Review Table Shells Before Filling Them with Results

Start with the question the table answers

A table shell specifies the structure and meaning of a planned output before the final results are inserted. Give it a clear purpose, analysis population, outcome or variable set, and time window. A polished grid is not useful if different readers infer different denominators from the same column heading.

The ICH E9 statistical principles describe detailed analysis planning and the planned summarization of study data. A table-shell review turns those decisions into an output specification that analysts and scientific reviewers can inspect together. It should not introduce a new analysis simply because a column looks empty.

Define each cell's ingredients

For every row and column, specify the source variable, eligible population, statistic, unit, and missing-data convention. Separate the number with an observation from the total eligible number when they differ. For categorical summaries, state whether percentages are calculated within rows, within columns, or against another defined denominator.

Describe the presentation of continuous measures, including the chosen spread statistic and quantile method where relevant. A heading such as “value” does not distinguish mean with SD from median with quartiles. Link the shell to the analysis specification instead of relying on a reviewer to infer the intended calculation.

Test a fictional denominator problem

Imagine a planned column for 60 randomized participants, of whom 54 provide the week-12 measurement and 18 meet a response definition. Eighteen divided by 54 is 33.3%; eighteen divided by 60 is 30.0%. The arithmetic does not determine which denominator answers the planned question.

Label the denominator and missing-outcome treatment explicitly. If the analysis uses imputation or a model, a simple observed-case percentage may not represent the primary estimate. These numbers are fictional and should appear only as clearly labeled test values, never as apparently observed study results.

Review interpretation before formatting

Check that effect estimates have their direction, reference group, units, and uncertainty identified. Confirm that adjusted results name the adjustment set or point to the relevant methods. Distinguish descriptive summaries from model-based estimates and from exploratory comparisons.

Do not add p-values to every row automatically. The planned purpose of a baseline description differs from a table of primary treatment effects. Where multiplicity or hierarchical testing applies, ensure the table and footnotes communicate the actual plan rather than an improvised collection of significance markers.

Use deliberately varied synthetic test values to check long labels, negative numbers, zero counts, missing entries, and wide intervals. Then inspect the rendered table for truncated footnotes, ambiguous symbols, and inaccessible reading order. Visual clarity should reinforce the statistical definitions.

Finish with a reviewed shell identifier, version, source mapping, and unresolved decisions. Keep changes traceable when outcomes or populations are revised. A successful shell lets another analyst explain how every displayed number will be produced before the real data are inserted, reducing avoidable disagreement after results become visible.

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