Audit the numbers in your abstract before submission

1. Inventory every numerical claim

Highlight sample sizes, event counts, percentages, dates, follow-up durations, effect estimates, intervals and P values. For each, record the exact result source, analysis population, outcome definition, time point and unit. A number copied from an early table may no longer describe the analysis used in the revised manuscript.

ICMJE recommends reporting absolute counts alongside percentages and uncertainty with estimates rather than relying on P values alone. Use that principle to check what a reader can reconstruct from the abstract. Do not add an interval or a statistical test that the responsible analyst has not verified.

Reference: ICMJE: manuscript preparation

2. Separate eligibility, observation and events

In a fictional cohort, 180 people were eligible for follow-up, the specified outcome was observed for 150, and 48 of those 150 had the event by the planned time point. Outcomes for the remaining 30 are unknown in the supplied record. The observed event proportion is 48/150 = 32%. Dividing by all 180 gives about 26.7%, but would treat the unknown outcomes as non-events if presented as the event proportion.

The abstract should state the observed denominator and the missingness. Whether the study requires a different analysis population, weighting, imputation or sensitivity analysis is a methodological question for the study team. The writing exercise cannot settle it. Reporting an available-case proportion accurately does not make that analysis sufficient.

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

3. Distinguish absolute and relative changes

Consider a separate fictional comparison with an observed event risk of 12/100 in group A and 20/100 in group B at the same time point, with complete outcome observation for both groups. The absolute difference A minus B is −8 percentage points. The risk ratio A/B is 0.60, corresponding to a 40% lower observed risk relative to B. These are different descriptions of the supplied numbers.

Do not call the difference an 8% relative reduction or turn an odds ratio into a risk ratio. Name the measure, reference group and direction. This example supplies no design, adjusted analysis or confidence interval, so the arithmetic does not establish a causal effect, statistical significance or the precision of the comparison.

4. Keep uncertainty and analysis status visible

Copy an estimate and its interval from the same approved model and analysis population. Check whether it is adjusted or unadjusted and whether the outcome and time point match the abstract sentence. A confidence interval from a subgroup cannot be attached to the overall estimate. If the interval is unavailable, ask the analyst to resolve that gap rather than inventing a plausible range.

Distinguish a prespecified principal result from an exploratory analysis. Do not replace the main result with whichever subgroup looks most favourable. If the abstract must mention a secondary finding, label its role and retain the main result and material uncertainty. Rounding should follow a consistent rule that does not reverse the apparent relationship.

5. Check units, timing and Arabic number placement

Confirm that a mean is not relabelled a median, a standard deviation is not called a confidence interval, and a count of visits is not described as a count of people. A participant can contribute more than one encounter; the distinction matters even when the arithmetic is correct. Name the period for rates and the scale for questionnaire scores.

In the Arabic version, read each numerical phrase aloud and inspect it in the actual layout. Verify that a minus sign, slash or interval endpoint has not moved visually in a confusing way. Compare “ثماني نقاط مئوية” with “انخفاض نسبي بنسبة 40%”: they describe different measures in the fictional comparison. The English and Arabic versions must preserve the same reference group.

6. Sign off a single results reconciliation

Use a table with columns for abstract wording, manuscript table, analysis output, discrepancy, responsible author and resolution date. Resolve discrepancies at their source. If the abstract is wrong, correct it; if the table or analysis is wrong, have the responsible author make and document the appropriate correction before recopying the number.

Repeat the check after revision and at proof stage, including the abstract entered directly into the submission system. Keep the final approved wording with its supporting output version. A clean reconciliation record establishes consistency of reporting; it is separate from statistical review, data verification and the scientific interpretation of the study.

Questions for a practical final check

• Every number has a result source and time point.
• The denominator is named and missing outcomes are not counted as zero.
• Absolute and relative measures are distinguished.
• Estimates and intervals come from the same verified analysis.
• Units and signs survive Arabic and English rendering.
• The final abstract, tables and metadata reconcile.