pmc Nat Med Nat Med 981 npgopen 9502015 Nature Medicine 1078-8956 1546-170X pmc-is-collection-domain yes pmc-collection-title Nature Portfolio PMC13190311 PMC13190311.1 13190311 13190311 41876649 10.1038/s41591-026-04253-5 4253 1 Article AI-based chest X-ray prioritization in the lung cancer diagnostic pathway: the LungIMPACT randomized controlled trial http://orcid.org/0000-0001-9598-189X Woznitza Nick 1 2 Smith Lesley 3 http://orcid.org/0000-0002-9536-9809 Rawlinson Janette 4 http://orcid.org/0009-0001-6422-7752 Au-Yong Iain 5 http://orcid.org/0009-0007-8006-9575 George Bindu 5 http://orcid.org/0000-0003-3408-1411 Djearaman Madava G. 6 Nair Arjun 1 http://orcid.org/0000-0001-8368-3406 Lee Richard W. 7 8 9 http://orcid.org/0000-0002-6412-7516 Navani Neal 2 10 Ndwandwe Siyabonga 11 http://orcid.org/0000-0002-4676-1257 Clarke Caroline S. 11 Creeden Andrew 12 Newsome Josh 12 Das Indrajeet 12 http://orcid.org/0009-0001-5031-111X Abaokporo Sylvia 5 Tucker Richard 5 Hathorn James 13 http://orcid.org/0000-0001-8410-7160 Baldwin David R. david.baldwin12@nhs.net 14 15 1 https://ror.org/042fqyp44 grid.52996.31 0000 0000 8937 2257 University College London Hospitals NHS Foundation Trust, London, UK 2 https://ror.org/02jx3x895 grid.83440.3b 0000 0001 2190 1201 Lungs for Living Research Centre, University College London, London, UK 3 https://ror.org/024mrxd33 grid.9909.9 0000 0004 1936 8403 Leeds Cancer Research UK Clinical Trials Unit, University of Leeds, Leeds, UK 4 Tipton, UK 5 https://ror.org/05y3qh794 grid.240404.6 0000 0001 0440 1889 Department of Radiology, Nottingham University Hospitals, Nottingham, UK 6 https://ror.org/014ja3n03 grid.412563.7 0000 0004 0376 6589 Radiology Department, University Hospitals Birmingham NHS Foundation Trust, Birmingham, UK 7 https://ror.org/0008wzh48 grid.5072.0 0000 0001 0304 893X The Early Diagnosis and Detection Centre, The Royal Marsden NHS Foundation Trust, London, UK 8 https://ror.org/043jzw605 grid.18886.3f 0000 0001 1499 0189 Institute of Cancer Research, London, UK 9 https://ror.org/041kmwe10 grid.7445.2 0000 0001 2113 8111 National Heart and Lung Institute, Imperial College London, London, UK 10 https://ror.org/02jx3x895 grid.83440.3b 0000 0001 2190 1201 Department of Thoracic Medicine, University College London Hospital, London, UK 11 https://ror.org/02jx3x895 grid.83440.3b 0000 0001 2190 1201 Research Department of Primary Care and Population Health, University College London, London, UK 12 https://ror.org/02fha3693 grid.269014.8 0000 0001 0435 9078 Radiology Department, University Hospitals of Leicester NHS Trust, Leicester, UK 13 https://ror.org/019g08z42 grid.507581.e Radiology Department, East Suffolk and North East Essex NHS Foundation Trust, Colchester, UK 14 https://ror.org/05y3qh794 grid.240404.6 0000 0001 0440 1889 Respiratory Medicine, Nottingham University Hospitals NHS Trust, Nottingham, UK 15 https://ror.org/01ee9ar58 grid.4563.4 0000 0004 1936 8868 University of Nottingham, Nottingham, UK 24 3 2026 2026 32 5 514029 1737 1744 6 10 2025 26 1 2026 24 03 2026 22 05 2026 22 05 2026 © The Author(s) 2026 2026 https://creativecommons.org/licenses/by/4.0/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . Prioritizing artificial intelligence (AI)-detected imaging findings may reduce the time to diagnosis of lung cancer. This prospective, multicentre, randomized controlled trial tested whether immediate AI prioritization of primary care-requested chest X-rays (CXR) influenced time to computed tomography (CT) and lung cancer diagnosis, the primary outcomes. Secondary outcomes included the number of urgent suspected lung cancer referrals, incidence and stage of lung cancer, times to urgent referral and treatment, concordance between AI and radiology reports, and algorithm accuracy. AI was available in both study arms, with AI prioritization randomized by day. Of 97,731 participant CXRs, 4,405 were excluded due to data compliance issues or failure of randomization, resulting in 93,326 CXRs analyzed (45,987 and 47,339 in the prioritization ‘on’ or ‘off’ arms, respectively). A total of 13,347 CTs were identified, with 2,766 performed within 14 days of CXR. Median (interquartile range) times to CT were 53 days (17–145) and 53 days (19–141), with and without AI prioritization, corresponding to a ratio of geometric means of 0.97 (95% confidence interval (CI) = 0.93–1.02; P = 0.31). When restricted to CTs performed within 14 days of CXR, the median time to CT was 8 days (5–11) in both groups. Lung cancer was diagnosed in 558 people (0.6% of CXRs). Median times to diagnosis were 44 days (26–90) and 46 days (24–105) respectively, with a ratio of geometric means of 0.98 (95% CI = 0.83–1.16; P = 0.84). No significant differences were observed in time to lung cancer referral (14 versus 15 days; P = 0.13), time to treatment (76 versus 72.5 days; P = 0.99) or stage at diagnosis ( P = 0.34). Discordance between AI and radiology reports occurred in 28,261 CXRs (30.3%) and expert radiology review identified actionable findings in 6,750 cases (23.9%). AI prioritization of CXR requested by UK primary care has no significant impact on the lung cancer pathway. Therefore, CXR AI deployments should not include worklist prioritization in this context. Future research should differentiate between primary pathway changes and the direct impact of AI. ISRCTN registration: 78987039 . In a large UK-based primary care randomized trial, immediate artificial intelligence (AI)-driven prioritization of chest X-rays did not significantly shorten time to computed tomography or lung cancer diagnosis versus standard workflow, suggesting that AI-based prioritization alone is unlikely to accelerate the lung cancer diagnostic pathway. Subject terms Medical imaging Outcomes research Respiratory signs and symptoms Non-small-cell lung cancer Computational models Small Business Research Initiative Healthcare. No SBRIC01P3039 pmc-status-qastatus 0 pmc-status-live yes pmc-status-embargo no pmc-status-released yes pmc-prop-open-access yes pmc-prop-olf no pmc-prop-manuscript no pmc-prop-legally-suppressed no pmc-prop-has-pdf yes pmc-prop-has-supplement yes pmc-prop-pdf-only no pmc-prop-suppress-copyright no pmc-prop-is-real-version no pmc-prop-is-scanned-article no pmc-prop-preprint no pmc-prop-in-epmc yes pmc-license-ref CC BY issue-copyright-statement © Springer Nature America, Inc. 2026 Main Lung cancer accounts for the highest proportion of cancer-related deaths in the United Kingdom (UK) and worldwide, primarily because it is common, is diagnosed at a late stage and/or late in the symptomatic period when patients have become too unwell for treatment. Survival rates in the UK are lower than those from high-income countries with comparable cancer registration data and it is thought that much of this is due to later diagnosis 1 – 3 . Diagnostic pathway delay can worsen outcomes in both early-stage and late-stage disease and increase patient and carer anxiety 4 – 6 . Patient feedback gathered from focus groups conducted during the study also supports the need for timely reporting. Patients who undergo investigations are often anxious about the results, with research suggesting that the time between testing and receiving the results is particularly worrying 7 – 10 . National cancer care pathways provide a method to accelerate and standardize the end-to-end patient journey 11 , 12 . In England, the National Optimal Lung Cancer Pathway (NOLCP) recommends rapid progression from chest X-ray (CXR) to computed tomography (CT), followed by assessment in a specialist clinic. The NOLCP mandates that patients with a suspicious CXR undergo CT within 72 h, preferably on the same day. This aims to reduce delays experienced by patients, including those referred through alternative routes 12 . A single attendance for initial diagnostic investigations may reduce delays in subsequent appointments and eliminate the chance of communication not being received or understood. Recently published research funded by Cancer Research UK, conducted by members of the current study team, found that immediate radiographer CXR reporting, with direct triage to same-day CT, significantly reduced time to diagnosis of lung cancer by almost half from a median of 63 days from CXR, to 32 days ( P = 0.03) compared to routine CXR reporting once the patient had left the department 13 . This demonstrated that immediate CXR reporting and prioritization are feasible and may accelerate early symptomatic diagnosis of lung cancer, which remains a major challenge despite clear referral criteria from the National Institute for Health and Care Excellence (NICE) 14 , 15 . Current UK guidance also recommends increasing the use of CXR, as there is some evidence that this improves survival