EUVIMEDEUVIMED entdecken
CrossrefInfokarte

Automation Bias in AI-Assisted Simulated Radiotherapy Decision-Making Among Undergraduate Radiological Sciences Students: A Controlled Experimental Study

Nouf M. Alzahrani, Lojain Abdulaal, Awad Alzahrani, Meshal Alzahrani, Nasser M. Alzahrani, Reem M. Althubaiti, Abir Bouaoun

International Medical Education · 2026-11-09 · 5 (4) · 127

DOI: 10.3390/ime5040127 ↗
AbstractBibliografische Angaben

Abstract

Background: Artificial intelligence (AI) may support clinical decision-making but introduce automation bias when incorrect recommendations influence users. This study aimed to determine the occurrence of harmful automation bias among undergraduate radiological sciences students during AI-assisted radiotherapy decision-making and to examine the influence of time pressure. Methods: In a controlled experiment, 54 final-year undergraduate radiological sciences students with no prior clinical radiotherapy placement experience completed 20 simulated radiotherapy decision-making scenarios with 13 correct and seven deliberately incorrect AI recommendations (1080 trials). Initial and final responses, AI advice acceptance, confidence, perceived workload, and attitudes toward AI verification were assessed. Students were assigned to time-pressure (n = 26) or no-time-pressure (n = 28) conditions. Harmful automation bias was operationally defined as switching from an initially correct response to an incorrect response matching AI advice. Results: Harmful automation bias occurred in 20/220 eligible trials (9.1%), affecting 15/54 students (27.8%). Incorrect AI advice was followed in 137/378 trials (36.2%). Overall accuracy remained similar before and after AI exposure (75.65% vs. 74.91%; p = 0.289). Time pressure increased workload, but no statistically significant association with harmful automation bias was observed (OR = 0.78; p = 0.654). Lower post-decision confidence was associated with harmful automation bias (OR = 0.599; p = 0.028). No statistically significant difference in harmful automation bias was observed between participants who supported AI verification and those who did not. Conclusions: Harmful automation bias can occur despite stable overall accuracy, suggesting that greater emphasis on critical appraisal, active verification, and independent judgement may be valuable in radiotherapy education.

Wissen, das zu Ihrer Arbeit passt.

Vom kostenlosen Einstieg über die persönliche Recherche bis zur vollständigen Research Suite.

Tarife ansehen →

Sind alle Verlagsvolltexte enthalten?

Nein. Bibliografische Nachweise, Abstracts und Volltexte sind unterschiedliche Inhalte. Ob ein Volltext verfügbar ist, hängt von der Quelle, Open Access und Ihren vorhandenen Lizenzen ab. Ein EUVIMED-Tarif ersetzt keine zusätzlichen Verlagsrechte.

Tarife ansehen →