30 April 2022 in Intestinal Research
Background/Aims: The incidence and prevalence of inflammatory bowel disease (IBD) is rising in Asia recently. The study aimed to obtain a comprehensive understanding of the current status of drug therapy and monitoring for IBD in Asia.Methods: A questionnaire investigation on drug therapy and monitoring for IBD was conducted right before the 6th Annual Meeting of Asian Organization for Crohn’s & Colitis. Questionnaires were provided to Asian physicians to fill out via emails between March and May 2018.Results: In total, responses of 166 physicians from 129 medical centers were included for analysis. Among the surveyed regions, the most average number of IBD specialist gastroenterologists and nurses was 4.8 per center in Taiwan and 2.5 per center in Mainland China, respectively. 5-Aminosalicylic acid/sulfasalazine (99.4%) was the most preferred first-line choice for mild-moderate ulcerative colitis (UC), meanwhile corticosteroid (83.7%) was widely applied for severe UC. The first-line medication for Crohn’s disease (CD) markedly varied as corticosteroid (68.1%) was the most favored in Mainland China, Japan, and South Korea, followed by infliximab (52.4%) and azathioprine (47.0%). Step-up strategy was preferred in mild-moderate UC (96.4%), while 51.8% of the physicians selected top-down treatment for CD. Only 25.9% and 17.5% of the physicians could test blood concentration of infliximab and antibody to infliximab in their hospitals, respectively.Conclusions: The current status of drug therapy and monitoring for IBD in Asia possesses commonalities as well as differences. Asian recommendations, IBD specialist teams and practice of therapeutic drug monitoring are required to improve IBD management in Asia.
14 July 2026 in SSRN Electronic Journal
High-quality event logs are essential for process mining, yet critical business information increasinglyresides in customer service conversations.Existing extraction methods based on Large LanguageModels (LLMs) struggle with hallucination and domain-knowledge gaps when processingsuch conversational text. We propose an LLMs-driven event log extraction framework purposebuiltfor customer service conversations. It orchestrates a retrieval–extraction–verificationpipeline: Retrieval-Augmented Generation (RAG) injects external domain knowledge to groundactivity recognition; a Self-Consistency module aggregates ten independent extraction passesthrough confidence-weighted voting. A dedicated verification Large Language Model (LLM)then adjudicates low-confidence predictions, jointly suppressing hallucination and recognitionerrors. Across three public customer-service dialogue datasets, LLMs-based Event Log ExtractionMethod (LLMs-EEM) attains an average F1-score of 93%, surpassing the NLI-BARTbaseline by 4 percentage points and standalone LLM methods by 8 percentage points. Theseresults confirm that the multi-stage collaborative design simultaneously improves extractionaccuracy and robustness in complex conversational environments.