BetaEntity Annotation Prototype
← Back to institutions

Annotated abstract

P94 Why iLFT? Identifying requesting trends using large language processing of >27,000 iLFT requests

gutjnl · 2025-10-06 · canonical JSON source

3 visible annotations · policy: published · automated confidence ≥ 75.00%

Document resource

Background Intelligent liver function testing (iLFT) was introduced in primary care in NHS Tayside in 2018 as a clinical decision-making tool to improve diagnostic accuracy and standardise treatment recommendations. If initial LFT results are abnormal, iLFT automatically reflex cascades to a series of aetiologically-specific investigations.Aim This study aims to investigate the indications for iLFT requests made from 2018–2023 and identify relevant temporal changes.Methods Language processing tools within KNIME analytics platform (KNIME AG, V4.1.3) were used to analyse all 27,664 iLFT requests made between August 2018 and July 2023. Requesting information inputted by clinicians was cleaned, including spell-checking and correcting common typing errors, term normalisation (e.g., replacing ‘excessive’, ‘elevated’, with ‘raised’), punctuation and stop word removal, and case standardisation.Automatically generated requesting artifacts from simultaneous additional investigation requests, including thyroid function test, CA125, qFIT, testosterone, UKELD score, anticoagulation screen and d-dimer, were removed.Data were processed by creation of n-grams, phrases of ‘n’ consecutive words, and identification of their document frequencies [DF] (the number of requests containing the phrase). 1-grams, 2-grams, and 3-grams were generated for all 27,664 requests and filtered for those with DF >100. The process was repeated for years 2018 and 2023, but only n-grams with DF >1% of requests that year were analysed. All n-grams were reviewed to remove nonsensical phrases. Overlapping terms were identified by Python (Python Software Foundation, Version 3.13) allowing adjustment of DFs to avoid double counting (e.g. ‘ALT’ DF reduced from 67 to 27 to exclude instances within ‘raised ALT’ which has DF of 40). Finally, n-grams and their DFs were examined to identify trends.Results ‘Raised ALT’ was the most common indication for iLFT in 2018, appearing in 26.8% of requests. In 2023, ‘tired’ became the most frequent, appearing in 24.6% of cases. Liver-related phrases including ‘raised alcohol’, ‘raised bilirubin’, ‘fatty liver’ and ‘ALT raised’ featured prominently in 2018, however, were absent from the top 20 in 2023. Instead, non-specific terms— ‘pain’, ‘review’, ‘paraesthesia’, ‘cough’ and ‘anaemia’— emerged more frequently. The proportion of iLFTs accompanied by additional investigations also increased from 18.3% to 55.9%, suggesting growing use of iLFT as a generic screening tool rather than for targeted liver assessment.Conclusion From 2018–2023, iLFT became more embedded and the cascading rate decreased because of changing practices to common usage over routine LFT requesting. These findings indicate a shift from liver-focused indications to broader, undifferentiated presentations commonly encountered in primary care.Abstract P94 Figure 1