Examining the Intersection between Gender, Community Health Workers, and Vector Control Policies: A Text Mining Literature Review

Abstract

Gender intersects with healthcare systems; this is equally true for arboviral vector control efforts. However, there is as yet no comprehensive analysis as to how vector control is gendered. Hence, our objective is to provide the first thematic scoping and spatial distribution of the literature on gender, community health workers, and vector control. The authors use a systematic review approach to collect the academic literature on gender, community health workers, and vector control in Web of Science, Scopus, and PubMed (7,367 articles). After applying the exclusion criteria, 2,812 articles were analyzed using machine learning techniques: text mining and quantitative text analysis. The authors use topic modeling to assess the thematic scope of the literature and analyze the spatial distribution of themes. Our results show that the literature’s spatial scope is strongly represented by the global south as research was conducted mainly in Latin America, Africa, and Asia, places with greater incidence of vector-borne disease and with health systems, which incorporate community healthcare workers. However, there are significant spatial heterogeneities in where and how research is conducted. The topic analysis reveals that the literature predominantly considers issues of sex (e.g., pregnancy) and gender as it relates motherhood. Gendered considerations occur upon implementation of vector control policies, rather than being mainstreamed into their development and delivery. There is a need to deepen the analysis to allow for gendered aspects to be understood beyond binary sex differences and/or reproductive health.

Automation of systematic reviews of biomedical literature: a scoping review of studies indexed in PubMed

“…The fourth tool, developed for public health purposes, Dextr [ 140 ] combined vector embedding text representation and deep learning. Further approaches included PECO tagging in a rapid evidence mapping study using SWIFT Review [ 125 ], extraction of geographic locations from the manuscript [ 141 ], extraction of endpoints as comparative claim sentences [ 142 ], data extraction from ClinicalTrials.gov for meta-analyses [ 143 ], and convenience tools to highlight relevant sentences [ 74 ], or extract data from graphs [ 144 ]. Finally, development of the REDASA COVID-19 dataset involved human experts in the loop, web-crawling, and a natural language processing search engine to provide a real-time curated open dataset for evidence syntheses to aid pandemic response [ 48 ].…”

“…From the 15 ASRs, four (26.7%) reviews automated the search [ 33 , 37 , 45 , 49 ], eleven (73.3%) the screening [ 45 , 67 , 80 , 100 , 103 , 110 , 111 , 126 , 133 , 136 ], two (13.3%) the full text selection [ 126 , 137 ], and one (6.7%) the data extraction phase [ 141 ]. One study did not report the software [ 100 ], six used open source software [ 33 , 37 , 45 , 110 , 137 , 141 ], and eight studies used off-the shelf tools [ 49 , 67 , 80 , 103 , 111 , 126 , 133 , 136 ]. Three studies (20.0%) reported recall with values between 96% and 100% [ 67 , 111 , 126 ].…”

Automation of systematic reviews of biomedical literature: a systematic review of studies indexed in PubMed

“…From the 15 ASLRs, four (26.7%) reviews automated the search 113 , 43,61,136 , eleven (73.3%) the screening 50,83,89,95,110,113,120,132,146,147 , two (13.3%) the full text selection 71,132 and one (6.7%) the data extraction phase 133 . One study did not report the software 110 , six used open source software 43,61,71,83,113,133 , and eight studies used off-the shelf tools 50,89,95,120,132,136,146,147 . Three studies (20.0%) reported recall with values between 96%-100% 89, 95,132 .…”

“…The fourth tool, developed for public health purposes, Dextr 148 combined vector embedding text representation and deep learning. Further approaches included PECO tagging in a rapid evidence mapping study using SWIFT Review 91 , extraction of geographic locations from the manuscript 133 , extraction of endpoints as comparative claim sentences 55 , data extraction from ClinicalTrials.gov for meta-analyses 97 , and convenience tools to highlight relevant sentences 130 , or extract data from graphs 88 . Finally, development of the REDASA Covid-19 dataset involved human experts in the loop, web-crawling and a natural language processing search engine to provide a real-time curated open dataset for evidence syntheses to aid pandemic response 127 .…”

“…The topics of ASLRs were usually broad, with on average 17952 records (range 962-52219) and 691 included studies (range 13-6305). From the 15 ASLRs, four (26.7%) reviews automated the search 113 , 43,61,136 , eleven (73.3%) the screening 50,83,89,95,110,113,120,132,146,147 , two (13.3%) the full text selection 71,132 and one (6.7%) the data extraction phase 133 . One study did not report the software 110 , six used open source software 43,61,71,83,113,133 , and eight studies used off-the shelf tools 50,89,95,120,132,136,146,147 .…”