How to Create a Study Characteristics Table for a Systematic Review

Description is part of the review’s reasoning

Systematic reviews commonly place a characteristics table before or beside the synthesis. Its apparent familiarity can hide a demanding analytical task. Primary reports differ in terminology, reporting depth and organisation. The review must convert those descriptions into a consistent comparison while avoiding categories that are more precise than the evidence. The resulting table becomes a map of the evidence base: readers use it to assess populations, interventions, settings, designs, follow-up and outcome measurement before they consider pooled estimates.

PRISMA 2020 requires the characteristics of included studies to be cited and presented, and its explanation and elaboration identifies tables as a useful method for comparison.^2,3 This is a reporting obligation, not a fixed template. Cochrane and JBI methodologies specify relevant data according to review type and planned synthesis.^1,4 The table should therefore be designed from the review question, protocol and evidence structure rather than inherited from an unrelated published review.

The central tension is compression. A review may hold hundreds of extracted variables, but a readable table can show only those necessary for interpretation. Compression becomes defensible when the team declares the unit, field definitions, selection and aggregation rules, and location of omitted detail. Without those controls, the table can make heterogeneous studies appear uniform or exaggerate differences that arise only from inconsistent labels.

Start with the questions a reader must answer

Field selection should follow interpretive purpose. To judge applicability, readers may need age, condition definition, severity, setting, geography, recruitment period and exclusions. To understand intervention heterogeneity, they may need components, dose, intensity, duration, provider and co-interventions. To understand outcome comparability, they may need instruments, definitions, timing and analysis population. To evaluate the evidence context, they may need design, sample size, follow-up, funding and conflicts.

Not every review needs all these fields, and broad labels may be insufficient. “Usual care” can describe materially different comparators across settings. “Adults” can hide a wide age distribution or mixed population. “Twelve-week outcome” can describe a fixed visit, a range or the nearest available observation. Select details that define the causal or descriptive comparison and explain the intended interpretation in the table notes.

A field should be removed when it does not help a reader understand applicability, heterogeneity, study conduct or the planned synthesis and when its omission does not conceal an important limitation. Space pressure is not, by itself, a scientific selection rule. Important complexity can be moved to an appendix or structured supplement rather than eliminated.

Make the row unit scientifically consistent

One row per study is common, but it must be implemented as a study entity rather than a publication citation. Companion reports should be linked and combined into the study description. If one report supplies baseline characteristics and another supplies long-term outcomes, the row should retain both sources in the evidence model. A report-level row structure can inflate the apparent number of independent investigations and make the same participants appear repeatedly.

Some evidence bases need nested or repeated rows. Multi-arm trials may require arm subrows so that interventions, sample sizes and participant characteristics remain distinct. Crossover studies may need sequence and period detail. Diagnostic studies may require separate thresholds or index-test variants. Qualitative syntheses may use finding-level rather than study-level displays. The table caption and notes should state the row rule so readers do not infer independence where none exists.

Stable study identifiers permit the table to connect to the extraction database, risk-of-bias record and results. The published citation label may change during copyediting; the stable ID should not. This continuity is especially important when reviews are updated or when several reports share similar author-year labels.

Describe populations without manufacturing homogeneity

Population fields should reflect the review’s eligibility and effect-modification hypotheses. Record how the condition or phenomenon was defined, key eligibility boundaries, baseline severity, demographic distribution and relevant comorbidities or contextual features. Do not reduce a mixed population to a single category unless the rule and proportion support that summary. If subgroup data were extracted for analysis, distinguish them from whole-study characteristics.

Denominators require explicit meaning. “N=120” could mean screened, enrolled, randomised, treated or analysed. State which number the table reports and, if arm sizes are shown, whether they are baseline or outcome-specific. Attrition is often better presented in dedicated results or risk-of-bias material, but the characteristics table should not use a denominator that readers will misinterpret as the analysed sample.

