Department of Neurology, Rutgers Robert Wood Johnson Medical School, New Brunswick, NJ, USA.
Headache. 2021 Jan;61(1):143-148. doi: 10.1111/head.14022. Epub 2020 Dec 8.
Non-headache literature inevitably influences headache research, but the way this interdisciplinary interaction occurs has seldom been evaluated.
Utilizing network analysis techniques within the PubMed Central (PMC) database, we illustrate a novel method by which to identify and characterize the important non-headache literature with significant impact within the headache world.
Using the National Center for Biotechnology Information E-utilities application programing interface and custom backend software, all PMC articles containing the words "headache(s)" and/or "migraine(s)" in the title were identified. This generated a list of "seed articles" to represent the body of primary headache literature. Articles referenced by the seeds were then found, generating the list of articles with one degree of separation from the seeds (first-degree neighbors). This was iterated twice more to find the second- and third-degree neighbors. A directed network graph was generated for each level of separation using these articles and their referential connections. The hyperlink-induced topic search (HITS) and PageRank algorithms were used on these graphs to find the top 50 articles in the network (hub and authority rank via HITS, general rank via PageRank). Removing seed articles from the ranked lists left the influential non-headache articles at each level of separation.
We extracted 6890 seed articles. The first-, second-, and third-degree models contained 16,451, 105,496, and 431,748 articles, respectively. As expected, most first-degree neighbors were part of the seed group (headache literature). Using HITS, at the second degree, only two seed articles were found in the top 50 hubs (none in the authorities); the majority of non-seed articles were basic neuroscience, involving ion channel function or cell signaling. At the third degree, there were no seeds and all articles involved imaging/structure of brain connectivity networks. PageRank gave more varied results, with 35/50 second-degree articles being seeds, and the remainder a mixture of articles describing rating scales (3), epidemiology/disease burden (3), basic statistical/trial methods (3), and mixed basic science (6). At the third degree, five were seeds; non-seed articles were represented heavily by genomic mapping studies, brain connectivity networks, and ion channel/neurotransmitter studies.
This work demonstrates the value of network citation analysis in the identification of interdisciplinary influences on headache medicine. Articles found with this technique via HITS identified and grouped basic science applicable to headache medicine at the molecular scale (ion channels/transmitters), and whole-brain scale (connectivity networks). Both groups have direct clinical correlates, with the former implicating pharmacological targets, and the latter implicating functional neuroanatomy and pathophysiology of various headache disorders. Likely, in-depth analysis of the whole network (rather than the top 50) would reveal further clusters where the relationship to headache may not be as immediately obvious. This may not only help to guide ongoing work, but also identify new targets where seemingly unrelated work may have future applications in headache management.
非头痛文献不可避免地会影响头痛研究，但这种跨学科相互作用的方式很少得到评估。
我们利用 PubMed Central (PMC) 数据库中的网络分析技术，展示了一种新的方法，可以识别和描述对头痛领域具有重要影响的重要非头痛文献。
使用美国国家生物技术信息中心的 E-utilities 应用程序编程接口和定制后端软件，我们确定了所有在标题中包含“头痛”和/或“偏头痛”字样的 PMC 文章。这生成了一份代表原发性头痛文献主体的“种子文章”列表。然后找到引用这些种子的文章，生成与种子有一个分离度的文章列表（第一级邻居）。这一过程迭代了两次，以找到第二级和第三级邻居。使用这些文章及其参考连接，为每个分离度生成一个有向网络图。使用超链接诱导主题搜索（HITS）和 PageRank 算法在这些图上找到网络中的前 50 篇文章（通过 HITS 找到的顶级和权威排名，通过 PageRank 找到的一般排名）。从排名列表中删除种子文章，就可以得到每个分离度的有影响力的非头痛文章。
我们提取了 6890 篇种子文章。第一、第二和第三级模型分别包含 16451、105496 和 431748 篇文章。如预期的那样，大多数第一级邻居都是种子组（头痛文献）的一部分。使用 HITS，在第二级，只有两篇种子文章在 50 个顶级枢纽中（没有在权威中）；大多数非种子文章都属于基础神经科学，涉及离子通道功能或细胞信号转导。在第三级，没有种子，所有的文章都涉及到脑连接网络的成像/结构。PageRank 给出了更多不同的结果，其中 35/50 篇第二级文章是种子，其余的是描述评分量表的文章（3）、流行病学/疾病负担（3）、基本统计/试验方法（3）和混合基础科学（6）。在第三级，有 5 篇是种子；非种子文章主要由基因组图谱研究、脑连接网络和离子通道/神经递质研究组成。
这项工作证明了网络引文分析在识别头痛医学中跨学科影响方面的价值。通过 HITS 技术找到的文章在分子水平（离子通道/递质）和全脑水平（连接网络）上识别和分组了适用于头痛医学的基础科学。这两个组都有直接的临床相关性，前者涉及到药理学靶点，后者涉及到各种头痛疾病的功能神经解剖学和病理生理学。很可能，对整个网络（而不是前 50 名）进行深入分析，将会揭示出更多的聚类，其中与头痛的关系可能不那么明显。这不仅有助于指导正在进行的工作，还可以确定看似不相关的工作在头痛管理中可能具有未来应用的新目标。
