Extract Zero-Event Studies Without Turning Silence into Zero

Establish what the report actually says

A zero event count is a result; an unmentioned outcome is missing information. Search the relevant table, supplement, registry results, and corrections before deciding which category applies. Record the exact event definition, observation period, and population that was assessed.

Cochrane's rare-event meta-analysis guidance discusses computational problems and interpretation when one or both groups have no events. It also highlights the ambiguity when an adverse event is not reported at all. Do not resolve that ambiguity by entering zero without evidence.

Preserve counts and ascertainment

For each arm, extract the event count, relevant denominator, follow-up, and how events were sought. Distinguish people with at least one event from the number of recurrent events. Note whether all participants had comparable opportunity for detection or whether missing follow-up limits interpretation.

Keep “reported zero,” “not reported,” and “unclear” as distinct extraction states. A spreadsheet that stores all three as zero can create false reassurance and alter the synthesis before the statistical method is even selected.

Work through a fictional record

Suppose a fictional trial explicitly reports no events among 50 participants in each of two arms over twelve weeks. The extraction should preserve 0/50 and 0/50 with the twelve-week window and source location. It should not automatically claim identical population risks or a risk ratio of one: the observed ratio zero divided by zero is undefined.

For one arm alone, a two-sided 95% Wilson interval for 0/50 extends from zero to approximately 7.13%. The Wilson tool illustrates why no observed events do not prove no underlying risk. This individual-arm illustration does not select a valid meta-analysis method.

Separate extraction from analysis corrections

Use the effect-measure audit worksheet to record the original counts and any later analytic handling separately. Adding 0.5 or another continuity correction changes derived estimates; it must not overwrite the reported data or be treated as additional observed participants.

Different synthesis methods handle sparse or double-zero studies differently. Avoid an automatic rule that discards such studies from the review merely because one calculator cannot process them. Conversely, the fact that an absolute difference is numerically defined does not establish that a particular rare-event meta-analysis performs well.

Discuss the effect measure, assumptions, sample imbalance, and sensitivity analyses in the synthesis plan. Retain a list of studies excluded from a particular pooled calculation with explicit reasons, while preserving their descriptive information where relevant.

Finish with a source-verifiable record that another reviewer can reconstruct. The useful outcome is an honest account of what was measured and observed, not a complete-looking numerical table achieved by treating silence as evidence of absence.

Source checked on 3 November 2026; fictional zero-event interval independently recalculated.