26kgg107 10.3205/26kgg107 urn:nbn:de:0183-26kgg1079 Meeting Abstract Erdös Erdös Judit J

Austrian Institute for Health Technology Assessment (AIHTA), Österreich

author Grabenhofer Grabenhofer Lena Malin LM

Austrian Institute for Health Technology Assessment (AIHTA), Österreich

author Götz Götz Gregor G

Austrian Institute for Health Technology Assessment (AIHTA), Österreich

author German Medical Science GMS Publishing House

Düsseldorf

610 20251030 engl This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. Dieser Artikel ist ein Open-Access-Artikel und steht unter den Lizenzbedingungen der Creative Commons Attribution 4.0 License (Namensnennung). M0662 107 Deutsche Gesellschaft für Allgemeinmedizin und Familienmedizin Deutsche Gesellschaft für Sozialmedizin und Prävention Netzwerk Evidenzbasierte Medizin e.V. Deutsche Gesellschaft für Medizinische Soziologie 60. Kongress für Allgemeinmedizin und Familienmedizin Gesundheit gestalten Digitalisierung und hausärztl. Praxis Göttingen 20260930 20261002 V-19-05 TextBackground: Artificial intelligence (AI) for chest X-ray interpretation is increasingly discussed in the diagnostic evaluation of suspected lung cancer. However, it remains unclear whether improvements in diagnostic performance translate into clinically meaningful benefits in real-world practice.Aim/scientific question: To assess the clinical relevance of AI-assisted chest X-ray interpretation for lung cancer detection in symptomatic patients based on available evidence.Topic for discussion: How should improvements in diagnostic performance be interpreted, and what study designs are required to demonstrate patient-relevant benefit and enable evidence linkage?Content: A health technology assessment (HTA) of AI-assisted chest X-ray interpretation for lung cancer detection was conducted. The available studies were primarily retrospective, had a considerable risk of bias, and focused on diagnostic performance outcomes. Results were heterogeneous. In some studies, AI support increased sensitivity compared to radiologists alone, while specificity did not improve consistently. Consequently, positive predictive values were often low in the study settings, suggesting a potential increase in false-positive findings under certain conditions. Some studies reported high negative predictive values for stand-alone AI systems, but sensitivity and specificity were variable. No study evaluated the impact of AI use on patient-relevant outcomes such as mortality, morbidity, or quality of life, and no evidence was available on downstream effects such as additional imaging, invasive procedures, or potential overdiagnosis. Furthermore, key aspects for clinical implementation, including organisational, economic, or workflow-related aspects, were not reported.Take Home Message: Current evidence on AI-assisted chest X-ray interpretation is limited to diagnostic performance and does not provide evidence of benefits for patient-relevant outcomes or clinical practice. Increased detection rates may lead to additional follow-up examinations without evidence of patient-relevant benefit. The evidence base is limited by retrospective study designs and a lack of validation in real-world clinical settings. Prospective studies evaluating patient-relevant outcomes and real-world implementation are required before routine use.Competing interests: None 0 0 0 0