Frye C Corbin, Akhund Ramsha, Murcy Mohammad, Veazey Lillie Grace, McLeod M Chandler, Osborne John D, Cochran Micah, Negrete Haleigh, Tridandipani Srini, Rothenberg Steven, Gillis Andrea, Fazendin Jessica, Chen Herbert, Lindeman Brenessa
Department of Surgery, University of Alabama at Birmingham, Birmingham, Alabama, USA.
University of Alabama at Birmingham, Heersink School of Medicine, Birmingham, Alabama, USA.
World J Surg. 2024 Dec;48(12):2907-2917. doi: 10.1002/wjs.12346. Epub 2024 Sep 17.
Adrenal gland incidentalomas (AGIs) are found in up to 5% of cross-sectional images. However, rates of guideline-based workup for AGIs are notoriously low. We sought to determine if a natural language processing (NLP)-informed AGI clinic could improve the rates of indicated biochemical evaluation and adrenal-specific imaging.
An NLP algorithm was created to detect clinically significant adrenal nodules from radiology reports of cross-sectional images at an academic institution. The NLP algorithm was applied to scans occurring between June 2020 and July 2021 to form a baseline cohort. The NLP algorithm was re-applied to scans from August 2021 to February 2023 and identified patients were invited to join an outpatient clinic dedicated to AGIs. Patients evaluated in the clinic from March 2022 to February 2023 were included in the intervention cohort. Statistical analysis utilized chi-square, t-test, and a multivariable logistic regression.
The baseline and intervention cohorts included 1784 and 322 unique patients, respectively. Patients in the intervention cohort were more likely to be female (59% vs. 51%, p = 0.01), be younger (60 ± 13.1 vs. 64 ± 13.2 years, p < 0.001), have smaller nodules (1.7 cm, IQR 1.4-2.1 vs. 1.8 cm, IQR 1.4-2.5 cm, p = 0.017), have had biochemical workup (99% vs. 13%, p < 0.001), and have had adrenal-specific imaging (40% vs. 11%, p < 0.001). In a multivariable analysis, intervention cohort patients were significantly more likely to have had biochemical workup (odds ratio ,OR 1209, confidence interval ,CI 434-5117, p < 0.001) and adrenal-specific imaging (OR 8.89, CI 6.42-12.4, p < 0.001).
The implementation of an NLP-informed AGI clinic was associated with a seven-fold increase in biochemical workup and a three-fold increase in adrenal-specific imaging in participating patients.
肾上腺偶发瘤（AGIs）在高达5%的横断面影像中被发现。然而，基于指南对AGIs进行检查的比例极低。我们试图确定一个基于自然语言处理（NLP）的AGI诊所是否能提高指定生化评估和肾上腺特异性成像的比例。
创建了一种NLP算法，用于从一所学术机构的横断面影像放射学报告中检测具有临床意义的肾上腺结节。将该NLP算法应用于2020年6月至2021年7月期间的扫描，以形成一个基线队列。将该NLP算法重新应用于2021年8月至2023年2月的扫描，并邀请识别出的患者加入一个专门针对AGIs的门诊。2022年3月至2023年2月在该诊所接受评估的患者被纳入干预队列。统计分析采用卡方检验、t检验和多变量逻辑回归。
基线队列和干预队列分别包括1784名和322名不同患者。干预队列中的患者更可能为女性（59%对51%，p = 0.01）、更年轻（60±13.1岁对64±13.2岁，p < 0.001）、结节更小（1.7cm，四分位间距1.4 - 2.1cm对1.8cm，四分位间距1.4 - 2.5cm，p = 0.017）、接受过生化检查（99%对13%，p < 0.001）以及接受过肾上腺特异性成像（40%对11%，p < 0.001）。在多变量分析中，干预队列患者接受生化检查的可能性显著更高（比值比，OR 1209，置信区间，CI 434 - 5117，p < 0.001）以及接受肾上腺特异性成像的可能性显著更高（OR 8.89，CI 6.42 - 12.4，p < 0.001）。
实施基于NLP的AGI诊所与参与患者的生化检查增加7倍以及肾上腺特异性成像增加3倍相关。
