Tables and Figures in Medical Papers: Common Mistakes

Tables and figures carry the evidence in a medical paper, and they are where small errors are easiest to see. Most can be caught in one careful pass before submission.

Checklist · 8 min read · Directive Publications Editorial Team

Who should do the final pass

Tables and figures are often built by several people, in different software, late in the writing. Each looks fine to its maker. The problems appear when they sit together: a sample size that differs between the abstract and Table 2, error bars defined in one legend but not the next, a blot panel whose original file nobody can find.

Give the final pass to one senior author, working from the analysis output and the original image files rather than from the draft. The sections below say what to look for, and the checklist at the end fits it into one sitting.

Mistake 1: the same numbers in three places

The ICMJE Recommendations on manuscript preparation are direct: do not repeat all the data in the tables or figures in the text, and do not duplicate data in graphs and tables. The text should summarize only the most important observations.

A working rule for each result:

A paragraph restating age, sex and smoking status for both groups, already in Table 1, can become one sentence citing the table.

The more damaging version is inconsistency: after a reanalysis, new numbers reach the table but not the abstract, or a percentage changes while its denominator does not. ICMJE asks for percentages to be given with the absolute numbers behind them, which makes mismatches easier to spot. Check every number that appears twice against the analysis output, not against another part of the draft.

Mistake 2: tables that cannot stand alone

Most table formatting for journals comes down to a short ICMJE list:

• Number tables in order of first citation, and cite every one in the text.
• Give each a short title a reader can understand without the text.
• Keep column headings short; put explanations and nonstandard abbreviations in footnotes. Footnote symbols vary, so check the instructions for authors.
• Say which measure of variation you report, such as standard deviation or standard error of the mean.

Two further points apply to trial reports. The CONSORT 2025 explanation and elaboration notes that standard errors and confidence intervals are inferential and are not appropriate for describing variability, so describe participants with the mean and standard deviation, or the median and percentiles for skewed data. It also states that significance testing of baseline differences is not recommended and should not be reported, so keep p values out of the baseline table.

Build tables with a table tool, not as pasted images or text aligned with spaces; ICMJE notes that tables which import directly into the journal's software avoid errors.

Mistake 3: figures that fall apart at print size

A figure that is clear on a monitor can be unreadable at column width. ICMJE asks for letters, numbers and symbols large enough to stay legible when the figure is reduced, because many journals do not redraw figures.

Figure resolution requirements vary by journal, so read the artwork instructions before exporting. As one example, the PLOS ONE figure guidelines ask for TIFF or EPS files at 300 to 600 dpi, with text in Arial, Times or Symbol at 8 to 12 point. Two of its points apply everywhere: a figure is only as sharp as its lowest-resolution element, and raising the resolution setting does not add detail. If a figure is too coarse, re-create it from the source data.

Then check what the figure says:

• Every axis has a label and a unit.
• Panels meant to be compared share a scale, or the legend says they do not.
• Bar charts start at zero, because bar length is the value.
• Groups are distinguished by shape or label, not color alone.
• Small samples show individual data points, since very different distributions can produce the same bar of means.

Mistake 4: legends that send readers back to the text

ICMJE asks for figures to be as self-explanatory as possible, partly because they are reused in slide presentations. Detailed explanations belong in the legend, not on the image, and every symbol, arrow or letter should be explained there. For the basic check of axes, sample sizes and error bars, see what reviewers wish authors knew about self-contained figures.

Figure legends in medical papers usually need four more things:

1. For photomicrographs, the internal scale and the staining method, both of which ICMJE asks authors to state.
2. For clinical images, what each arrow or arrowhead marks.
3. Any disclosure the image needs, such as rearranged lanes or a nonlinear adjustment.
4. A source and permission line for anything previously published.

Mistake 5: image edits nobody disclosed

Adjusting brightness or contrast is not in itself manipulation. PLOS ONE's guidance, like many journal policies, asks for adjustments to be applied consistently to the entire image and treats any adjustment that obscures, eliminates or misrepresents information, including background, as unacceptable. What crosses the line:

• Editing one region, or cleaning up a background with clone or erase tools.
• Combining bands from different gels, blots or exposures in one panel. PLOS ONE asks for separate panels, and for lanes rearranged within one image to be marked with dividing lines and explained in the legend.
• Cropping so tightly that background and non-specific bands disappear.
• Undisclosed nonlinear adjustments, such as gamma changes, which some journals require to be stated in the legend or methods.

ICMJE adds that before-and-after clinical photographs should use the same intensity, direction and color of light, and that editors may ask for original blot photographs. Keep unprocessed files where every author can reach them; the lab habits behind this are covered in research integrity responsibilities of a lab lead.

Mistake 6: patients who can be identified

The ICMJE guidance on protecting research participants says identifying information, including names, initials or hospital numbers, should not appear in photographs unless it is essential for scientific purposes and the patient, or a parent or guardian, has given written informed consent for publication, after being shown the manuscript. It also states that masking the eye region is inadequate protection of anonymity, and that de-identification must not distort scientific meaning.

Check scans, screenshots, slide labels and pedigrees for nonessential identifying details and remove them. If a patient could still be recognized, use the consent form your target journal requires; requirements vary by journal and jurisdiction.

Mistake 7: reused figures without permission

ICMJE says a previously published figure needs an acknowledgment of the original source and written permission from the copyright holder, irrespective of authorship or publisher, unless the material is in the public domain. That includes your own earlier figures. Tables built from another source's data also need permission and full acknowledgment.

Open access adds a step. The PLOS ONE licensing policy, for example, requires written permission to publish third-party content under CC BY 4.0 when the original licence does not already allow that use. A permission covering print reproduction alone may not be enough, so tell the rights holder which licence your article will carry.

Mistake 8: flow diagrams that do not add up

Some figures come with the study design. CONSORT 2025 asks trial reports for participant flow, including a flow diagram, giving for each group the numbers randomly assigned, receiving the intended intervention and analysed for the primary outcome, plus losses and exclusions with reasons. Systematic reviews use the PRISMA 2020 flow diagram of records identified, included and excluded. Add up every box, then check the totals against the baseline and outcome tables. A reviewer can do that arithmetic in a minute.

The final-pass checklist

1. Tables and figures cited in order; file names match caption labels
2. No result shown in full in both a table and a graph
3. Every repeated number, including denominators, matches the analysis output
4. Titles and legends make sense without the text
5. Abbreviations, symbols, error bars and tests defined
6. No standard errors describing participants; no baseline p values in trials
7. Text legible at column width; files meet the journal's resolution rules
8. Adjustments applied to whole images; splices and nonlinear changes disclosed
9. Unprocessed originals stored where every author can reach them
10. Written consent for identifiable patient images
11. Permission in hand for previously published figures and table data
12. Flow diagram totals match the tables

If you find one of these problems after publication, the route is a formal correction, not a quiet file swap; see when and how to correct the published record. Directive Publications collects its author guidelines and licensing policies on its for-authors page; read them before you submit your manuscript.