Clinicians’ cancer risk assessment among patients with pulmonary nodules: a qualitative study
Year: 2026
Type: article
Abstract: To the Editor: Pulmonary nodules present clinicians with a diagnostic dilemma.1 While a nodule may represent an early-stage and potentially curable lung cancer, under 10% of all pulmonary nodules are malignant,2 and a definitive diagnosis can only be made via a lung biopsy or surgical resection, both of which carry inherent procedural risks. Thus, clinical guidelines recommend cancer risk assessment to guide pulmonary nodule decision-making, with the goal of reserving biopsy or surgery only for those with higher cancer risk and a more favorable benefit-risk ratio.3 However, there exists no consensus on how best to perform risk assessment, resulting in marked heterogeneity in pulmonary nodule management.4 This variability persists despite the contemporaneous development and validation of several risk assessment tools designed to aid clinicians in performing pulmonary nodule risk assessment,2,5–9 including (1) clinical risk prediction models, which use statistical modeling to incorporate various patient and pulmonary nodule characteristics,5 and (2) commercial biomarkers, which employ blood-, respiratory-, or computed tomography (CT) radiomics-based assays.6,7 While several prior studies have assessed the diagnostic accuracy and clinical utility of such risk assessment tools,2,8,9 key knowledge gaps exist regarding clinicians’ perspectives on and implementation of these tools into practice. Given the current clinical equipoise in pulmonary nodule evaluation and complex, growing landscape of contemporary risk assessment tools, we sought to identify factors that influence clinicians’ pulmonary nodule risk assessment and determine perceptions and use of risk assessment tools. Semistructured interviews were performed from August 2024 to February 2025 with clinicians in the United States and Canada who manage outpatients with pulmonary nodules. We identified potentially eligible participants via professional society and clinical networks and approached them via recruitment emails in batches, using purposive sampling to ensure representation across professional roles (attending physicians, fellow physicians, and advanced practice providers [APPs]), specialties (general pulmonology [GP], interventional pulmonology [IP], and thoracic surgery [TS]), and geography (Northeast, Midwest, South, and West). Participants were recruited until data saturation was achieved and received a $50 ClinCard for completing interviews. This study was exempt from full review by the University of Pennsylvania Institutional Review Board (protocol #856438). We followed the Consolidated Criteria for Reporting Qualitative Health Research. Using an interview guide (available upon request to the corresponding author) developed by 3 investigators (R.Y.K., A.G., and K.A.R.) with collective expertise in qualitative methods and pulmonary nodule management, 2 investigators (R.Y.K. and A.G.) trained in qualitative interviewing conducted individual interviews. After verbal consent was obtained, participants completed a structured question survey4 before completing audio-recorded interviews on the pulmonary nodule risk assessment process, usage of risk assessment tools, and the ideal risk assessment tool. Deidentified data were transcribed verbatim through Datagain Transcription Services and quality checked. Investigators (R.Y.K., A.G., and L.A.) identified emergent themes and iteratively developed a codebook using inductive and deductive approaches by combining grounded theory with the Consolidated Framework for Implementation Research model. Using NVivo 15 software, 20% (n = 8) of transcripts were independently coded by 2 investigators (R.K. and L.A). After high coding reliability was achieved (κ = 0.92), one investigator (L.A.) used the final codebook to complete coding. Differences in interview themes by clinical subspecialty were explored. Of the 47 clinicians invited to participate, 40 (85%) completed interviews (Table 1). The median age was 40 years (IQR, 38-47 years), 22 (55%) identified as women, 30 (75%) as White race, and 27 (68%) as attending physicians. Most participants practiced in an academic (78%), urban setting (85%), and the Northeast region (55%). One transcript was lost during data collection. Thus, 39 transcripts were included in the qualitative analysis. Participant characteristics. Data are presented as No. (%) unless otherwise indicated. Abbreviations: APRN, advanced practice registered nurse; DO, doctor of osteopathic medicine; HMO, health maintenance