Abstract Introduction/Rationale Artificial intelligence (AI) tools could enhance the continuum of lung cancer care, including screening, diagnosis, and treatment. Yet physician perspectives regarding AI use during lung cancer care are uncharacterized, which could hinder implementation. Methods We conducted a national survey of Veterans Health Administration specialty physicians who care for patients with lung cancer. The survey was developed by adapting published survey questions assessing physician perspectives of AI to specifically ask about AI in lung cancer care. We reported descriptive statistics with medians and interquartile ranges (IQR) or counts and percentages. Responses from 5-point Likert scale items were summarized into three response categories: disagree, neutral, agree. We used multivariable linear regression models to test the association between frequency of AI use in personal life with overall excitement (0-10 scale, 0=least excited, 10=most excited) and concern (0-10 scale, 0=least concerned, 10=most concerned) about AI use in clinical care, while adjusting for participant characteristics (volume of lung cancer patients, years since residency, specialty, practice setting). We qualitatively analyzed open-ended responses. Results 330 of 1468 (23%) physicians responded. Participants represented pulmonology (54%), medical oncology (25%), radiation oncology (9.7%), thoracic surgery (7%), and other (4%). Figure 1 demonstrates physician perspectives about AI use in lung cancer care. We found a median excitement score for AI of 7 (IQR 3), and a median concern score for AI of 6 (IQR 3). Compared to those who use AI daily in their personal life, those who use AI less frequently had less excitement for AI in clinical care (βAI use few times/week -0.96 95% CI -1.70 to -0.22, βAI use few times/month -1.12 95% CI -1.89 to -0.42, βAI use once/month -1.99 -2.93 to -1.04, and βAI use never -2.02 -2.83 to -1.20), and greater concern about AI in clinical care (βAI use few times/week 0.15 95% CI -0.64 to 0.93, βAI use few times/month 0.56 95% CI -0.21 to 1.33, βAI use once/month 1.11 0.11 to 1.20, and βAI use never 1.19 0.33 to 2.05). In open-ended responses participants expressed that AI should augment, not replace, physician decision-making and that rigorous testing, continuous validation, and physician training (e.g., sandbox) on how to use AI tools would bolster trust. Conclusions Physicians demonstrated mixed excitement and concern regarding AI tools to support lung cancer care. Future implementation work should include strategies that address concerns, provide evidence on tool performance, and help providers understand how to use AI responsibly in practice. This abstract is funded by: VA National Oncology Program Office
Rucci et al. (Fri,) studied this question.