and the proportion of early-stage lung cancer diagnoses 15 – 18 . Artificial intelligence (AI) algorithms for assisted reporting of CXRs have been proposed to improve both the speed and accuracy of diagnosis and to help the overstretched radiology workforce. However, a recent review of the clinical utility of this technology by NICE concluded that there was insufficient evidence to make any recommendations other than to ensure products were carefully evaluated 19 . NICE cited the LungIMPACT study as the only investigation considered potentially capable of addressing important clinical questions 19 . The primary aim of the trial was to measure the impact of immediate AI-driven prioritization of abnormal CXRs for reporting on the time to CT and diagnosis of lung cancer. Secondary outcomes included impact on other steps in the pathway and detailed discordance reviews of AI in both arms. Results Descriptive statistics and consort diagram Between 17 July 2023 and 31 December 2024, a total of 97,731 CXRs were performed across five diverse National Health Service (NHS) Trusts, with a geographical spread and a mix of general and tertiary, as well as high-volume and low-volume, centers. Patients were followed up until 5 June 2025. Figure 1 shows the CONSORT diagram of the study population. After data cleaning, 93,326 CXRs from 86,945 patients were included in the study analysis. Of these, 45,987 (49.3%) CXRs were performed in sessions with AI prioritization for immediate review. Fig. 1 Study population for CXRs CONSORT diagram. The total number of CXRs assessed for eligibility, the numbers analyzed for the primary outcomes, and the number of AI discordances in each arm of the study. Table 1 presents the characteristics and demographics of the study population stratified by AI prioritization day. The mean age of the study population was 59 years and 46% were male. University Hospitals of Leicester performed 54% of the study CXRs, Nottingham University Hospitals 15%, East Sussex and North Essex Foundation Trust 11%, University College London Hospital 11% and University Hospitals of Birmingham 9%. Table 1 Study population (CXR) by AI prioritization Study population AI prioritization day No Yes Total ( n = 47,339) ( n = 45,987) ( n = 93,326) Age (years) 59.1 (17.4) 59.1 (17.5) 59.1 (17.4) Sex Male 21,987 (46.4%) 21,278 (46.3%) 43,265 (46.4%) Female 25,352 (53.6%) 24,709 (53.7%) 50,061 (53.6%) Hospital trust University College London Hospital 5,216 (11.0%) 5,360 (11.7%) 10,576 (11.3%) University Hospitals of Leicester 25,772 (54.4%) 24,287 (52.8%) 50,059 (53.6%) East Sussex and North Essex Foundation Trust 5,508 (11.6%) 5,148 (11.2%) 10,656 (11.4%) University Hospitals of Birmingham 3,878 (8.2%) 4,064 (8.8%) 7,942 (8.5%) Nottingham University Hospitals 6,965 (14.7%) 7,128 (15.5%) 14,093 (15.1%) Quarter Q3 2023 597 (1.3%) 632 (1.4%) 1,229 (1.3%) Q4 2023 6,127 (12.9%) 5,850 (12.7%) 11,977 (12.8%) Q1 2024 9,326 (19.7%) 9,839 (21.4%) 19,165 (20.5%) Q2 2024 12,986 (27.4%) 12,277 (26.7%) 25,263 (27.1%) Q3 2024 8,876 (18.7%) 8,535 (18.6%) 17,411 (18.7%) Q4 2024 9,427 (19.9%) 8,854 (19.3%) 18,281 (19.6%) Values are mean (s.d.) or n (%). Primary outcomes Among all patients, 13,347 had a valid CT scan and were included in the analysis, with a time window from CXR acquisition to CT scan analysis—6,674 with AI prioritization and 6,673 with no AI prioritization. A total of 558 patients were diagnosed with lung cancer and were included in the analysis of time to cancer diagnosis, 269 with AI prioritization and 289 with no AI prioritization. The median time to CT scan for all patients was 53 days (interquartile range (IQR) = 17–145 days) for AI prioritization days and 53 days (IQR = 19–141 days) for those with no AI prioritization. Similarly, the median time to lung cancer diagnosis was 44 days (IQR = 26–90 days) for AI prioritization days and 46 days (IQR = 24–105 days) for those with no AI prioritization (Table 2 ). There was no significant difference in the time from CXR acquisition to CT scan according to AI prioritization, with a ratio of geometric means of 0.97 (95% confidence interval (CI) = 0.93–1.02, P = 0.31). There was no difference in time to CT when restricted to CT scans performed within 14 days and CXRs with a CT referral ( P = 0.96 and P = 0.86, respectively; Table 2 ). There was no significant difference in the time from CXR acquisition to lung cancer diagnosis between AI prioritization days, with a ratio of geometric means of 0.98 (95% CI = 0.83–1.16, P = 0.84; Table 2 ). No statistically significant differences were found for either primary outcome at any site (Extended Data Table 1 ) or for any quarter (Extended Data Table 2 ). Primary outcomes by sex are presented in Extended Data Table 4 . Table 2 Time outcomes by AI prioritization, median and IQR Outcomes AI prioritization No Yes 95% CI P value ( n = 44,112) a ( n = 42,833) a n Median IQR n Median IQR Ratio of geometric means Primary outcomes Time to CT scan (days) All CT scans 6,673 53 (19–141) 6,674 53 (17–145) 0.97 (0.93–1.02) 0.31 CT scans within 14 days of CXR 1,314 8 (5–11) 1,452 8 (5–11) 1.00 (0.91–1.10) 0.96 CXRs coded to generate a CT referral 495 7 (4–9) 505 6 (4– 9) 1.02 (0.81–1.28) 0.86 Time to lung cancer diagnosis (days) 289 46 (24–105) 269 44 (26– 90) 0.98 (0.83–1.16) 0.84 Secondary outcomes Time to 2WW referral (days) 1,284 15 (6–50) 1,215 14 (5–53) 0.91 (0.80–1.03) 0.13 Time from CXR to CXR report (h) 44,078 47.0 (15.8–99) 42,814 34.1 (6.6–93.1) 0.85 (0.83–0.87 <0.001 Time to cancer treatment starting (days) 200 72.5 (43–120.5) 200 76 (37–114) 1.00 (0.84–1.19) 0.99 Statistical analyses were performed using a two-sided t- test on log-transformed outcomes and presented as the ratio of geometric means and 95% CIs. a Only includes the first CXR for patients with multiple CXRs in the study period. Secondary outcomes The median time to two-week-wait (2WW) referral was 14 days (IQR = 5–53 days) for AI prioritization days and 15 days (IQR = 6–50 days) for those with no AI prioritization. The median time to initiation of cancer treatment was 76 days (IQR = 38–114 days) for AI prioritization days and 73 days (IQR = 43–121 days) for those with no AI prioritization (Table 2 ). Extended Data Table 3 presents the stage and incidence of lung cancer in each arm. There were no differences in secondary outcomes by AI group. Table 3 presents the absolute numbers of true-positive (TP), false-positive (FP), true-negative (TN) and false-negative (FN) CXRs and the number of subsequent cancers diagnosed within each group after discordance reviews. Discordance reviews were completed for 26,505 of 28,261 discordant CXR reports (30.3%) of all CXRs. In 1,756 cases, discordance review was not performed because the category was not considered potentially serious and the prespecified limit of 1,000 reviews per category had been reached. The sensitivity and specificity relate to the radiology concordance review of the CXR, which was regarded as the reference standard. Table 3 shows a total of 254,349 discordant findings reviews; 5.5% were TP, 11.6% FP, 2.3% FN and 81.2% TN. There were 387 subsequent cancers diagnosed. More than one abnormality was often present on the CXR and patients with lung cancer may have had more than one CXR, so each lung cancer may be associated with multiple abnormality classifications. AI was positive for 354 of 387 cancers in the opacity category and 215 of 387 in the nodule category. However, 107 of 11,284 opacity classifications and 150 of 5,783 nodule classifications were judged to be FPs. AI was negative for nodule and cavity category in 20 of 343 and 9 of 1,395 CXRs, respectively, that subsequently resulted in a diagnosis of lung cancer. The agreement between AI and the radiology report for each classification is presented in Extended Data Table 5 . Table 4 shows the recommendations made following expert discordance review with the number of lung cancers diagnosed in the period following the CXR. Note that this table shows recommendations after discordance for any findings category. Table 3 Review outcomes and cancers diagnosed by abnormality for CXRs with review (numbers in parentheses indicate the number of cancers diagnosed within each group) Abnormality category True positive False positive False negative True negative Sensitivity Specificity Nodule 1,306 (65) 5,783 (150) 343 (20) 20,829 (152) 0.79 0.78 Cavity 40 (6) 257 (14) 35 (5) 27,929 (362) 0.53 0.99 Mediastinal widening 73 (2) 667 (3) 182 (17) 27,339 (365) 0.29 0.98 Hilar enlargement 186 (13) 2,997 (24) 169 (33) 24,909 (317) 0.52 0.89 Blunted CP/pleural effusion 1,935 (65) 2,827 (52) 284 (4) 23,215 (266) 0.87 0.89 Opacity 7,931 (247) 11,284 (107) 1,395 (9) 7,651 (24) 0.85 0.40 Cardiomegaly 1,425 (5) 2,141 (18) 