go-v.orgdoi

Over-reporting of p-values in the medical literature should be discouraged

Eric P. Brass · Clinical Investigation · 2012

  • DOI 10.4155/cli.12.35
  • Publisher OMICS Publishing Group
  • Type article
  • Issued 2012

Not cached yet. This record was fetched live for this request and is stored after the lookup. Reopening the page should serve the cached copy.

Clinical Investigation · OMICS Publishing Group · article · 2012

Abstract

The reporting of clinical trial data increasingly includes p‐values that are not based on prospective hypothesis testing. These p‐values often add little to the utility of the data presentation and have the potential to facilitate misinterpretation by the reader. This trend and the associated potential problems are illustrated in recent papers published in the literature. This over‐reporting of p‐values should be discouraged. Rigorous statistical methodologies have provided critical tools for designing, analyzing and interpreting clinical research.

Effective communication of clinical trial results, particularly to the non‐statistician, is an important responsibility of clinical investigators. The calculation of p‐values for predefined end points using prespecified methodologies greatly enhances the utility of trial results for the reader. However, there has been a substantial increase in the reporting of p‐values for comparisons that lack the rigor and robustness required for hypothesis testing. As highlighted by Fleming, the reporting of these p‐values adds little information to the data presentation and risks misleading the reader with respect to the proper interpretation of the data [1].

Recent publications from several major journals illustrate this problem. Ensuring the comparability between study arms in a clinical trial is important. Thus, Lee et al. list baseline characteristics of the study groups in Table 1 of their paper on statin effects in patients with acute myocardial infarctions [2]. This table appropriately provides values and indices of variability for each cohort. The table also provides p‐values for each characteristic – a total of 33 p‐values in this one table! How is the reader to interpret these values which range from <0.001 to 0.995? Is the p‐value more informative about potential imbalances then the values themselves which are so clearly provided?

No adjustment was made to these p‐values for multiple comparisons, and thus the reader who focuses on an individual low p‐value for a characteristic of interest may attach ‘significance’ to what could be a chance finding. If clinical features have the potential to impact the effect of an intervention there are robust strategies for dealing with this concern. In this case, Lee et al. used a propensity score analysis to adjust for potential confounders [2]. The p‐values in Table 1 of their paper thus cannot add to the reader’s interpretation of the study and may detract form the more formal propensity score analysis.

Similarly, Ogedegbe et al. report 20 p‐values in the table of baseline characteristics in their interesting paper on interventions to improve adherence in patients with hypertension, thus risking inappropriate inference by the reader [3]. Redon et al. report on patient characteristics based on outcomes and subgroups in Table 3 of their interesting paper on blood pressure effects in high‐risk diabetic “The proper use of p-values may be well understood in principle, but the cited examples and many others in recent publications suggest that misuse persists in practice. All editors should take steps to discourage the over-reporting of p-values in the literature.”

Open the resolved locator · 10.4155/cli.12.35

Metadata licence: OpenAlex metadata (CC0)

Registry
OpenAlex DOI index + API
Fetched
2026-11-19 23:52 UTC
Cache state
not cached yet (bulk-index+openalex)
Locator
https://doi.org/10.4155/cli.12.35
Metadata licence
OpenAlex metadata (CC0)
Scope
Metadata as returned by the resolver; abstracts are never generated here.