Artículos

Los indicadores de citación pueden medir el impacto o la utilidad de resultados de investigación de un artículo científico, sin embargo, este uso puede ser controversial. Factores intrínsecos y extrínsecos influencian la citación de un artículo, sin mencionar que el comportamiento en las citaciones puede variar entre áreas temáticas, lo cual dificulta las comparaciones entre artículos y disciplinas. Entender que el contexto puede afectar un análisis de citas es esencial para interpretar adecuadamente los indicadores. Por esta razón, buscan reconocerse los factores que inciden en la citación de los artículos de las revistas biomédicas colombianas indexadas en Scopus a través del uso de algoritmos de Machine Learning. Con los algoritmos ‘Gradient Boosting Classifier’ y ‘Light Gradient Boosting Machine’ identificamos características de importancia como el índice h del primer y el último autor, acceso abierto, número de autores, palabras clave del artículo, además del número de páginas. Estas características fueron relevantes para el área de interés y pueden brindar un contexto para futuros análisis, considerando que lo relevante de un artículo no debería ser cuántas citaciones atrae, sino si este ayuda a llenar vacíos en el conocimiento.

Introduction: Alcohol’s effects on cardiovascular disease (CVD) are controversial. Alcohol industry actors have shown particular interest in this subject, and been extensively involved through research funding, and in other ways, generating concerns about bias, particularly in reviews. Material & methods: We conducted a co-authorship network analysis of the primary studies included within a previous co-authorship study of 60 systematic reviews on the impact of alcohol on CVD. Additionally, we examined the relationships between declared alcohol industry funding and network structure. Results: There were 713 unique primary studies with 2832 authors published between 1969 and 2019 located within 229 co-authorship subnetworks. There was industry funding across subnetworks and approximately 8% of all papers declared industry funding. The largest subnetwork dominated, comprising 43% of all authors, with sparse evidence of substantial industry funding. The second largest subnetwork contained approximately 4% of all authors, with largely different industry funders involved. Harvard affiliated authors who at the review level formed co-authorship subnetworks with industry funded authors were seen at the primary study level to belong to the largest epidemiological subnetwork. A small number of key authors make extensive alcohol industry funding declarations. Conclusions: There was no straightforward relationship between co-authorship network formation and alcohol industry funding of epidemiological studies on alcohol and CVD. More fine-grained attention to patterns of alcohol industry funding and to key nodes may shed further light on how far industry funding may be responsible for conflicting findings on alcohol and CVD.

Background: In the last five years, there has been an accelerated growth in the scientific production about Artificial Intelligence and Healthcare by Scholars of the most diverse disciplines. Recently, the scientific corpus has been enriched with considerable literature reviews ranging from the overview of large collections of scientific documents to the recognition of the state of knowledge on specific aspects (e.g., in the medical field, ophthalmology, cardiology, nephrology, etc.). Methods: The methodological approaches belong to the scientific fields of bibliometrics and topic modeling. Following a bibliometric analysis of the literature on the subject, conducted on a vast collection of scientific contributions, we also searched for the “latent” themes in the semantic structures of these documents, identified the relationships between them and recognized the most likely to be investigated in the future. Results: Results show 24 topics about future trends in literature review connecting the field of AI and Healthcare. Conclusions: This bibliometric review of the literature on artificial intelligence and healthcare allows identifying of some privileged areas of attention by scholars of different disciplines. However, it also reveals the limits of hard clustering techniques, as demonstrated by the presence of some keywords in several groups. The numerous existing reviews must be integrated by reviews based on Topic Modeling techniques, which make it possible to identify topics, historical trends (classical and emerging topics), associations between the documents and to predict, on a probabilistic basis, which scientific fields will be most likely to see development in the future.