799 (4) 23,896 (360) 0.64 0.92 Pneumothorax 11 (0) 447 (8) 3 (0) 27,800 (379) 0.79 0.98 Other 1,068 (19) 1,730 (48) 2,567 (29) 22,896 (291) 0.29 0.93 Total 13,975 28,133 5,777 206,464 – – Table 4 Recommendation after expert discordance review Outcome recommendation n (%) Number of cancers diagnosed No actionable finding 21,511 (76.1%) 57 Suspected lung cancer—refer for CT and to lung cancer MDT 672 (2.4%) 232 Pulmonary nodule requiring CT 559 (2.0%) 52 Suspected other cancer—refer to other cancer MDT 52 (0.2%) 5 Incidental thoracic finding—requires respiratory review 488 (1.7%) 11 Incidental nonthoracic finding—requires other secondary care action 146 (0.5%) 0 Incidental findings for primary care action 3,077 (10.9%) 21 Number of CXRs reviewed n = 26,505. Note that no review was undertaken for 1,756 discordances. For common and less serious categories, a maximum of 1,000 reviews was required. MDT, multidisciplinary team. Post hoc analyses Patients with a confirmed diagnosis of cancer For patients diagnosed with cancer, the time to CT and time to cancer diagnosis were longest when both the radiologist and AI CXR reports were normal; median of 72 days to CT and median of 177 days to cancer diagnosis, compared to a median of 8 days to CT and 38 days to cancer diagnosis with both the radiologist and AI CXR reports were abnormal for any reason (Table 5 ). These differences were statistically significant, with the time to CT scan seven times higher ( P < 0.001) and the time to cancer diagnosis three times higher ( P < 0.001) for the radiologist and AI CXR reports both normal compared to both the radiologist and AI CXR reports abnormal (Table 5 ). Of the total of 558 cancers, 53 were in the radiology-normal/AI-abnormal group, with a median time to diagnosis of 106 days—potentially within a window when CT would have been able to diagnose early-stage cancer and 2.27 times longer geometric mean than when both radiology and AI were abnormal (median time to diagnosis of 38 days). In contrast, the ratio of geometric means was only 1.24:1 for the 22 cancers in the radiologist-abnormal/AI-normal group (median time to diagnosis of 50 days). Duration of patients with a lung cancer diagnosis for AI prioritization on and off is visualized in Extended Data Fig. 1 . Table 5 For those diagnosed with cancer, time to CT scan and time to cancer diagnosis by radiologist and AI CXR report normal/abnormal, median days and ratio of geometric means CXR report agreement n Median days (IQR) Ratio of geometric means 95% CI P value Time to CT Radiology and AI normal 30 72 (32–154) 7.35 (3.63–14.9) <0.001 Radiology normal AI abnormal 51 46 (9–121) 4.70 (2.70–8.17) <0.001 Radiology abnormal and AI normal 20 15 (7–22.5) 1.59 (0.68–3.73) 0.29 Radiology and AI abnormal 446 8 (4–21) 1.0 – – Time to cancer diagnosis Radiology and AI normal 33 177 (91–227) 3.43 (2.47–4.75) <0.001 Radiology normal AI abnormal 53 106 (52–187) 2.27 (1.75–2.96) <0.001 Radiology abnormal and AI normal 22 50 (30–75) 1.24 (0.83–1.84) 0.29 Radiology and AI abnormal 450 38 (22–72) 1 – – Statistical analyses were performed using linear regression on log-transformed outcomes and presented as the ratio of geometric means and 95% CIs. All hypothesis testing was conducted using two-sided tests. Timings relevant to the NOLCP A total of only 172 CT scans were completed on the same day as the CXR, with 88 in the prioritization ‘on’ and 84 in the prioritization ‘off’ arms. This is the recommended best practice in the NOLCP. There were 477 CT scans within the maximum time of 72 h from CXR as set out in the NOLCP, with 241 in the prioritization ‘on’ and 236 in the prioritization ‘off’ arms. Discussion This large, multisite randomized study has shown that AI prioritization of CXRs did not improve the speed of the lung cancer diagnostic pathway. This means that this element of AI functionality, which introduces additional complexity and cost to AI installation and may add time to clinical workflows, is not required to accelerate the lung cancer diagnostic pathway. A detailed health economics evaluation will be published separately, but costs are both considerable and avoidable, given the results reported in this study. Several other important findings from the secondary outcomes and the exploratory analyses are discussed below. CXR, often the first test for patients with lung cancer referred for diagnosis, is the most frequently performed imaging investigation, with more than 7 million performed annually in England, including ~2.2 million CXRs referred from primary care alone 20 . Following either an abnormal CXR or presentation in a high-risk patient, the NOLCP mandates rapid progression to CT. Current (April 2025) median reporting times for CXRs referred from primary care (2 days) and waiting time for CT chest (15 days) do not meet the NOLCP standards (72 h from abnormal CXR to CT) 20 . The lower number of CT scanners per capita 21 , as well as chronic shortage of consultant radiologists and radiographers and an ever-increasing workload, may explain the lack of compliance. One potential strategy to optimize progression from CXR to CT and bring forward the diagnosis is to use an AI tool to prioritize CXRs with suspicious findings for urgent reporting and improve the accuracy of interpretation, thereby selecting patients most likely to benefit from a same-day CT scan. However, this approach needs careful evaluation, as AI might have the opposite effect of not increasing accuracy or speed and creating extra workload, for example, through FP findings or altered clinician behavior. Vigilance fatigue is of particular concern, that is, when the review of all CXRs flagged as abnormal either desensitizes the reader to potential subsequent FN CXRs labeled as normal by the AI or causes the reader to lose trust in the AI’s accuracy (the ‘cry wolf’ phenomenon) 22 . This is one reason why NICE has not recommended any AI products for CXR interpretation in England 23 . The LungIMPACT study has examined both the above elements of CXR use through its primary outcomes and several other aspects with the secondary outcomes, including analysis of discordances and the likelihood of a cancer diagnosis with the addition of AI. It has been shown conclusively that AI prioritization had no impact on time to CT or time to lung cancer diagnosis. In the secondary outcome analysis, further steps of the pathway were examined; the study has quantified the number of FPs and FNs and shown the number of lung cancers expected to be diagnosed (Tables 3 and 4 ). Exploratory analyses also showed differences in time to CT and time to lung cancer diagnosis across the different combinations of discordance (Table 5 ). A significant reduction in the median time from CXR acquisition to report was observed, from 47 h to 34.1 h; however, no significant differences were detected in any of the timings measured as primary or secondary outcomes (Table 2 ). The lack of improvement in AI prioritization is likely because, although the time to CXR report was shortened, it was not sufficiently reduced in the prioritization arm to influence the clinical pathway in the same way as immediate radiographer triage did in our previous study 13 . This may, in turn, be explained by the limited capacity within Trusts to both report immediately and organize downstream tests and appointments, even though this is a recommendation in the NOLCP. However, this finding could not have been predicted with certainty because our previous study showed a marked effect on time to diagnosis under similar service constraints, albeit on a smaller scale. It could be argued that pathway change should have been mandated in the prioritization arm to ensure all prioritized CXRs were reviewed by a radiologist before the patient left the department. However, this represents a pathway change that may have an impact on its own and has little to do with AI-based prioritization (see review of other studies below). Furthermore, as shown in Extended Data Fig. 2 , immediate review of the CXR was permitted, according to local procedure, in both arms of the study. This ensured that AI prioritization was tested rather than the effect of primary pathway change. A large number of discordance reviews were completed (26,505), representing 28.4% of the randomized CXRs and 94% of all discordances. Although results are presented for eight classifications, the ‘other’ category includes an additional nine findings (none of which were relevant to lung cancer). Expert radiology review was considered the gold standard. Compared with radiologists, AI FP findings comprised 11.6% of the total across all categories. Some of these will be from the same CXR (proportion of FP plus FN of all reviewed CXRs was 33,910/28,261; ratio = 