INCLUSION AND EXCLUSION CRITERIA

Each PubMed record includes a variable number of fields as described on the PubMed website. Errata, Letters, Biographies, Comments, Directories, News, and Retractions of publication, as specified in Publication Type field (PT), were not considered as research publications and were excluded. Records belonging to conferences or congresses were also omitted because many of them are likely to be published as full articles thereafter, thereby creating duplicate records for the same investigation. Finally, records in which the First Author Field (FAU) was not filled were not considered either, because no information about the author’s gender could be obtained.

It is common practice in medical research that “the first author is the person whose work underlies the paper as a whole,” whereas the last authorship “indicates a person whose work or role made the study possible without necessarily doing the actual work,” and is normally given to a person in a senior position. Because those were considered the most relevant positions to our purpose, only the first and last author’s gender and country of affiliation were analyzed. The influence of a First Author Female (FAF) and Last Author Female (LAF) was studied.

COUNTRY ASSIGNMENT

Affiliation Field (AD) was explored to assign the first and last author’s country of affiliation. When this included more than 1 country (eg, overseas fellowships), the one with the lower gross domestic product per capita (more likely the author’s country of origin) was assigned.

Because US authors tend not to include their country of affiliation, a second search for the state and largest cities (>100 000 inhabitants) was performed in the unassigned records. Finally, a manual search was carried out for the rest of the unidentified affiliations.

From what we observed, the way in which the affiliation field is filled in PubMed records depends on the journal itself. Some journals do not provide the affiliation, whereas others enlist the affiliation for each author listed, even when shared. When the affiliation field was filled exclusively for one of the authors, we automatically assigned it to both the first and last authors.

Finally, countries were classified as majority English speaking based on the specifications of the Government of the United Kingdom, to determine whether that could be a confounding factor.

GENDER ASSIGNMENT

To increase the accuracy of the gender classification, this was performed at 3 levels. First, the Gender API (Gender-API.com, Germany) was used to classify all of the possible combinations of Name/Country (where available). As suggested by the API provider, including country increases the accuracy of the gender prediction. Only those predictions with ≥95% accuracy, and based on at least 10 samples, were considered as possibly correct, whereas the rest were rejected.

Registers of given birth names and their frequency are publicly available from many national statistics services (eg, United States and Spain) and were compiled to create lists of male/female names. Only the names used at least 10 times, and with a frequency >95% for one of the genders were included (eg, Diana was used for females 99.7% of the times; Michael was used for males 99.5% of the times) and the whole data set was classified using these lists. Records classified as belonging to different genders by the API and lists were set as undetermined.

Finally, to reduce the number of unidentified records, authors were searched for in professional profiles (hospital/lab websites, Google Scholar, ResearchGate, LinkedIn, Aminer, etc) using their names, affiliations, and/or the title of their publications. Authors were manually searched until at least either the first or last author was identified for 90% of records in each journal; however, 90% could not be reached for Eye (London) , because in many cases this journal provides initials instead of full names. A large percentage of these manually searched records corresponded to authors with a Chinese affiliation, because the romanization of Chinese names tends to make it difficult to assign a gender.

To verify the accuracy of the gender assignment strategy, a random subset of 265 authors classified unanimously by both the API and the name lists were searched manually. After verification that the API and the name list agreed in ≥95% cases—discrepancies were rare (<0.3%)—and that the accuracy of the prediction was around 95%, a classification by only the API or the lists (based on at least 10 samples and 95% accuracy/frequency) was deemed sufficient to assign the gender. Finally, the current gender of the editors-in-chief in the JCR top 30 was also analyzed.

RESEARCH TOPICS

To analyze authors’ interests and contributions in the field, 30 research topics were scrutinized. Records were assigned to 1 or more topics based on their MeSH terms and keywords (fields MH and OT, where available).

TIMING

Dates of receipt, review, acceptance, and publication in PubMed were extracted and the intervals received-reviewed (t _REV ), received-accepted (t _ACC ), and received-published (t _PUB ) were calculated for each record. Not all records contained all of the dates and so the statistical analysis for each interval was only performed on the subsets for which the interval was available. Timing analysis was performed for all records and in various predefined subgroups (ie, English names, Asian names, specific topics).

STATISTICS

Matlab R-2020b (The MathWorks Inc) was used for data curation and analysis and JMP Pro 15 (SAS Institute Inc) was used for statistical analyses. The cut-off for significance was considered to be .05.

Mixed models were used for the analyses, and the full factorial equation was:

Stay updated, free articles. Join our Telegram channel

Join