organization; IQR, interquartile range; MD, doctor of medicine; PA-C, physician assistant–certified. Geographic regions were defined as Northeast (Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont, and eastern Canada), Midwest (Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, and Wisconsin), South (Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, and West Virginia), and West (Alaska, Arizona, California, Colorado, Hawaii, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming). Clinicians consistently reported integrating pulmonary nodule, patient, and population (ie, geographic or clinic-specific) characteristics, but variably incorporated clinical guidelines and risk assessment tools to categorize cancer risk as either low, intermediate, or high. Compared to those in IP or TS, GP clinicians more frequently referenced guidelines as influencing their risk assessment, citing the Fleischner Society and Lung-RADS guidelines most. IP clinicians largely relied on clinical intuition or “gestalt.” “To tell you the truth, a lot of it is gestalt more than anything” (participant 7, IP). TS clinicians emphasized pulmonary nodule characteristics. “I look at how it looks on the CT scan … if I have a patient [with] a 1.5 centimeter nodule [and] we don’t know what it is, that person’s probably headed to the OR” (participant 18, TS). On a 5-point scale, the median rating for overall clinical impact of risk assessment tools was 3 (IQR, 2.5-4). Clinicians used clinical risk prediction models more frequently than commercial biomarkers, with the Mayo Clinic and Brock models mentioned most. However, use frequency ranged from “most of the time” to “never.” IP clinicians used models most while TS clinicians least. Rather than using them to determine cancer risk assessment, clinicians instead used models to (1) cross-check clinical intuition, (2) communicate risk to patients, and (3) train less experienced clinicians. Participants identified discrete opportunities for improvement: (1) inconvenience of use, (2) limited impact on medical decision-making, and (3) lack of generalizability to specific patient populations (Table 2). Few clinicians used commercial biomarkers, although the Nodify Lung, Percepta, and Lung Cancer Prediction Convolutional Neural Network tests were mentioned. Respondents generally expressed uncertainty regarding biomarkers in their current state. “I have not incorporated a lot of biomarkers … I’m just not impressed with what they can give us” (participant 16, IP). Clinical risk prediction model: common uses and opportunities for improvement. Abbreviations: GP, general pulmonology; IP, interventional pulmonology; P, participant; PET, positron emission tomography; TS, thoracic surgery. Tasked with envisioning the ideal risk assessment tool (Table 3), 70% of participants agreed that it should prioritize not missing a lung cancer diagnosis (ie, increased sensitivity, reduced false negative rate) over avoiding a potentially unnecessary procedure (ie, increased specificity, reduced false positive rate). Clinicians frequently cited comprehensive data integration, efficiency, and artificial intelligence as critical for ideal tool development. “I think the ideal tool would incorporate clinical characteristics, biomarkers when they become better validated and integrated into our workflow, and radiomic characteristics … I think that is how we’re going to be able to have a better marker” (P31-IP). Ideal risk assessment tool prioritization. Abbreviations: GP, general pulmonology; IP, interventional pulmonology; NLST, National Lung Screening Trial; P, participant; TTNA, transthoracic needle aspiration. In this qualitative study, clinicians inconsistently integrated clinical guidelines and risk assessment tools into pulmonary nodule evaluation. Among those who did use risk assessment tools, more used clinical risk prediction models than commercial biomarkers. Rather than guiding medical decision-making, risk assessment tools were instead used to cross-check clinical intuition, communicate risk to patients, and train inexperienced clinicians. Challenges plaguing current risk assessment tools included inconvenience of use, limited impact on medical decision-making, and inadequate generalizability. Regarding an ideal risk assessment tool, priorities included sensitivity over specificity, user-friendliness, efficiency, comprehensive risk assessment, artificial intelligence, and integration into existing clinical workflows. While prior qualitative and survey studies have focused on patient perspectives and shared decision-making for pulmonary nodule