External validation of a convolutional neural network artificial intelligence tool to predict malignancy in pulmonary nodules.

BACKGROUND: Estimation of the risk of malignancy in pulmonary nodules detected by CT is central in clinical management. The use of artificial intelligence (AI) offers an opportunity to improve risk prediction. Here we compare the performance of an AI algorithm, the lung cancer prediction convolutional neural network (LCP-CNN), with that of the Brock University model, recommended in UK guidelines. METHODS: A dataset of incidentally detected pulmonary nodules measuring 5-15 mm was collected retrospectively from three UK hospitals for use in a validation study. Ground truth diagnosis for each nodule was based on histology (required for any cancer), resolution, stability or (for pulmonary lymph nodes only) expert opinion. There were 1397 nodules in 1187 patients, of which 234 nodules in 229 (19.3%) patients were cancer. Model discrimination and performance statistics at predefined score thresholds were compared between the Brock model and the LCP-CNN. RESULTS: The area under the curve for LCP-CNN was 89.6% (95% CI 87.6 to 91.5), compared with 86.8% (95% CI 84.3 to 89.1) for the Brock model (p≤0.005). Using the LCP-CNN, we found that 24.5% of nodules scored below the lowest cancer nodule score, compared with 10.9% using the Brock score. Using the predefined thresholds, we found that the LCP-CNN gave one false negative (0.4% of cancers), whereas the Brock model gave six (2.5%), while specificity statistics were similar between the two models. CONCLUSION: The LCP-CNN score has better discrimination and allows a larger proportion of benign nodules to be identified without missing cancers than the Brock model. This has the potential to substantially reduce the proportion of surveillance CT scans required and thus save significant resources.

Sequential screening for lung cancer in a high-risk group: randomised controlled trial: LungSEARCH: a randomised controlled trial of Surveillance using sputum and imaging for the EARly detection of lung Cancer in a High-risk group.

BACKGROUND: Low-dose computed tomography (LDCT) screening detects early-stage lung cancer and reduces mortality. We proposed a sequential approach targeted to a high-risk group as a potentially efficient screening strategy. METHODS: LungSEARCH was a national multicentre randomised trial. Current/ex-smokers with mild/moderate chronic obstructive pulmonary disease (COPD) were allocated (1:1) to have 5 years surveillance or not. Screened participants provided annual sputum samples for cytology and cytometry, and if abnormal were offered annual LDCT and autofluorescence bronchoscopy (AFB). Those with normal sputum provided annual samples. The primary end-point was the percentage of lung cancers diagnosed at stage I/II (nonsmall cell) or limited disease (small cell). RESULTS: 1568 participants were randomised during 2007-2011 from 10 UK centres. 85.2% of those screened provided an adequate baseline sputum sample. There were 42 lung cancers among 785 screened individuals and 36 lung cancers among 783 controls. 54.8% (23 out of 42) of screened individuals versus 45.2% (14 out of 31) of controls with known staging were diagnosed with early-stage disease (one-sided p=0.24). Relative risk was 1.21 (95% CI 0.75-1.95) or 0.82 (95% CI 0.52-1.31) for early-stage or advanced cancers, respectively. Overall sensitivity for sputum (in those randomised to surveillance) was low (40.5%) with a cumulative false-positive rate (FPR) of 32.8%. 55% of cancers had normal sputum results throughout. Among sputum-positive individuals who had AFB, sensitivity was 45.5% and cumulative FPR was 39.5%; the corresponding measures for those who had LDCT were 100% and 16.1%, respectively. CONCLUSIONS: Our sequential strategy, using sputum cytology/cytometry to select high-risk individuals for AFB and LDCT, did not lead to a clear stage shift and did not improve the efficiency of lung cancer screening.

Lung cancer stage-shift following a symptom awareness campaign.

BACKGROUND: Lung cancer outcomes in the UK are worse than in many other developed nations. Symptom awareness campaigns aim to diagnose patients at an earlier stage to improve cancer outcomes. METHODS: An early diagnosis campaign for lung cancer commenced in Leeds, UK in 2011 comprising public and primary-care facing components. Rates of community referral for chest X-ray and lung cancer stage (TNM seventh edition) at presentation were collected from 2008 to 2015. Linear trends were assessed by χ2 test for trend in proportions. Headline figures are presented for the 3 years pre-campaign (2008-2010) and the three most recent years for which data are available during the campaign (2013-2015). FINDINGS: Community-ordered chest X-ray rates per year increased from 18 909 in 2008-2010 to 34 194 in 2013-2015 (80.8% increase). A significant stage shift towards earlier stage lung cancer was seen (χ2(1)=32.2, p<0.0001). There was an 8.8 percentage point increase in the proportion of patients diagnosed with stage I/II lung cancer (26.5% pre-campaign vs 35.3% during campaign) and a 9.3% reduction in the absolute number of patients diagnosed with stage III/IV disease (1254 pre-campaign vs 1137 during campaign). INTERPRETATION: This is the largest described lung cancer stage-shift in association with a symptom awareness campaign. A causal link between the campaign and stage-shift cannot be proven but appears plausible. Limitations of the analysis include a lack of contemporary control population.