Demographic reporting can be incomplete or use historical categories. Preserve reported terminology where accuracy requires it, and distinguish the source description from any review classification. “Not reported” is more accurate than inference from names, setting or photographs. If sex and gender are conflated in the report, state the limitation rather than silently selecting one construct.

Make intervention and comparator labels reproducible

Intervention descriptions should contain the features that determine what was actually compared. A name alone may hide components, dose, schedule, duration, delivery mode, provider, adherence support and co-interventions. Complex interventions may require a structured component framework or a linked appendix. The table should allow readers to distinguish meaningful variants without forcing them to consult every primary report.

Comparator descriptions deserve equal attention. “Control” is not a treatment. State placebo, no intervention, waiting list, attention control, usual care or active comparator and describe relevant components. Usual care can differ by country, site and time; where this variation affects interpretation, retain it. A comparative effect is defined by both sides of the contrast.

When several intervention arms are combined for synthesis, do not rewrite the original design as if one arm existed. Retain arm descriptions and explain the analytical combination elsewhere. The characteristics table may show separate arms or a clearly marked grouped display, but it should not conceal the multi-arm structure.

Distinguish outcome availability from outcome results

The characteristics table can show which outcomes were measured, the instruments used and the relevant follow-up periods. It should not duplicate detailed effect estimates that belong in results tables and plots. This division keeps the table focused on comparability while preserving the context required to understand why studies enter different syntheses.

Outcome labels should identify the construct and measurement. “Quality of life” may encompass instruments with different domains and directions. “Adverse events” may refer to any event, serious events, withdrawals or a specific harm. State the measurement frame used to judge whether studies address a common outcome. Where outcome availability is uncertain or only author supplied, use an explicit status.

Time points require a declared rule. If the protocol defines windows, show the selected observation or the available range consistently. Avoid reporting one study’s exact week and another study’s broad phrase without explanation. Differences in follow-up can be a source of clinical heterogeneity, not merely a formatting nuisance.

Let missingness remain information

Blank cells obscure whether information was absent, irrelevant, unclear or omitted to save space. Use distinct internal codes and define the reader-facing display. “Not reported in available sources” is a claim about the searched report set. “Not applicable” is a logical judgment. “Unclear” means the report contains information that cannot be interpreted confidently. These states can reveal patterns in primary-study reporting and should not be collapsed casually.

Author contact may resolve a field, but the value should retain its source status. If the clarification changes an important characteristic, a footnote can tell readers that the information was supplied after publication. The full contact event and correspondence remain outside the table in the controlled log.

Do not use “NR” without defining it. Abbreviations save space only when they do not force repeated decoding. A table note should define abbreviations, denominators, time windows, summary statistics, grouping decisions and any non-obvious source rules.

Treat table construction as controlled compression

The authoritative extraction dataset should be normalised enough to preserve entities and provenance. The publication table is a view generated from that dataset. Document how repeated fields are combined, how categories are ordered, which time point or denominator is selected, and what moves to an appendix. This separation prevents page-layout constraints from changing the underlying scientific record.

Ordering communicates emphasis. Alphabetical order is neutral with respect to effect size but can separate related designs. Grouping by design, intervention or setting may improve interpretation but should not imply a hierarchy of quality. Chronological order can show development over time. Choose and state the rule. Avoid sorting by observed result in a characteristics table because it can make a post hoc pattern appear structurally important.

Footnotes should resolve exceptions, not hold a second hidden table. If many cells need unique notes, the field may be too compressed or the row unit may be wrong. Consider arm subrows, multiple panels or a supplementary table. The goal is not minimum length; it is accurate comparison with manageable cognitive load.

Audit the table against neighbouring review products

Study IDs and counts should agree with the included-study set and PRISMA flow record. Design labels and outcome names should agree with the methods. Intervention groups and denominators should agree with the analysis dataset. Funding and conflicts should agree with risk-of-bias or other assessment records where those domains are used. Differences may be legitimate because products answer different questions, but they require an explanation.