evaluation,4,10 to our knowledge this is the first qualitative study to provide an in-depth description of clinicians’ contemporary approach to cancer risk assessment and use of risk assessment tools among patients with pulmonary nodules. A key finding from this study is that clinicians who routinely manage pulmonary nodules do not primarily use risk assessment tools to guide cancer risk assessment or medical decision-making. Instead, clinicians described using these tools for alternative purposes for which they were not initially designed, such as patient communication and clinical education. We identified several perceived barriers to routine risk assessment tool usage, which highlight targets for improvement, including increased efficiency and ease of use, demonstrated impact on clinical decision-making, and generalizability to patient populations not currently represented in existing validation cohorts. Thus, it is imperative that future studies evaluating pulmonary nodule risk assessment tools address clinical effectiveness (ie, impact on influencing management decision-making), implementation strategies, and alternative uses, such as patient communication and clinical education. Our study has limitations. First, it was limited to a small number of participants with expertise in pulmonary nodule evaluation and management, which could limit the generalizability of our findings. However, we included attending physicians, physicians-in-training, and APPs among 3 clinical subspecialties across North America and achieved thematic saturation, enabling us to capture a range of perspectives. Moreover, we are currently conducting a US nationwide survey to supplement these qualitative data with quantitative results. Additionally, we did not interview patients as part of the current study, which limited our ability to understand the patient point of view regarding pulmonary nodule risk assessment. Future studies are required to more fully explore themes regarding shared decision-making and patient perspectives on modern-day pulmonary nodule evaluation and management. In conclusion, there exists heterogeneity in pulmonary nodule risk assessment and tool use. The factors influencing approaches to cancer risk assessment identified in this study may provide opportunities for clinicians, researchers, industry partners, and health system leaders alike to develop, test, and implement innovative strategies to improve upon current pulmonary nodule evaluation practices. Conceptualization, investigation, methodology, and funding acquisition: R.Y.K. Data curation, formal analysis: R.Y.K., A.G., and L.A. Methodology: A.B. and K.A.R. Supervision: A.V. and K.A.R. Visualization: R.Y.K. and L.A. Writing—original draft: R.Y.K. Writing—review & editing: All authors. R.Y.K. directly accessed and verified the underlying data reported in the manuscript. Supplementary material is available at Annals of the American Thoracic Society online. Please see the ICMJE disclosure forms, which have been provided as supplementary material. R.Y.K. is an Associate Editor for Annals of the American Thoracic Society and Chair of the American Thoracic Society Thoracic Oncology Assembly Early Career Professionals Working Group, has stock in Amgen, Inc., is a consultant for Guidepoint, and receives research grant funding to his institution from Optellum Ltd, Veracyte, DELFI, and the National Cancer Institute of the National Instiutes of Health. A.V. a consultant for Intuitive Surgical and Johnson & Johnson and receives research grant funding to his institution from Median Technologies, Optellum Ltd., Astra Zeneca, the National Comprehensive Cancer Network, Lungevity Foundation, and the National Cancer Institute of the National Institutes of Health. K.A.R. receives research grant funding to her institution from Pfizer, Astra Zeneca, Lung Cancer Research Foundation, and the National Comprehensive Cancer Network. Research reported in this manuscript was supported by the American Cancer Society (ACS IRG-22–150-41-IRG) and the National Cancer Institute of the National Institutes of Health (K08CA279881). Kim RY, Gupta A, Bilger A, et al. A qualitative analysis of clinicians’ approaches to pulmonary nodule evaluation and management. In: C31. Breaking Through the Fog: AI, IP and Novel Insights to Improve Lung Cancer Diagnosis and Management. American Thoracic Society International Conference, San Francisco, CA, May 20, 2025. No artificial intelligence tools were used in writing this manuscript.
Language: English
Subfield: Pulmonary and Respiratory Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7125481919