier and a snapshot

Build the audit from one stable participant identifier and a dated data extract. Assign each participant a status at each relevant stage rather than trying to reconstruct membership from aggregate percentages. Keep screening identifiers linked to randomized identifiers through an approved mapping when these differ.

Record the outcome and follow-up window beside the analyzed count. Someone can be analyzed for one endpoint and lack another. Avoid a single universal “completed” flag that obscures these distinctions.

Reconcile a fictional study

Suppose 150 people were screened, 30 were excluded before randomization, and 120 were randomized: 60 per arm. In arm A, 58 received the allocated intervention and two did not; 55 contributed to the stated analysis and five did not. In arm B, those counts are 57 and three, then 54 and six.

The study-flow reconciler checks whether these stage totals balance. These invented counts satisfy the displayed arithmetic, but they do not show why people were excluded or whether exclusion was justified. Treatment-receipt and analysis categories describe different dimensions; their counts should not simply be added together.

Investigate discrepancies at person level

Use the follow-up reconciliation worksheet to record each mismatch, source, and resolution. Check for duplicated identifiers, different data cutoff dates, late results, and people incorrectly counted as both excluded and analyzed for the same endpoint.

If several reasons apply to one participant, retain all relevant reasons but specify a mutually exclusive hierarchy when presenting a partition whose counts must sum. The calculator cannot detect overlap in aggregate reason categories. Arithmetic agreement is necessary for a coherent flow diagram, but it cannot prove completeness or absence of bias.

Finish with a dated reconciliation note linking the flow diagram to the analysis dataset. Keep unresolved discrepancies visible and update both outputs when their underlying definitions change. This workflow complements a variable-level missing-data audit by showing where participants move through the study.

Source checked on 31 October 2026; fictional arithmetic independently recalculated.