Study Characteristics Table Template for Systematic Reviews

A descriptive table can quietly determine how readers understand the evidence

The study characteristics table is often the first place where readers evaluate the evidence base. It shows who was studied, where, under what design, which interventions or exposures were compared, and what outcomes and follow-up periods were available. PRISMA 2020 item 17 requires authors to cite each included study and present its characteristics; the explanation and elaboration notes that tabular presentation facilitates comparison.^2,3 The requirement does not prescribe a universal column set, because relevant characteristics depend on the review question and design.

A table can mislead even when every cell is copied accurately. One row per article can count companion publications as independent studies. A single “sample size” column can mix randomised, enrolled and analysed denominators. An intervention label can hide dose, intensity or co-interventions. A blank cell can mean not reported, not applicable or not extracted. Compression therefore requires explicit rules and a richer source record behind the display.

The table should answer a defined reader question: what features are necessary to judge applicability, compare the studies and understand the planned synthesis? It should not become an indiscriminate catalogue. Every column should have a declared meaning and source, and every row should represent the chosen unit consistently.

Choose the row unit before choosing the columns

For many intervention reviews, the main row represents a study, with arm information nested inside the row or presented on separate lines. Other designs may require a cohort, comparison, diagnostic threshold or qualitative finding as the meaningful display unit. The unit should follow the scientific comparison, not the publication format. Multiple reports can contribute to one row, and one complex study may legitimately require more than one display row if the table states the rule.

Keep stable study IDs in the internal dataset (see the data extraction form template) and retain the linked report citations. If a row draws population information from one report and outcomes from another, that provenance should remain recoverable. The publication table can cite the study’s report set or principal report according to journal style, but internal linkage must not be lost.^1,8

Select characteristics by interpretive purpose

Core domains often include design, setting, recruitment period, eligibility, population, sample size, intervention or exposure, comparator, outcome measurement, follow-up and funding. The exact fields must be adapted. Cluster trials may require the number and type of clusters; crossover trials require sequence and period information; diagnostic reviews require index-test and reference-standard details; prevalence reviews require sampling frame and case definition. A field is justified when it informs applicability, heterogeneity, bias assessment or synthesis interpretation.

Define denominators. If the table displays total sample size while analyses use outcome-specific denominators, say which number appears and where attrition is shown. Separate arm-level information when interventions or populations differ. Do not combine distinct doses or components into a single label merely to fit the column width. Concision must not change the exposure that readers believe was studied.

Outcome columns should describe what was measured and when, not reproduce every numerical result. Results belong in results tables, forest plots or synthesis text. The characteristics table provides the measurement context needed to interpret those results. State instruments or operational definitions when they materially affect comparability.

Make missingness and uncertainty visible

Use explicit display labels such as “not reported”, “unclear from available reports” and “not applicable”, backed by distinct internal codes. Avoid a dash unless its meaning is defined in a note. A blank can make incomplete extraction indistinguishable from absent reporting. When information is author supplied, identify that status in the internal record (track these interactions using the missing data contact log and request templates) and consider a footnote if it materially changes interpretation.

Reported study language should not be silently modernised into a more precise category. Preserve the source description and, where needed, add the review’s operational classification. If sex and gender terminology is inconsistent or setting is only indirectly inferred, state the evidential limitation rather than presenting an unqualified category.

Compress by rule, not by convenience

A publication table is narrower than the extraction dataset. Declare which fields are combined, which values are selected when several exist and what is moved to an appendix. Repeated intervention components can be organised using a standard structure; complex eligibility criteria can be summarized with a link or appendix; detailed provenance remains in the audit record. The table should not be the only surviving copy of characteristics data.

Test the draft with a reader who was not involved in extraction. Ask whether rows can be compared, units are visible, abbreviations are defined and footnotes resolve rather than create ambiguity. Accessibility also matters: avoid meaning conveyed only by colour, use real table headers, keep notes close to the affected cells and provide a downloadable structured version where feasible.

Predefine aggregation rules when a field contains several legitimate values. A multicentre study may span countries and settings; an arm may contain several intervention components; eligibility may be broader than the analysed population. Listing every value can overwhelm the display, whereas choosing one can misrepresent the study. Use a structured summary, arm subrows or an appendix and preserve the complete values in the source dataset. The table note should tell readers which reduction was applied.

Ensure that classifications remain separate from observations. For example, the report may describe care as “standard treatment”, while the review classifies it as an active comparator after examining its components. Store both statements. The display can use the review classification when it serves comparison, provided the source description and coding rule remain recoverable. This distinction supports later recoding and prevents a reviewer-created category from being presented as a direct quotation from the study.

Use the builder as a publication-planning surface

The browser builder creates a concise row and export. It does not store multiple reports, arm-level repeated records or immutable provenance. Create the authoritative dataset first, then populate this view using declared selection rules. The accompanying Word template includes a data dictionary, study-level table, arm-level table, outcome-context table and display-planning notes so that the published view remains connected to its source.

Before publication, reconcile the displayed study count with the included-study set, check every citation and study ID, verify denominator meanings, define all abbreviations and compare the table against the methods and synthesis. A polished table does not validate the extraction; it makes the evidence architecture visible to readers.

Verify the generated view against its source records

Table verification should sample neither only complex studies nor only convenient ones. Check every stable study ID and citation, then target fields whose compression can change interpretation: denominators, intervention components, follow-up windows, mixed populations and author-supplied characteristics. Compare the displayed value with the underlying study- and arm-level records and confirm that footnotes explain exceptions. If the table is generated by code, test the mapping and sorting rules with known cases.

Record the table version used in the manuscript. Copyediting can alter abbreviations, alignment or notes in ways that affect meaning, so compare the final proof with the verified source. A reproducible export helps, but human review remains necessary for semantic errors that pass structural checks.

Resolve common display conflicts explicitly

If one study reports baseline age for all participants and another reports age separately by arm, choose whether the table will show a total or arm-specific value and apply that rule consistently. If only medians are available for some studies, do not label the column “mean age”; identify the summary statistic in the cell or create separate fields. When recruitment spans several countries, decide whether geography or healthcare setting is the more relevant comparison and retain the complete description in the underlying record.

Complex interventions require similar discipline. A short intervention name may be useful for navigation, but components, intensity and duration determine comparability. Use a concise label plus a structured detail field or appendix. For comparators, avoid a single “control” category when usual care, waiting list and active treatment differ. If the table groups them for space, the grouping rule and original description must remain visible.

Where a source does not report a characteristic, display the defined missingness label rather than borrowing information from an ineligible or unrelated report. If the team infers a classification from several sources, mark it as a review interpretation and cite the evidence. These controls prevent visual regularity from being mistaken for empirical uniformity.

Hand off the table with its interpretation contract

Deliver the publication table together with the field dictionary, stable study IDs, abbreviation list, selection and aggregation rules, source dataset version and unresolved exceptions. This contract allows a co-author, editor or update team to change presentation without changing the meaning. If journal formatting forces a field to be shortened or moved, confirm that the rule still applies and that the omitted detail remains accessible in a supplement governed by the overall systematic review data extraction and data management plan.

Keep one verified, machine-readable table as the authoritative publication view. Avoid parallel copies in email, slides and word-processing files that can diverge. A change to a denominator, group label or footnote should propagate from the controlled source and create a new version. The publication proof then becomes a checked rendering of the evidence view rather than an independent dataset.