Can artificial intelligence replace manual search for systematic literature? Review on cutaneous manifestations in primary Sjögren’s syndrome

Abstract

Objectives Manual systematic literature reviews are becoming increasingly challenging due to the sharp rise in publications. The primary objective of this literature review was to compare manual and computer software using artificial intelligence retrieval of publications on the cutaneous manifestations of primary SS, but we also evaluated the prevalence of cutaneous manifestations in primary SS. Methods We compared manual searching and searching with the in-house computer software BIbliography BOT (BIBOT) designed for article retrieval and analysis. Both methods were used for a systematic literature review on a complex topic, i.e. the cutaneous manifestations of primary SS. Reproducibility was estimated by computing Cohen’s κ coefficients and was interpreted as follows: slight, 0–0.20; fair, 0.21–0.40; moderate, 0.41–0.60; substantial, 0.61–0.80; and almost perfect, 0.81–1. Results The manual search retrieved 855 articles and BIBOT 1042 articles. In all, 202 articles were then selected by applying exclusion criteria. Among them, 155 were retrieved by both methods, 33 by manual search only, and 14 by BIBOT only. Reliability (κ = 0.84) was almost perfect. Further selection was performed by reading the 202 articles. Cohort sizes and the nature and prevalence of cutaneous manifestations varied across publications. In all, we found 52 cutaneous manifestations reported in primary SS patients. The most described ones were cutaneous vasculitis (561 patients), xerosis (651 patients) and annular erythema (215 patients). Conclusion Among the final selection of 202 articles, 155/202 (77%) were found by the two methods but BIBOT was faster and automatically classified the articles in a chart. Combining the two methods retrieved the largest number of publications.

Rheumatological features of Whipple disease

“…A systematic literature review was performed using the artificial intelligence software BIBOT, as previously described 13 , using the keywords detailed in Supplementary Table S1 which included “ Tropheryma whipplei”, “ Clinic” and “Classical—Classic Whipple disease”.…”

Towards a universal definition of disease activity score thresholds: the AS135 score

“…This application will provide clinicians with a quick tool to convert the main scores used in rheumatology to the AS135 score and thus allow clinicians to access the level of disease activity, even without knowledge of the thresholds used for a speci c score. Our use of BIBOT, which is not a classic systematic method to review the literature, for the literature review is a potential limitation, but we previously demonstrated that this software is at least as e cient as a manual systematic literature review and that the combination of BIBOT and manual review is the best way to conduct a literature review (12). Another limitation of the study is the use of a linear model to approximate the distribution of each score, but selected scores have been constructed to be interpreted as a linear scale, which makes the approximation performed by AS135 very acceptable We noticed that communication between rheumatologists and primary care physicians can sometimes be altered by scores that are too speci c, especially because of thresholds that differ from one score to another.…”

“…Therefore, each sentence is automatically segmented into linguistic units, and the main topics identi ed are compared to two lists of keywords: a validation list containing relevant terms and an exclusion list of terms re ecting speci c exclusion criteria. The BIBOT program has been upgraded since its last utilization (12) and is now able to classify articles in order of interest. Instead of being simply selected or rejected, the articles are divided into four different categories: Platinum, Gold, Silver and Bronze.…”

Wearable activity trackers and artificial intelligence in the management of rheumatic diseases

“…A French team developed BIBOT, a software based on natural language processing methods, to perform a systematic literature review on cutaneous manifestations of primary Sjögren’s syndrome; BIBOT was much faster than manual search and automatically classified the articles of interest in a chart. Reliability of this tool was good, given that among 202 relevant articles, 155 (77%) were selected by both AI and manual method [ 23 ].…”