Virtual Library

Start Your Search

Emma O'Dowd



Author of

  • +

    MA05 - Lung Cancer Screening (ID 174)

    • Event: WCLC 2020
    • Type: Mini Oral
    • Track: Screening and Early Detection
    • Presentations: 1
    • +

      MA05.06 - Lung Cancer Screening – Cumulative Results from Five UK-Based Programmes (ID 3527)

      11:45 - 12:45  |  Author(s): Emma O'Dowd

      • Abstract
      • Presentation
      • Slides

      Introduction

      Lung cancer remains the leading cause of cancer related death globally. Low-dose CT (LDCT) screening of high-risk individuals reduces lung cancer specific mortality. An important requirement for any screening programme is to minimise harms, especially in those who do not have cancer. Data from randomised controlled trials (RCT) is often used as the primary source from which to extrapolate risks of harm but they do not reflect modern, real-world practice. In this paper we present cumulative data on screening harms from five UK-based lung cancer screening programmes.

      Methods

      In the United Kingdom (UK), several implementation pilots and research studies have demonstrated that screening can be successfully delivered within or aligned to the NHS. These include: UK Lung Cancer Screening Trial (UKLS), Lung Screen Uptake Trial, Manchester Lung Health Checks, Liverpool Healthy Lung Project and Nottingham Lung Health MOT. Most sites, other than UKLS, used the British Thoracic Society (BTS) nodule management guidelines. Positive screening results were defined as those referred for more than a repeat LDCT. False positives were those positive screens without an eventual diagnosis of lung cancer. Harms were categorised according to the need for further imaging, invasive investigations and/or surgery. Complications were categorised as per the National Lung Screening Trial (NLST).

      Results

      A total of 11,815 screening LDCTs were performed across the five projects between 2016 and 2020. Overall, 85.5% of screening scans were categorised as negative, 10.5% as indeterminate and 4% as positive. Lung cancer detection was 2.1%, ranging from 1.7% to 4.4% across sites. The surgical resection rate was 66.0%. Details of the cumulative reported harms are summarised in Table 1.

      Table 1. Details of cumulative reported harms

      Reported screening related harm

      Total % (n)

      Per 1000 screening scans

      False positive rate

      As a proportion of all LDCT scans

      1.9% (219)

      17

      As a proportion of all positive scans

      (i.e. false discovery rate)

      46.7% (219)

      -

      Invasive investigation* for benign disease (excluding surgery)

      0.5% (61)

      5

      Surgical resection for benign disease

      As a proportion of all surgeries

      4.6% (8)

      1

      As a proportion of all LDCT scans

      0.07% (8)

      -

      Major complication+ from invasive

      investigation/treatment for benign disease

      0% (0)

      0

      Deaths from invasive investigation/treatment for benign disease

      0% (0)

      0

      *image guide biopsies or bronchoscopic procedures; +as defined by NLST

      Conclusion

      Discussion: This collaborative work provides up-to-date data on lung cancer screening performance and harms. The rate of positive (4%) and false positive (1.9%) screening results were significantly lower than NLST and the majority of European screening trials other than NELSON. Harms from investigation and treatment of non-malignant disease was minimised with no reported major complications or deaths. This provides reassurance that with the use of evidence-based practice and experienced lung MDTs, harms from false positive results can be minimised within screening. This information is important in the planning of larger scale implementation of lung cancer screening within the UK and beyond.

      Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.

  • +

    P42 - Screening and Early Detection - Risk Modelling and Artificial Intelligence (ID 177)

    • Event: WCLC 2020
    • Type: Posters
    • Track: Screening and Early Detection
    • Presentations: 1
    • Moderators:
    • Coordinates: 1/28/2021, 00:00 - 00:00, ePoster Hall
    • +

      P42.01 - AI Assistance for Pulmonary Nodule Stratification: An Multiple-Reader Multiple-Case Study (ID 3481)

      00:00 - 00:00  |  Presenting Author(s): Emma O'Dowd

      • Abstract
      • Slides

      Introduction

      Improving stratification of patients with indeterminate pulmonary nodules (IPNs) can lead both to earlier diagnosis of lung cancer and to reduced scanning and reduced intervention in cases of benign disease. AI-based decision support software has been shown to outperform conventional risk models at classifying IPNs as low or high risk, but its performance in addition to clinician assessment has yet to be investigated. We report the results of a Multiple-Reader Multiple-Case reader evaluation comparing reader performance for both radiologists and pulmonologists on an IPN risk stratification task with and without AI assistance from the previously-published Lung Cancer Prediction Convolutional Neural Network (LCP-CNN).

