Introduction

The late 1980s saw an acceleration in efforts to highlight and address the gender bias in science and medical research.[1,2] There are issues of justice associated with such a bias, and there are significant empirical concerns about the generalizability of gender biased findings and their applicability to clinical practice.[3] The efforts to address the imbalance has resulted in the establishment of several peak bodies to address gender in science, targeted funding streams to enhance both opportunities for women scientists and extend women's participation in research, and policies that require researchers to address specifically the gender implications of their design, recruitment strategies and findings. A recent “call for action” is seeking to include gender in research impact assessment.[4] Similarly, key biomedical journals increasingly require gender reporting.[2,5]

A recent review of progress on gender equality was undertaken across a range of indicators. including on authorship of research papers. The prevalence of female first authors in major medical journals increased from 27%-37% between 1994–2014, but this result was not consistent across journals.[6] We sought to investigate the gender gap further with a focus on the gender bias in clinical case reports.

“The [clinical] case report is the archetypical medical article”, at least that was a view expressed in JAMA in 1968.[7] Today, the case report is less common, having been overtaken by research papers of experimental and quasi-experimental studies. Nonetheless the clinical case report maintains a key role in the medical literature as a vehicle for reporting unusual disease presentations or outbreaks,[8,9] as a hypothesis generator,[10,11] and most typically as a pedagogic tool.[12–14] In their “Instructions to Authors”, for example, the British Medical Journal, and the Lancet highlight the place of the case report in this latter role. The British Medical Journal mentions the requirement for “real cases … suitable for presentation in specifically educational formats”, and the Lancet observes that “novelty is not essential, but at least one broadly useful learning point is.”

A case report could be about any patient or any clinical condition, and it is unreasonable to expect that a single case report is gender balanced. In their specificity they are necessarily biased. As a corpus of case reports within a general and internal medicine journal, however, one would hope to see a representative sample of the patient population, gender balanced.

Materials and methods

Results

The initial search identified 2,742 case reports. Forty (40) case reports were excluded because the gender of the patient was not identified; and a further 120 case reports were excluded because they related to patients of both sexes. The final sample of case reports was 2,582, accounting for 94.2% of the initially identified reports. The flow diagram of case report inclusion is shown in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study flow chart of clinical case report selection.

https://doi.org/10.1371/journal.pone.0177386.g001

Of the 2,582 case reports, just under half (46.8%) appeared in the New England Journal of Medicine (NEJM), 23% appeared in the British Medical Journal (BMJ), 17.4% appeared in The Lancet (Lancet), 8.7% appeared in the Journal of the American Medical Association (JAMA), and 4.1% appeared in the Annals of Internal Medicine (AnnInternMed). Of the case reports, 1,168 (45.2%) involved female patients and 1,414 (54.8%) involved male patients. Table 1 shows the sex distribution of the case reports for each journal.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. A word frequency analysis of the MeSH text associated with the 2,582 case reports aggregated into four categories for each gender: body/organ, disease, symptom, and investigation/treatment.

https://doi.org/10.1371/journal.pone.0177386.t001

A series of logistic regression models were developed with the sex in the case report as a dichotomous outcome variable, and the journal as a potential explanatory factor for observed variation in the proportion of female case reports. A random intercepts logistic regression model was developed to estimate the variance associated with journal as a random intercept. The associated variance was extremely small (4.0x10^-14) and the inclusion of the journals as a fixed effect in a second model reduced the variance to zero. A Wald test for inclusion of the fixed effects showed no significant difference between the two models (Chi-sqr = 4.178, df = 4, p = 0.382) indicating that the data from the 2,582 case reports could be directly pooled.

Fig 2 shows a combined plot and table of the data associated with each journal; the number of female case reports and the total; and the 95% confidence interval surrounding the individual journal estimate. The pooled result is shown last.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The gender bias in clinical case reports among the “big five” journals.

https://doi.org/10.1371/journal.pone.0177386.g002

Squares and the horizontal lines represent the measures of effect, e.g. the proportion of case reports about women, and associated confidence intervals for each journal, and the diamond indicates the summary measure.

The NEJM (.44; 95%CI:.41–.47) and BMJ (.45; 95%CI:.41–.49) were the only two journals that showed a statistically significant bias against case reports of female patients. The Annals of Internal Medicine was the only journal that had a point estimate in the direction of a bias against male patients (.53; 95%CI:.43–.63). The Annals of Internal Medicine, however, published relatively few case reports (n = 105) and the confidence intervals were consequently wide. The remaining two journals had point estimates in the direction of a bias against female patients; however, neither of the individual estimates were significantly different from 0.5. In the pooled analysis, there was a statistically significant gender bias in case reports against female patients (0.45; 95%CI: 0.43–0.47). The 95% Confidence intervals of the individual journals’ estimates including the Annals of Internal Medicine overlapped the pooled estimate indicative of a gender bias.

To clarify that the gender bias was not a “NEJM problem” we recalculated the pooled estimate with only four of the journals, excluding the NEJM data (n = 1,444). The bias remains, albeit slightly smaller, and with wider confidence intervals because of the smaller sample size (.46; 95%CI:.437–.497, p = .006).

Discussion

There were 10% more case reports about male patients than female patients identified in the review. What appears to be a relatively small effect in cross-section should really be seen as a potentially accumulating effect over time; and small accumulating effects can have a substantial impact.[19]

The concern is that a bias in clinical perception and decision making may be reinforced with continuous exposure towards an over representation of case reports involving male patients—supporting historical biases in clinical medicine and clinical research.[20] Medicine in males may be regarded as more main stream, more interesting, more indicative of what is a normal disease process. Because the case reports have become an integral part of medical education, the risks of gender biased exposure to “interesting clinical medicine” is potentially compounded, particularly in junior clinicians who are still laying down the matrix of expertise.[21] Unfortunately, the nature of these kinds of influences on decision making are likely to be subtle, hard to recognise in one's own decision making, and even when one is alert to the bias, formidable to overcome.[3]

What is surprising, is that the textual analysis of the MeSH words identified a remarkable similarity between case reports about male and female patients. If similar kinds of clinical cases are catching the eyes of authors and editors, is there a need to represent those clinical cases with a gender imbalance? Are there inherently more interesting characteristics in male than in female patients?

It is impossible to know the source of the observed bias, but there are three potential explanations. First, the apparent bias is an artifact of random noise in the data. This seems unlikely given the narrowness of the confidence interval around the estimate, but it is certainly a possibility. The second possibility is that there is an authorial bias. Authors of case reports are submitting more case reports about male patients than female patients to journals, and the journals are publishing them in the gender ratio with which they are received. The third possibility is that there is an editorial bias. Authors submit a gender balance of equally meritorious case reports and editors are more likely to approve for publication those reports about male patients. The extension of explanation two and three is that some interplay arises between an authorial and an editorial bias.

The sex and gender equity in research guidelines (SAGER) developed by the European Association of Science Editors were designed primarily to guide authors in preparing their manuscripts, but they are also useful for editors, as gatekeepers of science, to integrate assessment of sex and gender into all manuscripts as an integral part of the editorial process.”[2] In developing guidelines to ensure that gender is appropriately reflected in research the focus is on the individual article.[2] With respect to case reports, the role of the editor needs to extend beyond the impossible task of identifying bias in any specific case report, and should include an overview of the journal's corpus of published work.