1.2:1). However, feedback from reporters indicated that most of the FPs are rapidly dismissed, so the potential for extra workload will be dependent on the user becoming familiar with the strengths and weaknesses of the AI. An important finding from our study relates to the number of cancers subsequently diagnosed in the cohort with a positive AI finding that was dismissed by the radiologist or reporting radiographer and the apparent association with time to diagnosis. We are aware of only a few cases in which there was a definite error on behalf of the reporter after review and this merits a detailed evaluation of the identified cases to establish whether they can be appropriately managed, perhaps using a combination of baseline risk 24 and CXR AI findings. It is also important to establish the reason for FNs. These were relatively few in number, with over half in the ‘other’ and cardiomegaly categories (3,366/5,777). However, for the nodule category, with only 343 AI FNs, 20 cancers were diagnosed, representing 5% of the cancers in that category (Table 3 ). A major strength of the trial is its randomized controlled design, which does not require obtaining individual consent from participants. This meant minimal influence on the normal clinical pathway and more equitable inclusion of groups underrepresented in research. Information provided in the radiology departments indicated that participants could opt out of the study; however, only one participant did so. In addition, the trial was multicentre, encompassing NHS hospitals of different sizes and catchment demographics. The automated data acquisition followed by manual verification is another strength of the study and the independent analysis of the results with no influence from either the AI developers or the investigators. Another major strength is the number of discordance reviews completed by expert radiologists and reporting radiographers. This has provided perhaps the largest reliable dataset on this aspect. Review/recall methodology is a form of radiology peer review, advocated as an effective method of quality assurance 25 – 29 , but it is seldom, if ever, done on this scale. A strength of the trial is also the use of two clinically meaningful, real-world, primary outcomes: time to lung cancer diagnosis and time to CT. Both are highly relevant to the goal of achieving earlier diagnosis in lung cancer, but the former is more dependent on the many steps involved in the pathway. In addition, secondary outcomes provided insight into other potential impacts, including the number of lung cancer diagnoses, the stage at diagnosis and the number of urgent referrals, which in our previous work showed trends towards being more favorable in the immediate reporting arm, but with too few cancers to allow a statistically significant result ( n = 49) 13 . A limitation of the study is that the primary outcomes evaluated only the impact of AI prioritization, not the presence or absence of AI in the clinical workflow. This was necessary because of the study’s nonconsenting nature. Consenting would have been impractical and could itself have influenced the primary outcome through greater-than-normal scrutiny of imaging within the context of the busy NHS. Furthermore, the use of AI, rather than no use at all, could improve triage quality and potentially the speed of diagnosis. However, it would likely have less influence on CT requests unless the AI were highly accurate. By choosing time to CT as an additional primary outcome, we can be more confident that a comparison of AI use versus no AI would not show a meaningful difference. A limitation is the use of a single AI product, which will inevitably have different performance characteristics from others. This issue is likely to persist as algorithms are updated. However, the fact that only prioritization was tested here means differences in AI accuracy had less effect on the coprimary outcomes than on the discordances. A further limitation of the study was the difficulty in identifying the best way to measure time to CT. Including all CTs was necessary to ensure that any potential adverse effect of AI prioritization, such as ‘reassurance’ from a normal AI finding, were captured. This was because we found that many patients having CXRs go on to have CTs, which may have no relationship to the CXR—for example, due to subsequent respiratory assessment or emergency department attendance, even many days after the first referral. While this finding is of interest in itself, it may have diluted this primary outcome by adding many CTs that were not relevant to the hypothesis. This explains why the time to CT, for all downstream scans, was so long (indeed longer than the time to lung cancer diagnosis). We therefore had to complete restrictive sensitivity analyses using the code, where used, signifying a CT was triggered and also by limiting the time after CXR to less than 14 days, having determined that these were likely related to the initial CXR. This was the case even when the CXR was normal, but the risk remained high enough to warrant referral for subsequent CT. Despite this limitation, the results clearly show no difference in time from CXR to CT according to prioritization day across all analyses, each of which included a more than adequate number of observations (Table 3 ). Another limitation relates to the discordance reviews, which have been reported according to differences between the AI flag classification and the final report. Because reporters had the AI available at the time, we cannot determine how many findings would be independently found by the AI or whether the human would have found something without AI assistance. However, the purpose of the discordance review was merely to confirm that the reports were indeed discordant, and this is what is presented. There were a small number of discordances in which the AI was correct and the reporter incorrect; these have not yet been fully quantified or analyzed. These cases, together with a detailed analysis of downstream outcomes, are the subject of a further project. In England, there has been a phased implementation of lung cancer screening for people aged 55 to 74 at high risk of developing lung cancer. From Table 1 and Extended Data Table 1, the cancer rate per patient can be calculated, demonstrating that University College London Hospital, which has an established screening program, had a rate of 0.38%, whereas University Hospitals of Birmingham, which commenced screening in 2022, had a rate of 0.53%. However, the rate for University Hospitals of Leicester (without a program) and Nottingham University Hospitals (program started in 2023) was 0.7%. Our study is unable to determine whether the screening program is reducing the early-stage detection by CXR. The health economics analysis is underway and considers the comparison made in the study, as well as focusing on the differences between using AI at all and not using AI, in a planned secondary analysis. Research on the impact of AI prioritization of the CXR is very limited. A prospective study is currently underway in Glasgow, Scotland, assessing the impact of AI prioritization on time to CT among patients with lung cancer 30 ; this is the only other randomized prospective study we are aware of. The research on the impact of CXR AI on clinical outcomes and accuracy is also limited. Some information comes from company product specifications, which show favorable area under the curve (AUC) values across a variety of classifications 31 , 32 . In a retrospective study of 6,006 CXRs from a respiratory department, AI assistance improved radiologists’ detection performance for nodules/masses, consolidation and pneumothorax, increasing the AUC from 0.861 to 0.886 ( P = 0.003; ref. 33 ). Another retrospective study using 2,568 CXRs acquired in a range of settings showed an AUC for radiologists of 0.71, which improved to 0.81 with AI assistance, but also noted a remarkable AUC of 0.96 with the AI model alone 34 . A further retrospective study of 563 emergency unit CXRs showed improvement in the accuracy of imaging diagnosis among nonradiology residents 35 . A small, single site ( n = 68 lung cancers), nonrandomized before-and-after 12-month service evaluation conducted in Scotland with another CXR AI algorithm could not attribute the small (mean 51 versus 58 days) reduction in time to treatment to the use of AI as multiple pathway changes were made (additional radiologist, administrative support, pathway navigator) and resulted in an increase in cost of £3.59 per CXR 36 . Another single-trust before-and-after comparison of AI found a shorter time from CXR to CT chest from 6 to 3.6 days 37 . However, as with the Scottish service evaluation, there