Extent of evidence synthesis in biomedical research: a MeSH-driven analysis of neglected and well-explored areas

Background

Previous evaluations raised concerns about redundant meta-analyses in specific fields. Guidelines have been developed to determine when updates are necessary, aiming to curb this issue and minimize unnecessary research efforts [1, 2]. Conversely, other areas remain underexplored, showcasing a notable lack of evidence synthesis summaries.

Objective

To map the diversity of meta-analysis topics and identify areas that are over- or under-represented by employing a large-scale, Medical Subject Headings (MeSH)-driven analysis across the entire spectrum of biomedical research.

Methods

We searched for meta-analyses published in Medline between 01/1990 and 09/2024. Our analysis was focused on the assigned primary MeSH headings along with the available subheadings for each article, which are used to categorize and describe the content of the study. We excluded reports with missing primary MeSH terms or subheadings. We compared the assigned terms to those listed in the National Library of Medicine’s complete list of MeSH terms [3]. We measured and visualized the frequency distribution of primary MeSH terms across the meta-analyses. To evaluate the primary MeSH term diversity among the available subheadings, we calculated the coefficient of variation of MeSH headings as the ratio of the standard deviation to the mean. Higher values indicated a greater variability in MeSH terms frequencies, suggesting a more diverse distribution of the MeSH terms in the corresponding subheadings. We visualized the derived coefficient of variation against the standard deviation. All analyses were conducted in Python 3.12.

Findings

We identified 162,478 unique reports of meta-analysis listed in Medline. Overall, 449,639 primary MeSH headings (14,072 unique terms) and 76 unique subheadings were extracted. The median number of assigned primary MeSH terms in each meta-analysis was 3 (interquartile range: 2–4, minimum: 1, maximum: 17). Compared to the currently available MeSH vocabulary list in Medline, 16,692 (out of 30,764) terms were not assigned to any meta-analysis. As shown in Fig. 1, there is a disproportionate frequent use of a small subset of terms compared to the vast majority. The most frequently assigned MeSH terms pertain to oncology, cardiovascular, and infectious disease fields, while the least frequently assigned terms are related to basic sciences, health services, and surgical procedures. Figure 2 illustrates the diversity of the corresponding primary MeSH terms across the 76 subheadings. Subheadings such as “Therapeutic Use,” “Drug Therapy,” “Prevention & Control,” and “Methods” show high coefficients of variation and standard deviation values, indicating significant variability in the MeSH terms associated with them. In contrast, subheadings represented by lower points in Fig. 2 exhibit less variability, suggesting more consistent application of MeSH terms and a narrower range of topics of interest.

Fig. 1

Frequency of the individual primary Medical Subject Headings terms assigned across the meta-analyses. The primary MeSH terms on the x-axis are arranged in alphabetical order, as specified in the National Library of Medicine’s comprehensive list of MeSH terms

Fig. 2

Coefficients of variation distribution for the primary MeSH headings across the corresponding subheadings. The x-axis represents the coefficient of variation, highlighting the relative variability in each subheading compared to its mean, while the y-axis shows the standard deviation, indicating absolute variability. The size of each bubble corresponds to the number of meta-analysis contributing to each of the 76 subheadings

Discussion

Our MeSH-driven analysis reveals the large-scale extent of uneven evidence synthesis efforts across biomedical research, with several fields notably underrepresented. Previous studies have raised concerns about redundancy in meta-analyses within specific fields. However, these evaluations were typically limited to particular research areas. Building on this foundation, we employed a comprehensive, MeSH-driven approach to uncover significant disparities in evidence synthesis efforts across the full spectrum of biomedical research. Notably, we observed a disproportionate focus on oncology, cardiovascular, and infectious disease fields, with basic sciences, health services, and surgical procedures remaining less explored. In some topics, rare conditions and limited primary studies may explain the lack of meta-analyses; however, such circumstances would particularly benefit from evidence synthesis [4, 5]. In our analysis, we focused only on meta-analyses, excluding other forms of synthesis (e.g., scoping or systematic reviews) that might be present in such underrepresented fields. These findings should be considered with certain limitations in mind. First, while MeSH provides a comprehensive biomedical vocabulary, its evolving nature may omit emerging topics or previously used terms, especially in fast-evolving fields. Nonetheless, we matched all major headings to terms from the latest MeSH list [3]. Second, MeSH terms serve as proxies for topics, but their ambiguity and multiple meanings can cause misinterpretations in specific research contexts.

Despite limitations, our analysis shows significant disparities in evidence synthesis across biomedical fields over three decades, with some areas heavily saturated and others potentially understudied. Further research is needed to understand the causes of this imbalance beyond disease prevalence and to identify neglected areas that would benefit from systematic evaluation and synthesis of existing evidence.