PubMed growth patterns and visibility of journals of Sub-Saharan African origin

Abstract

In their extensive analysis of African biomedical publications, Hofman et al. [1] found that in ten years' time (1995–2004), the number of publications with authors from Sub-Saharan Africa (SSA) indexed in PubMed had increased by more than 40%. As they correctly indicate, PubMed is not an ideal database to monitor the geographical distribution of authors, as only 1 address per paper is included (and not in a standardized way). But still, when looking at the 10 most prolific SSA countries, an increase from 2,073 papers listed in 1995 to 2,929 papers in 2004 suggests that the international visibility of African research is clearly on the rise. Yet if one takes a closer look at the distribution patterns of the PubMed database (Table 1), currently featuring over 18.7 million bibliographic references from 1950 onward, it appears that the (adjusted) overall increase of listed papers from 1995–2004 is roughly the same as that reported for the African papers (39.64% vs. 41.29%). The increase of the average number of papers per journal and the emergence of new (often open access) journals indexed in PubMed are probably major factors to account for this trend. As is clear from column 4, the annual increase of records does not show a stable pattern (e.g., ranging from 0.12%–8.26% in just 3 years, then back to 2.30%). Table 1 Evolution of the number of PubMed records 1989–2008 (counted on April 1, 2009) Moreover, the correct identification of publication years for the last decade can constitute a potential pitfall when using PubMed as a tool for bibliometric applications. With the practice of listing “Epub before print” descriptions, both the date of the online posting of individual papers and the date of their actual publication in a complete issue are included in the date fields. For many papers, this results in 2 different years of publication (e.g., e-publication on November 12, 2007, and publication of the full issue on February 15, 2008; the reverse also occurs, but apparently to a far lesser extent). In this way, a naive search for publication years may yield up to 10% incorrectly accorded bonus records (e.g., year 2007 in Table 1). This double publication year phenomenon starts around the year 2000 and becomes really conspicuous from the year 2005 onward. For the adjustments in Table 1, I have given priority to the most recent year as the “true” date of publication, but this may be too simple a rule to correctly pinpoint them all. Another complication for date searching in post-1999 publications is that there is always a difference in the number of articles retrieved using the “date of publication” limit and the number retrieved using the “date of publication [dp]” field, which gets increasingly larger over time. Differences of several hundreds of records per year occur. Although this difference represents less than 0.1% of the annual totals, it is difficult to decide which method is the most accurate. I have chosen the “limits” technique, invariably yielding the largest numbers of records. Table 1 also shows a relatively stable availability of free full-text—as indicated by the “Links to free full text” limits option—for around 20%–23% of all listed papers for the last decade. Counts were made at the end of the first quarter of 2009, so the apparent setback in 2008 probably reflects the embargoes of 6–12 months of many semi-open access journals (e.g., those cohosted in the PubMed Central repository). For the penultimate decade, relative full-text availability is much lower but still over 10%. As this period largely predates the web era, the results of retrospective digitization are clearly visible. Both PubMed and PubMed Central are continually being updated with older material (but again a very small fraction of total updates).