Grolleau Emmanuel, Couraud Sébastien, Jupin Delevaux Emilien, Piegay Céline, Mansuy Adeline, de Bermont Julie, Cotton François, Pialat Jean-Baptiste, Talbot François, Boussel Loïc
University of Lyon, Claude Bernard University, 43 boulevard du 11 Novembre 1918, 69100 Villeurbanne, France; Acute Respiratory Disease and Thoracic Oncology Department, Lyon Sud Hospital, Hospices Civils de Lyon, 165 Chemin du Grand Revoyet, 69495 Oullins-Pierre-Bénite, France.
University of Lyon, Claude Bernard University, 43 boulevard du 11 Novembre 1918, 69100 Villeurbanne, France; Acute Respiratory Disease and Thoracic Oncology Department, Lyon Sud Hospital, Hospices Civils de Lyon, 165 Chemin du Grand Revoyet, 69495 Oullins-Pierre-Bénite, France; EMR-3738 Therapeutic Targeting in Oncology, Lyon Sud Hospital, Hospices Civils de Lyon, 165 Chemin du Grand Revoyet, 69495 Oullins-Pierre-Bénite, France.
Respir Med Res. 2024 Nov;86:101136. doi: 10.1016/j.resmer.2024.101136. Epub 2024 Aug 22.
Pulmonary nodules are a common incidental finding on chest Computed Tomography scans (CT), most of the time outside of lung cancer screening (LCS). We aimed to evaluate the number of incidental pulmonary nodules (IPN) found in 1 year in our hospital, as well as the follow-up (FUP) rate and the clinical and radiological features associated with FUP.
We trained a Natural Language Processing (NLP) tool to identify the transcripts mentioning the presence of a pulmonary nodule, among a large population of patients from a French hospital. We extracted nodule characteristics using keyword analysis. NLP algorithm accuracy was determined through manual reading from a sample of our population. Electronic health database and medical record analysis by clinician allowed us to obtain information about FUP and cancer diagnoses.
In this retrospective observational study, we analyzed 101,703 transcripts corresponding to the entire CTs performed in 2020. We identified 1,991 (2 %) patients with an IPN. NLP accuracy for nodule detection in CT reports was 99 %. Only 41 % received a FUP between January 2020 and December 2021. Patient age, nodule size, and the mention of the nodule in the impression part were positively associated with FUP, while nodules diagnosed in the context of COVID-19 were less followed. 36 (2 %) lung cancers were subsequently diagnosed, with 16 (45 %) at a non-metastatic stage.
We identified a high prevalence of IPN with a low FUP rate, encouraging the implementation of IPN management program. We also highlighted the potential of NLP for database analysis in clinical research.
肺结节是胸部计算机断层扫描（CT）中常见的偶然发现，大多数情况下并非在肺癌筛查（LCS）时发现。我们旨在评估我院一年内发现的偶然肺结节（IPN）数量、随访（FUP）率以及与随访相关的临床和放射学特征。
我们训练了一种自然语言处理（NLP）工具，以从一家法国医院的大量患者中识别提及肺结节存在的记录。我们使用关键词分析提取结节特征。通过对我们人群样本的人工阅读来确定NLP算法的准确性。临床医生对电子健康数据库和病历的分析使我们能够获得有关随访和癌症诊断的信息。
在这项回顾性观察研究中，我们分析了2020年进行的全部CT对应的101,703条记录。我们识别出1991名（2%）有IPN的患者。CT报告中结节检测的NLP准确性为99%。在2020年1月至2021年12月期间，只有41%的患者接受了随访。患者年龄、结节大小以及印象部分对结节的提及与随访呈正相关，而在新冠肺炎背景下诊断出的结节随访较少。随后诊断出36例（2%）肺癌，其中16例（45%）处于非转移阶段。
我们发现IPN的患病率较高，但随访率较低，这促使实施IPN管理计划。我们还强调了NLP在临床研究数据库分析中的潜力。