was extensive pathway redesign within the ‘after’ pathway, which the authors recognized as a significant limitation in attributing improvements to AI. Storey and colleagues also used a weak reference standard, namely lung cancer suspected on CT chest, rather than the robust reference standard of confirmed lung cancer diagnosis in the current study. Although some of these studies show more favorable results for the classifiers shown in Table 3 , they may not be directly comparable as the CXRs were acquired in a variety of settings, retrospectively selected and in some, enriched to include specific findings 34 . In the present study, cases were unselected and, in the context of primary care, requested a CXR, and were much greater in number. In addition, we were able to show the number of downstream lung cancer diagnoses for each classification. The LungIMPACT trial has shown that immediate AI prioritization had no impact on important clinical outcomes for patients with suspected lung cancer in the English NHS and that this tool, as assessed, does not currently require potentially time-consuming and costly efforts to include in clinical services. This is likely to apply to the use of any CXR algorithm in this way, because the vast majority of cancers were detected where AI was abnormal and prioritization was therefore tested in a large number of cases. Even modest differences in accuracy are unlikely to change the primary outcome. Furthermore, this finding may extend to other settings where there is a relatively rapid review of images through the usual clinical pathway. Although the study applies directly to the English NHS, some findings may apply to other healthcare systems with established pathways. Because AI prioritization, although associated with a shortened time to CXR report, did not influence more important clinical outcomes, services would need to implement additional pathway changes to potentially replicate our previous findings. A recommendation could be that any AI-flagged abnormality prompts an immediate human review and, if confirmed, the implementation of a downstream bundle of investigations. The large-scale analysis of secondary outcomes shows the considerable potential burden of FP results, using radiologist and radiographer review as the gold standard, while also showing the considerable number of lung cancers diagnosed in this medium-to-high-risk group. This supports ongoing efforts to better manage these patients. Methods Study design A prospective, multicenter, randomized controlled trial was conducted between July 2023 and December 2024 across five NHS Trusts in England. Participants aged 18 years or older, attending for a primary care-requested CXR, were block-randomized by day and site to either immediate AI prioritization of reporting or no AI prioritization. Anteroposterior or posteroanterior views were analyzed and lateral (not routinely performed in the UK) views were excluded. Pre-allocation to the intervention or routine care was performed using random sampling via a 1:1 randomization method for Monday-to-Friday routine imaging whole-day sessions. Days were randomly allocated to minimize the potential impact of different levels of staffing and capacity. All imaging (CXR or CT) was performed as part of routine care, with the final decision made by the clinicians. Extended Data Fig. 2 illustrates the pathway and randomization. A single CXR AI (qXR, Qure.ai Technologies) was used in the study. The AI was applied to both arms at the time of image acquisition, so that the reporter had access to the AI-marked images at the time of reporting in both arms of the study. Extended Data Fig. 2 shows that radiographers could, at their discretion, flag abnormalities that may require further action, which may result in a CT scan, sometimes performed on the same day, which is a preferred option in the NOLCP. Radiographer flags could also be applied to any other (noncancer) findings where they consider that action is potentially needed. The usual clinical pathway was followed where any action was considered to be necessary or optimal. Where there was no flag, the CXR was later reported by a radiologist or reporting radiographer according to local usual clinical practice. AI algorithm qXR (v 4.0, Qure.ai Technologies, India) is a class IIb CE-certified deep learning algorithm already in routine clinical use in some NHS Hospitals. It functions as a clinical decision support tool to assist healthcare professionals in identifying abnormalities (29 classes/types) on CXRs. Images were pseudonymized, transferred to the cloud-based qXR for analysis and then returned to the local Picture Archiving and Communication System with image mark-ups and abnormality classification. On intervention days, an active notification was sent to the worklist for any ‘qXR-suspected-abnormal’ cases, allowing prioritization for immediate reporting by radiology. The software algorithm did not undergo updates or changes for the duration of the study to ensure consistency. Training on the use of the AI was provided to all reporters during the implementation of the AI at each hospital and before study commencement. Training covered intended use as a clinical decision support tool, the different AI findings, radiology worklist prioritization and the user interface/AI results. Sample size Based on data from previous work 13 , the median time to lung cancer diagnosis was 63 days in the standard reporting group and using a conservative reduction of 10 days, we calculated that 265 cases per group would be needed to detect a difference with 95% power. The expected prevalence of lung cancer in this cohort (primary care referrals for CXR) is 0.6% 13 and thus 100,000 CXRs were judged sufficient (600 cancer cases). For time to CT, the NHS Diagnostic Imaging Dataset records a median of 15 days between request and CT, but data are not segregated between urgent scans and those that are routine or for follow up 20 . The mean and standard deviation of the time between abnormal CXR and CT chest are unknown. Cohen’s d (effect size) was used to estimate sample size with power 0.1, 0.3 and 0.5 (considered small, moderate and large effect sizes) 38 . Assuming approximately equal distribution of the number of scans taken, 4,000 CT scans were found to provide adequate power to detect even the smallest effect size (Cohen’s d = 0.1). A clinically meaningful difference in time from abnormal CXR to CT chest of three days was defined, using the maximum time recommended in the NOLCP. Data Data were collected from existing routine clinical data sources using electronic methods and transferred to a single database. Data verification was performed by manually checking the dates of CT and lung cancer diagnosis by the investigators without knowledge of the study arm. Primary and secondary outcomes with invalid dates that generated negative time intervals were excluded during the data cleaning process. Safety and discordance All CXRs (in both arms) were evaluated for discordance between AI findings and the radiology report. This was done for each of the 29 AI class findings that the AI detected on the CXR, grouped into eight categories based on radiological–clinical relevance and are reported in the results with a ninth category of ‘other’. CXRs with a positive AI class finding not taken up for further testing (radiology report normal for that finding) were reviewed. All AI findings that were either potentially serious, such as pneumothorax, or potentially indicative of cancer, such as nodules and hilar enlargement, were also reviewed. CXRs where the only disagreement between AI positive and report negative were unlikely to be suspicious for cancer were capped for pragmatic reasons; cardiomegaly (1,000), blunted costophrenic angle (1,000), pleural effusion (20%) and fibrosis (20%). All reviews were undertaken by local experienced thoracic radiologists or reporting radiographers with the facility for a further opinion. The primary outcomes were as follows: (1) the difference in time (in days from CXR acquisition) to the diagnosis of patients with lung cancer who had CXRs with AI support at the time of CXR acquisition and prioritization for immediate review and those that had no immediate read but the AI read was available at the time of reporting and (2) the difference in time (in days from CXR acquisition) to CT for patients who had CXRs with AI support at the time of CXR acquisition and prioritization for immediate review and those that had no immediate read but the AI read was available at the time of reporting. The secondary outcomes were as follows: (1) days to urgent lung cancer referral as defined by the time between CXR acquisition and lung cancer referral (2WW); (2) days