Estimate Review Screening Work Without Hiding Assumptions

Separate records, reports, and decisions

A search result is not necessarily a unique study. Deduplication can remove repeated records, and several reports may describe one study. Start the workload estimate with the unit entering each stage: title-and-abstract records, full-text reports, and disagreements requiring resolution.

Cochrane's searching and study-selection guidance explains the distinction between studies and reports and the importance of a planned selection process. A workload estimate should support that process without weakening eligibility checks simply to meet a deadline.

Measure a small pilot

Time a realistic set of screening decisions after reviewers understand the criteria. Record whether the timing includes opening the record, reading the abstract, assigning a reason, and resolving uncertainty. A speed measured on obvious exclusions may underestimate the time needed for a difficult topic.

Estimate title-and-abstract work separately from full-text retrieval, translation, review, and disagreement resolution. Record the number of independent reviewer passes required at each stage. One thousand records screened twice require two thousand initial decisions, before any reconciliation.

Calculate a fictional workload

Suppose 1,200 deduplicated records require two reviewer passes, averaging 30 seconds per record per reviewer. That first-stage workload is 1,200 × 2 × 30 / 3,600 = 20 reviewer-hours. If 120 full texts each require two passes averaging ten minutes, the second stage adds 120 × 2 × 10 / 60 = 40 reviewer-hours.

At an assumed 10% disagreement at both stages, there are 120 title-and-abstract disagreements and 12 full-text disagreements. At two minutes of total team resolution effort per case, reconciliation adds 4.4 hours, giving 64.4 team hours. These are invented assumptions, not productivity benchmarks. The screening-workload tool fixes two reviewer passes and rounds full-text and disagreement counts upward separately.

Convert effort into a realistic schedule

Reviewer-hours measure work, not elapsed calendar time. At eight total team hours available each working day, 64.4 hours equals 8.05 equivalent working days. Dependencies, uneven availability, and delayed full texts can extend that schedule. Retrieval, training, deduplication, extraction, and appraisal remain outside the tool's estimate.

Use the screening workload worksheet to record the pilot sample, timing method, pass counts, and capacity assumptions. Avoid multiplying by the number of reviewers twice: reviewer passes create work, while available reviewer time supplies capacity.

Update the estimate after the pilot and at a planned checkpoint. Compare actual throughput and full-text yield with the original assumptions, then revise the remaining work transparently. A lower-than-expected yield is not a reason to relax eligibility criteria, and a deadline is not evidence that a faster screening procedure retains adequate quality. Finish with a range of scenarios and a clear statement of which activities remain outside the numerical estimate.

Source checked on 2 November 2026; fictional workload independently recalculated.