PMC:7799291 / 52149-55193
Annnotations
{"target":"http://pubannotation.org/docs/sourcedb/PMC/sourceid/7799291","sourcedb":"PMC","sourceid":"7799291","source_url":"https://www.ncbi.nlm.nih.gov/pmc/7799291","text":"Systematic review automation\nSystematic reviews aim to synthesize results over all relevant published studies on a topic, providing the highest quality of evidence and recommendations for clinical and public health decisions. They have become a fixture in the biomedical literature, with many established protocol around their registration, production, publication and update [10, 15, 81]. We refer to them here because the systematic review framework is useful to keep in mind when discussing evidence summary and information overload. In Figure 2, we show the steps of systematic review construction [38]. Indeed, many of the text mining tasks we discuss previously can be framed in the context of systematic review construction. For example, search and QA can help to identify relevant documents and spans of text, table completion helps to extract structured evidence from different studies and multi-document summarization is a way of aggregating evidence across studies.\nFig. 2. The process of systematic review construction (left) and example systems that assist with several steps (right). Systematic reviews have played an important role in the scientific response to COVID-19. Rapid reviews, which condense and shorten the typically months- or years-long systematic review process [39, 89], have been common. For example, rapid reviews have been published addressing research questions on infection and mortality rates [31], clinical characteristics in different subpopulations [27, 32, 65], symptoms of disease [66, 84], drug repurposing [78], COVID-19 management policies [108], as well as interactions between COVID-19 and other diseases and comorbities [20, 69, 112, 113]. Due to the large number of COVID-19 reviews, numbering in the thousands, the ones we have chosen to cite here are ones that use COVID-19 corpora like CORD-19 or LitCovid as a source of studies in addition to traditional databases like PubMed.\nAs the number of publications on COVID-19 has grown, it becomes increasingly difficult and expensive to produce and update these reviews. Systems that assist with or automate parts of the review process are needed. Several existing systems focus on automating parts of the systematic review process more broadly [90]. These systems focus on supporting the identification of relevant studies [5, 64, 70, 71, 94] or extracting PICO elements [22, 42, 55, 62]. The recently released Trialstreamer system allows users to discover new clinical trials using PICO-based search [61]. ASReview [5, 94], Rayyan [64] and Trialstreamer [61] all have COVID-19 modules that allow users to focus exclusively on COVID-19 papers.\nThe processes around creating systematic reviews have matured over the past several decades. Reviews provide trusted evidence to clinicians and policymakers and are useful for addressing information overload, as they survey and summarize information across numerous studies. Targeted methods and systems that assist in or automate systematic reviews for COVID-19 could be very impactful going forward.","divisions":[{"label":"title","span":{"begin":0,"end":28}},{"label":"p","span":{"begin":29,"end":976}},{"label":"figure","span":{"begin":977,"end":1098}},{"label":"label","span":{"begin":977,"end":984}},{"label":"caption","span":{"begin":986,"end":1098}},{"label":"p","span":{"begin":986,"end":1098}},{"label":"p","span":{"begin":1099,"end":1930}},{"label":"p","span":{"begin":1931,"end":2642}}],"tracks":[]}