to treatment start for patients with lung cancer as defined by the time between CXR acquisition and cancer treatment start date; (3) agreement between qXR and human readers for present/absent findings on CXRs referred from primary care, by each classification; (4) number of urgent lung cancer (2WW) referrals; (5) incidence of lung cancer; (6) stage of lung cancer at diagnosis and (7) algorithm sensitivity, specificity and agreement between report and AI (Kappa). Data handling of multiple CXRs per patients Some patients had multiple CXRs over the study period. For the main analysis of the primary outcomes, we calculated time to CT and time to cancer diagnosis from the first CXR (to avoid duplication and overcounting outcomes). Sensitivity analysis (described below) was also conducted to assess the robustness of this assumption. Statistical analysis Statistical analysis was conducted by an independent statistician (LS) without input from the study investigators. Statistical analyses were performed using Stata/MP 19.5 (StataCorp). Baseline characteristics of the study population were tabulated using summary statistics overall and by intervention group (immediate AI prioritization or no AI prioritization; Table 1 ). No statistical testing was carried out on this data. Graphical representations of the time to lung cancer diagnosis and time to CT scan between the two groups were generated using histograms and boxplots. As the time distributions for both these outcomes exhibited a right-skewed distribution, a log transformation was used for formal statistical testing. A t test on the log transformation of these data was used to test the null hypothesis that mean values on the log-transformed scale are equal. This is equivalent to the null hypothesis that the geometric means on the untransformed scale are equal. The primary outcome, time to CT scan, was further examined by restricting the sample to (1) CT scans within 14 days of CXR or (2) CXRs that were flagged for CT referral only. This was because, in many cases, CTs were performed for other reasons much later, independently of the CXR request, and were unrelated to the hypothesis being tested. On manual review of data, when the time to CT exceeded 14 days, the CT was never triggered by the index CXR randomized for the trial. In addition, some CXRs were coded by radiology to indicate that a CT had been recommended. Other time-related endpoints (time to urgent referral and time to treatment starting) were also assessed using a t test on log-transformed time duration data. The number of urgent referrals, incidence of lung cancer and stage of lung cancer at diagnosis were compared between the two groups using chi-squared tests. The secondary endpoint of agreement between the reporter and the AI was summarized as overall agreement, measured using the kappa statistic. Subgroup analyses were performed by site and year quarter for both primary endpoints. Sensitivity analyses addressing the issue of multiple CXRs per patient were conducted by identifying the closest CXR before the CT scan as the index CXR or by excluding patients with multiple CXRs. Further exploratory analysis (not prespecified) of the combined arms was conducted for those diagnosed with cancer, examining only whether both the radiologist and AI CXR reports were normal or abnormal. CXR reports were flagged as normal if all features were recorded as absent; if any feature was recorded as present, the CXR report was classified as abnormal. Comparisons of time to CT scan and time to cancer diagnosis between radiologist and AI CXR reports as normal or abnormal were made using linear regression, with the outcome variables log-transformed as described above. Ethical and safety considerations Favorable ethical approval was obtained from the East of England—Cambridge East Research Ethics Committee (23/EE/0014, 21 February 2023). The study was undertaken with strict adherence to recommended CONSORT guidelines and Good Clinical Practice 39 . The data was held securely, and information governance rules were rigorously followed by all persons involved in the management of the trial protocol or data at the site level, as well as by the investigators. The study did not directly recruit patients; rather, it evaluated health service delivery. Individual patients were not randomized; rather, their CXRs were. Consent was not obtained from patients, but in each department, clear messaging indicated that AI was being used as part of a research study, and details on how to opt out of the study were provided, as agreed by the Ethics Committee. No patient-identifiable data were available outside the direct clinical care team and data were pseudonymized before analysis. The health economic analysis will be reported separately. The study protocol is available in the Supplementary Information . Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information, details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-026-04253-5. Supplementary information Reporting Summary Peer Review File Extended data Extended Data Fig. 1 Duration on pathway by patient. This data visualization is limited to those diagnosed with cancer and valid dates for all events presented in the pathway, n = 382, 192 with AI prioritization and 190 with no AI prioritization. Extended Data Fig. 2 Pathway and randomization. The radiographer may, at their discretion, flag abnormalities that may require further action and this may result in a CT scan being done, sometimes on the same day, which is a preferred option in the NOLCP. The radiographer flag can also happen for any other (noncancer) findings where they consider that action is potentially needed. The usual clinical pathway was followed, where action was confirmed to be necessary or optimal. Where there was no flag, the CXR was later reported by a radiologist or reporting radiographer according to local usual clinical practice. Extended Data Table 1 Primary outcomes stratified by site, ratio of geometric means comparing AI prioritization groups Extended Data Table 2 Primary outcomes stratified by quarter, ratio of geometric means comparing AI prioritization groups Extended Data Table 3 Secondary outcomes by AI prioritization Extended Data Table 4 Primary outcomes stratified by sex, median and interquartile range and ratio of geometric means comparing AI prioritization groups Extended Data Table 5 Agreement between radiologist reports and Qure AI report by abnormality Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Extended data is available for this paper at 10.1038/s41591-026-04253-5. Supplementary information The online version contains supplementary material available at 10.1038/s41591-026-04253-5. Acknowledgements This work was supported by the Small Business Research Initiative (SBRI) Healthcare (grant SBRIC01P3039 to N.W. and D.R.B.). The funder undertook regular reviews of project progress and funding allocation. We are indebted to the Qure.ai Technologies team for their work on the implementation of the AI and for ensuring all interactions with the NHS ran smoothly and that data were collected in full compliance with data protection requirements. We thank the NHS information technology, research and innovation teams for their help in navigating this relatively new area of research. Qure.ai Technologies had no influence on the study design or analysis. The latter was conducted by a statistician independently of the trial investigators. Qure.ai Technologies assisted in the deployment of qXR, working with the NHS to integrate with local IT systems, ensure pseudonymization and the correct prioritization method, which could be switched on or off according to the randomization. Author contributions N.W. and D.R.B. conceived the study design and wrote the study protocol, obtained ethics approval, maintained oversight of the study and drafted the first version of the manuscript. L.S. designed the statistical analysis plan and conducted all statistical analyses. S.N. and C.S.C. assisted with data cleaning and took part in analytical discussions. J.R. contributed/coordinated PPI work for this study, including the patient panel and part of the study team, including governance. A.N., R.W.L. and N.N. contributed to the study protocol and provided oversight of the study. I.A.Y., B.G., M.G.D., I.D. and J.H. contributed to the study protocol and coordinated data collection at sites. A.C., J.N., S.A. and R.T. performed data collection. N.W., D.R.B., L.S., S.N., C.S.C., A.N., R.W.L. and N.N. edited the final version of the manuscript. I.A.Y., B.G., M.G.D., I.D., J.H., A.C., J.N., S.A. and R.T. reviewed and approved the final version of the manuscript. Peer review Peer review information Nature Medicine thanks Stephen Lam