      Methods

      A pool of 12 readers interpreted 100 non-contrast chest CTs each with at least one IPN. The reader breakdown was 7 radiologists, 5 pulmonologists. 7 were UK and 5 US, with representation from both specialties in both geographic zones. They ranged in experience from registrar (resident) to consultant (attending). Readers interacted with viewing software in which they could scroll through axial slices, and adjust window/level settings, but were blinded to patient clinical information. Readers first estimated the likelihood of malignancy for each nodule independently (solo-LoM) as a percentage. The reader was then provided with the AI score as a number from 1-10 and allowed to update assessment of LoM (assisted-LoM). The dataset comprised 50 histologically-diagnosed primary lung cancers (median 10.1mm, IQR 8-13mm) and 50 benign nodules (median 8.8mm, IQR 7-11mm). The nodules were detected incidentally at six EU centres, and were all 5-15mm in size at detection, in patients of 18+ years without a history of malignancy in the past 5 years. Performance was analysed comparing the Area Under the ROC curve (AUC) for the solo-LoM and assisted-LoM. CIs and P-values were calculated using bootstrapping.

      Results

      The average pre-AI AUC over all readers was 76.6 (95%CI 68.7-83.7), and 84.3 (77.2-90.6) when assisted by AI (P<.0001). The mean improvement over the set of readers is 7.7 points of AUC (95%CI 4.6-11.0). Table 1 shows the performance on a per-reader basis. All readers improved when assisted by the AI, and the improvement was significant in 10/12 readers at the 0.05 level.

      Table 1: Performance on per-reader basis

      Reader #

      AUC pre-AI

      AUC post-AI

      AUC improvement

      P value

      1

      81.1 (72.1-89.1)

      88.0 (81.0-93.8)

      6.9 (1.8-12.4)

      .004

      2

      74.5 (64.5-83.5)

      81.9 (73.4-89.2)

      7.4 (3.2-12.3)

      <.001

      3

      80.7 (71.3-89.1)

      83.1 (74.0-91.2)

      2.5 (0.0-5.4)

      .024

      4

      76.5 (66.4-85.7)

      87.6 (80.3-93.7)

      11.2 (4.3-18.6)

      <.001

      5

      80.2 (71.3-87.9)

      82.1 (73.6-89.3)

      1.9 (-0.6-4.4)

      .066

      6

      71.6 (61.2-81.0)

      83.7 (75.5-90.7)

      12.1 (6.7-18.3)

      <.001

      7

      70.6 (60.1-80.0)

      85.5 (77.8-92.2)

      14.9 (7.3-23.3)

      <.001

      8

      76.4 (66.6-85.1)

      85.2 (77.4-91.9)

      8.8 (1.2-16.7)

      .012

      9

      80.3 (71.3-88.3)

      82.3 (73.8-89.8)

      2.0 (-1.2-5.1)

      .105

      10

      78.1 (68.6-86.5)

      85.4 (77.5-92.1)

      7.3 (2.6-12.3)

      <.001

      11

      72.0 (61.6-81.3)

      82.6 (74.1-90.1)

      10.6 (5.0-16.8)

      <.001

      12

      76.8 (67.2-85.5)

      84.1 (76.0-91.1)

      7.3 (2.9-12.2)

      <.001

      Conclusion

      Radiologists and pulmonologists were able to significantly improve their assessment of the likelihood of malignancy for an IPN when assisted by AI score.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.