Do not import risk-of-bias judgments into a descriptive field unless the table is explicitly designed to show them. Characteristics describe what was studied and reported; risk-of-bias assessments evaluate threats to validity under a separate method. Likewise, avoid placing pooled results in the characteristics table. Keeping these products distinct helps readers identify whether a claim is description, evaluation or synthesis.

Before publication, verify every row against the linked report set, test abbreviations and notes, and have an independent reader interpret sample size, timing and groups. A table that only its extractor understands is not publication ready.

Design for accessibility and reuse

Use true table headers, descriptive captions and a reading order that survives narrow screens and assistive technology. Do not use colour as the only signal. Keep units in headers or cells rather than relying on visual position. Avoid merged cells that make relationships difficult to parse. If the website requires horizontal scrolling, retain row labels and provide a downloadable structured file.

Reuse requires the data dictionary, stable IDs and machine-readable values. A PDF table may support reading but is a weak preservation format. Deposit or retain the underlying table in CSV or another non-proprietary form when permissions and policy allow. Remove unnecessary personal information and maintain the source-provenance record in the approved repository.

For living reviews, generate the table from a versioned dataset rather than editing an old display manually. Record additions, corrections and changes in categorisation. This keeps the published view aligned with the evidence base and makes update differences explainable.

Recognise when a single table is the wrong solution

Extreme heterogeneity can make one rectangular comparison misleading. If the evidence contains incompatible designs, fundamentally different intervention structures or many study-specific outcome definitions, use multiple tables, panels or a structured appendix. The decision should follow the interpretive task, not a convention that every review must have one master table. State the relationship between views and keep the stable study ID visible across them.

A characteristics table also cannot repair sparse reporting. When critical fields are absent, display that limitation and explain how it affects applicability or synthesis. Do not infer precision for the sake of visual completeness. The honest output may contain explicit uncertainty and a smaller number of comparable fields.

Conclusion: the table should make comparison more accurate, not merely easier

A strong study characteristics table uses a consistent scientific unit, includes fields that serve interpretation, defines denominators and time frames, preserves visible missingness, and declares how rich data were compressed. It remains connected to a fuller extraction record and does not trespass on the distinct functions of risk-of-bias assessment or results presentation.

The paired study characteristics table template provides a practical data dictionary and display builder. Adapt it to the review design and journal context. The adequacy of the table must be judged by whether an informed reader can understand who and what was studied, why studies differ and which uncertainties remain.

References

1. Li T, Higgins JPT, Deeks JJ. Chapter 5: Collecting data. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5. Cochrane; 2024. Access the current chapter.
2. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71.
3. Page MJ, Moher D, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ. 2021;372:n160. doi:10.1136/bmj.n160.
4. Aromataris E, Lockwood C, Porritt K, Pilla B, Jordan Z, editors. JBI Manual for Evidence Synthesis. Adelaide: JBI; 2024. doi:10.46658/JBIMES-24-01.
5. Buscemi N, Hartling L, Vandermeer B, Tjosvold L, Klassen TP. Single data extraction generated more errors than double data extraction in systematic reviews. J Clin Epidemiol. 2006;59(7):697-703. doi:10.1016/j.jclinepi.2005.11.010.
6. Jonnalagadda SR, Goyal P, Huffman MD. Automating data extraction in systematic reviews: a systematic review. Syst Rev. 2015;4:78. doi:10.1186/s13643-015-0066-7.
7. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst Rev. 2015;4:1. doi:10.1186/2046-4053-4-1.
8. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. 2nd ed. Chichester: Wiley; 2019.
9. Mathes T, Klaßen P, Pieper D. Frequency of data extraction errors and methods to increase data extraction quality: a methodological review. BMC Med Res Methodol. 2017;17:152. doi:10.1186/s12874-017-0431-4.
10. Li T, Vedula SS, Scherer R, Dickersin K. What comparative effectiveness research is needed? A framework for using guidelines and systematic reviews to identify evidence gaps and research priorities. Ann Intern Med. 2012;156(5):367-377. doi:10.7326/0003-4819-156-5-201203060-00009.