How Unintentional Bias Weakens Medical and Scientific Writing

The EQUATOR Network — Enhancing the QUAlity and Transparency Of health Research — maintains a library of more than 500 reporting guidelines for different study designs and research types. The CONSORT statement (Consolidated Standards of Reporting Trials) for randomized controlled trials, the STROBE statement (Strengthening the Reporting of Observational Studies in Epidemiology) for observational studies, and the PRISMA statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) are among the most widely adopted. Their adoption by major medical journals is nearly universal.

These reporting guidelines exist because bias in the presentation of research results — not only in the conduct of the research — is a significant and documented problem. A study that is conducted without bias can be presented in a way that misrepresents its findings. Conversely, a study with known limitations can be reported in ways that contextualize those limitations honestly and give readers the information needed to assess the evidence appropriately. The guidelines formalize what unbiased, complete, transparent reporting looks like — and peer reviewers use them, explicitly or implicitly, to evaluate every manuscript they assess.

Confirmation Bias in Framing

Confirmation bias in scientific writing is the tendency to present results, select citations, and frame conclusions in ways that support the hypothesis the authors began with, while minimizing or omitting contrary evidence. It is rarely intentional and often undetectable by authors who are deeply familiar with the field — they genuinely do not notice the asymmetry in how they've framed the evidence because the framings that favor their hypothesis feel like accurate descriptions of the literature.

Peer reviewers, approaching a manuscript from the outside, are positioned to see this asymmetry more clearly. Specific patterns that reviewers flag:

• Selective citation of supporting evidence — citing studies that support the authors' hypothesis while failing to cite equally well-conducted studies with contrary findings
• Asymmetric emphasis in results sections — presenting positive findings with detailed description and analysis while presenting negative findings briefly, in supplementary materials, or framed as "unexpected"
• Introduction framing that presents the authors' position as consensus — characterizing an ongoing scientific debate as settled in the direction that supports the current study's hypothesis
• Post-hoc analysis presented as pre-specified — reporting exploratory analyses in the same language and emphasis as the pre-specified primary outcomes

CONSORT 2010, the current version of the randomized trial reporting checklist, specifically requires that primary and secondary outcomes be defined as pre-specified or post-hoc, that the flow of participants through the trial be reported for all groups, and that both significant and non-significant results be reported with appropriate statistical detail. A manuscript that does not meet these requirements is not fully CONSORT-compliant, which many journals require as a condition of acceptance.

Citation Bias and Its Consequences

Citation bias — the systematic tendency to cite research that supports a particular conclusion — is related to confirmation bias in framing but has specific structural manifestations in scientific writing. Research published in PLOS ONE and BMJ has documented that trials with positive results are cited more frequently than trials with negative or null results, and that systematic reviews citing a narrow, non-representative body of primary literature produce conclusions that overstate the certainty of the evidence.

The practical consequence of citation bias in a manuscript is that the literature review and discussion sections present a picture of the evidence base that is systematically more supportive of the authors' conclusions than the actual literature supports. Reviewers who are themselves familiar with the field will notice which studies are not cited and may question why they are absent.

The remediation is methodological rather than a writing fix: a systematic search of the literature that includes defined search terms, defined databases, defined date ranges, and defined inclusion and exclusion criteria — documented transparently enough that another researcher could replicate the search. This is the literature review standard for systematic reviews; it is increasingly expected for primary research introductions as well, at least at higher-impact journals.

Reporting Bias in Results Presentation

Reporting bias in results sections is distinct from the design or conduct of the study — it is a writing phenomenon. Statistical reporting choices that constitute reporting decisions include: whether to report absolute or relative risk, how to define and present subgroup analyses, which timepoint to use for time-to-event outcomes, and how to describe findings that trend toward significance without reaching a pre-specified threshold.

The absolute vs. relative risk distinction is among the most consequential. A treatment that reduces a three-year mortality rate from 2% to 1% can be described as reducing mortality by 50% (relative risk reduction) or by 1 percentage point (absolute risk reduction). Both statements are accurate. Only one provides the information needed to assess clinical meaningfulness. The National Number Needed to Treat (NNT) — in this case 100, meaning 100 patients must be treated to prevent one death — is the number that most clearly represents clinical impact, but it appears in a minority of trial reports.

The STROBE statement for observational studies, CONSORT for randomized trials, and TRIPOD (Transparent Reporting of a Multivariable Prediction Model) for prediction models all require complete, non-selective reporting of pre-specified outcomes. Journals that require EQUATOR checklist completion at submission use the checklist as a systematic instrument for identifying reporting gaps before peer review begins — which means reporting gaps identified at that stage result in editorial desk rejection before the manuscript reaches a reviewer.

Gender, Racial, and Cultural Bias in Clinical Research Reporting

NIH's policy on Sex as a Biological Variable (SABV), implemented in 2016 and reinforced in subsequent funding policy, requires that NIH-funded research study both sexes in cell and animal studies, and that clinical research be designed to support analysis of sex differences in outcomes and responses. The policy reflects a documented historical pattern: clinical trial populations have been predominantly male, and treatment effects established in male-predominant populations have been extrapolated to female patients without evidence.

Reporting of trial participant demographics, required by CONSORT, includes sex and age and increasingly includes race and ethnicity. Manuscripts that do not report these demographic characteristics, or that report them without discussing their implications for generalizability, are manuscripts with reporting gaps that journals increasingly require authors to address.

The generalizability question extends to the Discussion section, where authors are expected to describe the limitations of their findings — including population-specific limitations. A trial conducted in a predominantly White, high-income, academic medical center population that claims broad applicability of its findings without discussing this limitation is a manuscript making a generalizability claim that the data do not support. Reviewers familiar with health disparities research will identify this gap.

How the EQUATOR Guidelines Apply to Manuscript Assessment

The EQUATOR Network's reporting guidelines are organized by study design. The most widely used are:

• CONSORT — for randomized controlled trials, including a 25-item checklist and flow diagram requirement
• STROBE — for observational studies (cohort, case-control, cross-sectional), with design-specific extensions
• PRISMA — for systematic reviews and meta-analyses, with extensions for network meta-analysis, diagnostic test accuracy, and other review types
• STARD — for diagnostic accuracy studies
• SPIRIT — for clinical trial protocols
• ARRIVE — for animal research in vivo studies

Authors who complete these checklists before submission — rather than as a submission requirement for a specific journal — find that the process identifies reporting gaps and framing problems that would otherwise surface in peer review. The checklist is not merely a compliance exercise: it is a structured way of asking whether every element that a critical reader will look for is present in the manuscript.

Statistical analysis reporting is an area where the checklists are particularly valuable. CONSORT requires that statistical methods be described in sufficient detail that the analyses could be reproduced by a knowledgeable reader with access to the original data, that confidence intervals be reported for all primary outcomes, and that the statistical analysis plan for subgroup analyses and sensitivity analyses be pre-specified and distinguishable from exploratory analyses. Manuscripts that do not meet these requirements typically receive reviewer requests for additional statistical reporting that delay acceptance.