and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Mattia Andreoletti, in collaboration with the Nature Medicine team. Data availability All data for the study is stored by the study sponsor. Access to fully anonymised data can be requested from the sponsor at: Research & Innovation, Nottingham University Hospitals NHS Trust, Queen’s Medical Centre, B Floor, Medical School, Derby Road, Nottingham NG7 2UH, UK. Email: nuhnt.researchsponsor@nhs.net. Data would normally be available within two months of the request. Code availability The code for the AI algorithm is not available, as this, like many others in this field, is a commercial product. The code for the statistical analysis is available from L.S. and has been uploaded to GitHub; see https://github.com/LFairleySmith/LungIMPACT . Competing interests D.R.B. declares honoraria for speaking and education from AstraZeneca and Boehringer Ingelheim, unrelated to the current study. He has received research grants from the National Institute of Health Research, Cancer Research UK, Horizon Europe, Ruth Strauss Foundation, SBRI, Innovate UK, Yorkshire Cancer Research, the Royal Castle Foundation, and UK Research and Innovation. N.W. declares consultancy fees from InHealth, Apollo Radiology (UK) and SMR Health & Tech unrelated to the current study and a travel grant from Qure.ai Technologies to attend the European Congress of Radiology 2023. R.W.L. is funded by the Royal Marsden National Institute for Health and Care Research (NIHR) Biomedical Research Centre and The Royal Marsden Cancer Charity; his institution receives compensation for time spent in a secondment role with NHS England for the Lung Health Check Programme and as National Specialty Lead for the National Institute for Health and Care Research; he has received research funding from Cancer Research UK, Innovate UK (cofunded by GE Healthcare, Roche Diagnostics, Optellum, Elliptica and RNA Guardian), SBRI (including as a co-applicant with Qure.ai), RM Partners Cancer Alliance and NIHR (including co-applicant on grants with Optellum); he has received honoraria, speaker and advisory fees, and/or hospitality and travel expenses from Cancer Research UK, Roche Diagnostics, Johnson & Johnson, Guardant Health, AstraZeneca and King Faisal Specialist Hospital and Research Centre, Saudi Arabia; and he also undertakes private medical practice. These activities are unrelated to the present study, aside from the SBRI funding received in collaboration with Qure.ai. N.N. is supported by a Medical Research Council Clinical Academic Research Partnership (MR/T02481X/1). N.N. has also received grant funding from the National Institute of Health Research, Cancer Research UK, Horizon Europe, Ruth Strauss Foundation, and the Engineering and Physical Sciences Research Council. This work was partly undertaken at University College London Hospitals NHS Foundation Trust/University College London, which received a proportion of funding from the UK Department of Health’s NIHR Biomedical Research Centre’s funding scheme. N.N. reports honoraria for nonpromotional educational talks, conference attendance or advisory board participation from Amgen, AstraZeneca, AXANA, BeiGene, Boehringer Ingelheim, Bristol Myers Squibb, EQRx, Fujifilm, Guardant Health, Intuitive, Janssen, Eli Lilly and Company, Merck Sharp & Dohme, Olympus Corporation, Roche and Sanofi. The remaining authors declare no competing interests. References 1. Araghi M International differences in lung cancer survival by sex, histological type and stage at diagnosis: an ICBP SURVMARK-2 Study Thorax 2022 77 378 390 10.1136/thoraxjnl-2020-216555 34282033 Araghi, M. et al. International differences in lung cancer survival by sex, histological type and stage at diagnosis: an ICBP SURVMARK-2 Study. Thorax 77 , 378–390 (2022). 34282033 10.1136/thoraxjnl-2020-216555 2. Arnold M Progress in cancer survival, mortality, and incidence in seven high-income countries 1995-2014 (ICBP SURVMARK-2): a population-based study Lancet Oncol. 2019 20 1493 1505 10.1016/S1470-2045(19)30456-5 31521509 PMC6838671 Arnold, M. et al. Progress in cancer survival, mortality, and incidence in seven high-income countries 1995-2014 (ICBP SURVMARK-2): a population-based study. Lancet Oncol. 20 , 1493–1505 (2019). 31521509 10.1016/S1470-2045(19)30456-5 PMC6838671 3. Walters S Lung cancer survival and stage at diagnosis in Australia, Canada, Denmark, Norway, Sweden and the UK: a population-based study, 2004–2007 Thorax 2013 68 551 564 10.1136/thoraxjnl-2012-202297 23399908 Walters, S. et al. Lung cancer survival and stage at diagnosis in Australia, Canada, Denmark, Norway, Sweden and the UK: a population-based study, 2004–2007. Thorax 68 , 551–564 (2013). 23399908 10.1136/thoraxjnl-2012-202297 4. Navani N Lung cancer diagnosis and staging with endobronchial ultrasound-guided transbronchial needle aspiration compared with conventional approaches: an open-label, pragmatic, randomised controlled trial Lancet Respir. Med. 2015 3 282 289 10.1016/S2213-2600(15)00029-6 25660225 PMC4648022 Navani, N. et al. Lung cancer diagnosis and staging with endobronchial ultrasound-guided transbronchial needle aspiration compared with conventional approaches: an open-label, pragmatic, randomised controlled trial. Lancet Respir. Med. 3 , 282–289 (2015). 25660225 10.1016/S2213-2600(15)00029-6 PMC4648022 5. Yang CJ Impact of timing of lobectomy on survival for clinical stage IA lung squamous cell carcinoma Chest 2017 152 1239 1250 10.1016/j.chest.2017.07.032 28800867 Yang, C. J. et al. Impact of timing of lobectomy on survival for clinical stage IA lung squamous cell carcinoma. Chest 152 , 1239–1250 (2017). 28800867 10.1016/j.chest.2017.07.032 6. Khorana AA Time to initial cancer treatment in the United States and association with survival over time: an observational study PLoS ONE 2019 14 e0213209 10.1371/journal.pone.0213209 30822350 PMC6396925 Khorana, A. A. et al. Time to initial cancer treatment in the United States and association with survival over time: an observational study. PLoS ONE 14 , e0213209 (2019). 30822350 10.1371/journal.pone.0213209 PMC6396925 7. Paul C Cancer patients’ concerns regarding access to cancer care: perceived impact of waiting times along the diagnosis and treatment journey Eur. J. Cancer Care (Engl.) 2012 21 321 329 22111696 10.1111/j.1365-2354.2011.01311.x PMC3410528 Paul, C. et al. Cancer patients’ concerns regarding access to cancer care: perceived impact of waiting times along the diagnosis and treatment journey. Eur. J. Cancer Care (Engl.) 21 , 321–329 (2012). 22111696 10.1111/j.1365-2354.2011.01311.x PMC3410528 8. Capaccio E Podestà A Morcaldi D Sormani MP Derchi LE How often do patients ask for the results of their radiological studies? Insights Imaging 2010 1 83 85 10.1007/s13244-009-0003-1 22347908 PMC3259305 Capaccio, E., Podestà, A., Morcaldi, D., Sormani, M. P. & Derchi, L. E. How often do patients ask for the results of their radiological studies? Insights Imaging 1 , 83–85 (2010). 22347908 10.1007/s13244-009-0003-1 PMC3259305 9. Pahade J Reviewing imaging examination results with a radiologist immediately after study completion: patient preferences and assessment of feasibility in an academic department AJR Am. J. Roentgenol. 2012 199 844 851 10.2214/AJR.11.8064 22997377 PMC3786865 Pahade, J. et al. Reviewing imaging examination results with a radiologist immediately after study completion: patient preferences and assessment of feasibility in an academic department. AJR Am. J. Roentgenol. 199 , 844–851 (2012). 22997377 10.2214/AJR.11.8064 PMC3786865 10. Cancer Patient Experience Survey 2011/12—National Report (Department of Health, 2012); https://assets.publishing.service.gov.uk/media/5a7c1cc1e5274a25a9140877/Cancer-Patient-Experience-Survey-National-Report-2011-12.pdf 11. Optimal Care Pathway for People with Lung Cancer (Cancer Council Australia, 2021); https://www.cancer.org.au/assets/pdf/lung-cancer-optimal-cancer-care-pathway 12. National Optimal Lung Cancer Pathway (Version 4.0) (National Health Service England, 2024); https://roycastle.org/app/uploads/2024/03/National-Optimal-Lung-Cancer-Pathway-V4-01Jan2024.pptx 13. Woznitza N Impact of radiographer immediate reporting of X-rays of the chest from general practice on the lung cancer pathway (radioX): a randomised controlled trial Thorax 2023 78 890 894 10.1136/thorax-2022-219210 36351688 PMC10447363 Woznitza, N. et al. Impact of radiographer immediate reporting of X-rays of the chest from general practice on the lung cancer pathway (radioX): a randomised controlled trial. Thorax 78 , 890–894 (2023). 36351688 10.1136/thorax-2022-219210 PMC10447363 14. Suspected Cancer: Recognition and Referral (National Institute for Health and Care Excellence, 2015); https://www.nice.org.uk/guidance/ng12 26180880 15. Bradley SH Achieving earlier diagnosis of symptomatic lung cancer Br. J. Gen. Pract. 2024 75 40 43 39725539 10.3399/bjgp25X740493 PMC11684445 Bradley, S. H. et al. Achieving earlier diagnosis of symptomatic lung cancer. Br. J. Gen. Pract. 75 , 40–43 (2024). 39725539 10.3399/bjgp25X740493 PMC11684445 16. Kennedy MPT Lung cancer stage-shift following a symptom awareness campaign Thorax 2018 73 1128 1136 10.1136/thoraxjnl-2018-211842 29950525 Kennedy, M. P. T. et al. Lung cancer stage-shift following a symptom awareness campaign. Thorax 73 , 1128–1136 (2018). 29950525 10.1136/thoraxjnl-2018-211842 17. Bradley SH General practice chest X-ray rate is associated with earlier lung cancer diagnosis and reduced all-cause mortality: a retrospective observational study Br. J. Gen. Pract. 2025 75 e323 e332 10.3399/BJGP.2024.0466 39740925 PMC11966530 Bradley, S. H. et al. General practice chest X-ray rate is associated with earlier lung cancer diagnosis and reduced all-cause mortality: a retrospective observational study. Br. J. Gen. Pract. 75 , e323–e332 (2025). 39740925 10.3399/BJGP.2024.0466 PMC11966530 18. Bradley SH Kennedy MPT Callister MEJ Early-stage lung cancer associated with higher frequency of chest x-ray up to three years prior to diagnosis Prim. Health Care Res. Dev. 2022 23 e66 10.1017/S1463423622000573 36321523 PMC9641667 Bradley, S. H., Kennedy, M. P. T. & Callister, M. E. J. Early-stage lung cancer associated with higher frequency of chest x-ray up to three years prior to diagnosis. Prim. Health Care Res. Dev. 23 , e66 (2022). 36321523 10.1017/S1463423622000573 PMC9641667 19. Artificial Intelligence for Analysing Chest X-ray Images (National Institute for Health and Care Excellence, 2022); https://www.nice.org.uk/advice/mib292 20. Diagnostic Imaging Dataset Annual Statistical Release 2023/24 (NHS England, 2024); https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/2024/11/DID-Annual-Statistical-Release-2023-24.pdf 21. Life Sciences Competitiveness Indicators 2024: Summary (Department of Science, Innovation and Technology & Department of Health and Social Care, 2024); https://www.gov.uk/government/publications/life-sciences-sector-data-2024/life-sciences-competitiveness-indicators-2024-summary 22. Bernstein MH Can incorrect artificial intelligence (AI) results impact radiologists, and if so, what can we do about it? A multi-reader pilot study of lung cancer detection with chest radiography Eur. Radiol. 2023 33 8263 8269 10.1007/s00330-023-09747-1 37266657 PMC10235827 Bernstein, M. H. et al. Can incorrect artificial intelligence (AI) results impact radiologists, and if so, what can we do about it? A multi-reader pilot study of lung cancer detection with chest radiography. Eur. Radiol. 33 , 8263–8269 (2023). 37266657 10.1007/s00330-023-09747-1 PMC10235827 23. Artificial Intelligence (AI)-derived Software to Analyse Chest X-rays for Suspected Lung Cancer in Primary Care Referrals: Early Value Assessment (National Institute for Health and Care Excellence, 2023); https://www.nice.org.uk/guidance/hte12 24. Bradley SH Estimating lung cancer risk from chest X-ray and symptoms: a prospective cohort study Br. J. Gen. Pract. 2021 71 e280 e286 10.3399/bjgp20X713993 33318087 PMC7744041 Bradley, S. H. et al. Estimating lung cancer risk from chest X-ray and symptoms: a prospective cohort study. Br. J. Gen. Pract. 71 , e280–e286 (2021). 33318087 10.3399/bjgp20X713993 PMC7744041 25. Strickland NH Quality assurance in radiology: peer review and peer feedback Clin. Radiol. 2015 70 1158 1164 10.1016/j.crad.2015.06.091 26223739 Strickland, N. H. Quality assurance in radiology: peer review and peer feedback. Clin. Radiol. 70 , 1158–1164 (2015). 26223739 10.1016/j.crad.2015.06.091 26. Chetlen AL Collaborative learning in radiology: from peer review to peer learning and peer coaching Acad. Radiol. 2020 27 1261 1267 10.1016/j.acra.2019.09.021 31636005 Chetlen, A. L. et al. Collaborative learning in radiology: from peer review to peer learning and peer coaching. Acad. Radiol. 27 , 1261–1267 (2020). 31636005 10.1016/j.acra.2019.09.021 27. Capizzano AA Artificial intelligence in neuroradiology: a smart prospective peer reviewer Acad. Radiol. 2021 28 94 95 10.1016/j.acra.2020.06.029 32763061 Capizzano, A. A. Artificial intelligence in neuroradiology: a smart prospective peer reviewer. Acad. Radiol. 28 , 94–95 (2021). 32763061 10.1016/j.acra.2020.06.029 28. Rao B Utility of artificial intelligence tool as a prospective radiology peer reviewer—detection of unreported intracranial hemorrhage Acad. Radiol. 2021 28 85 93 10.1016/j.acra.2020.01.035 32102747 Rao, B. et al. Utility of artificial intelligence tool as a prospective radiology peer reviewer—detection of unreported intracranial hemorrhage. Acad. Radiol. 28 , 85–93 (2021). 32102747 10.1016/j.acra.2020.01.035 29. Woznitza N Reporting radiographer peer review systems: a cross-sectional survey of London NHS trusts Radiography (Lond.) 2021 27 173 177 10.1016/j.radi.2020.07.014 32771302 Woznitza, N. et al. Reporting radiographer peer review systems: a cross-sectional survey of London NHS trusts. Radiography (Lond.) 27 , 173–177 (2021). 32771302 10.1016/j.radi.2020.07.014 30. Duncan SF Radiograph accelerated detection and identification of cancer in the lung (RADICAL): a mixed methods study to assess the clinical effectiveness and acceptability of Qure.ai artificial intelligence software to prioritise chest X-ray (CXR) interpretation BMJ Open 2024 14 e081062 10.1136/bmjopen-2023-081062 39306349 PMC11418533 Duncan, S. F. et al. Radiograph accelerated detection and identification of cancer in the lung (RADICAL): a mixed methods study to assess the clinical effectiveness and acceptability of Qure.ai artificial intelligence software to prioritise chest X-ray (CXR) interpretation. BMJ Open 14 , e081062 (2024). 39306349 10.1136/bmjopen-2023-081062 PMC11418533 31. CXR Edge Performance Guide (Annalise, 2025); https://www.tga.gov.au/sites/default/files/2025-06/FOI%2025-0127%20Documents.pdf 32. RAIQC supports evaluation of Lunit INSIGHT CXR AI tool for enhancing clinician chest X-ray interpretation. RAIQC https://www.raiqc.com/news/raiqc-supports-evaluation-of-lunit-insight-cxr-tool/ (2024). 33. Jin KN Diagnostic effect of artificial intelligence solution for referable thoracic abnormalities on chest radiography: a multicenter respiratory outpatient diagnostic cohort study Eur. Radiol. 2022 32 3469 3479 10.1007/s00330-021-08397-5 34973101 PMC9038825 Jin, K. N. et al. Diagnostic effect of artificial intelligence solution for referable thoracic abnormalities on chest radiography: a multicenter respiratory outpatient diagnostic cohort study. Eur. Radiol. 32 , 3469–3479 (2022). 34973101 10.1007/s00330-021-08397-5 PMC9038825 34. Seah JCY Effect of a comprehensive deep-learning model on the accuracy of chest x-ray interpretation by radiologists: a retrospective, multireader multicase study Lancet Digit. Health 2021 3 e496 e506 10.1016/S2589-7500(21)00106-0 34219054 Seah, J. C. Y. et al. Effect of a comprehensive deep-learning model on the accuracy of chest x-ray interpretation by radiologists: a retrospective, multireader multicase study. Lancet Digit. Health 3 , e496–e506 (2021). 34219054 10.1016/S2589-7500(21)00106-0 35. Rudolph J Artificial intelligence in chest radiography reporting accuracy: added clinical value in the emergency unit setting without 24/7 radiology coverage Invest. Radiol. 2022 57 90 98 10.1097/RLI.0000000000000813 34352804 Rudolph, J. et al. Artificial intelligence in chest radiography reporting accuracy: added clinical value in the emergency unit setting without 24/7 radiology coverage. Invest. Radiol. 57 , 90–98 (2022). 34352804 10.1097/RLI.0000000000000813 36. Artificial Intelligence-assisted Clinician Review of Chest X-rays for Suspected Lung Cancer (Healthcare Improvement Scotland, 2024); https://shtg.scot/media/2526/2024-07-18-chest-x-ray-assessment-v14.pdf 37. Storey M Early clinical evaluation of AI triage of chest radiographs: time to diagnosis for suspected cancer and number of urgent CT referrals NEJM AI 2025 3 AIcs2500539 10.1056/AIcs2500539 Storey, M. et al. Early clinical evaluation of AI triage of chest radiographs: time to diagnosis for suspected cancer and number of urgent CT referrals. NEJM AI 3 , AIcs2500539 (2025). 38. Cohen, J. Statistical Power Analysis for the Behavioural Science, Second Edition (Laurence Erlbaum Associates, 1988). 39. Liu X Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension Lancet Digit. Health 2020 2 e537 e548 10.1016/S2589-7500(20)30218-1 33328048 PMC8183333 Liu, X. et al. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Lancet Digit. Health 2 , e537–e548 (2020). 33328048 10.1016/S2589-7500(20)30218-1 PMC8183333