Author of

• 20177

  +

  P2.13 - Radiology/Staging/Screening (ID 714)

  • Event: WCLC 2017
  • Type: Poster Session with Presenters Present
  • Track: Radiology/Staging/Screening
  • Presentations: 1
  • Moderators:
  • Coordinates: 10/17/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)

  • 23494

    +

    P2.13-025 - Selecting the Risk Cut off for the LLP Model (ID 9519)

    09:30 - 09:30  |  Presenting Author(s): Kevin Ten Haaf
    • Abstract

    Background:
    The application of risk prediction models for the selection of individuals for lung cancer (LC) screening requires risk thresholds to distinguish between individuals eligible and ineligible for screening. However, little is known about the performance of risk prediction models across different risk thresholds. The UKLS trial utilised the Liverpool Lung Project risk model (LLP~v2~) with a risk threshold of 5% for 5-year LC incidence as the selection criteria in the trial. The UKLS yielded a 1.7% LC detection rate at baseline, which was higher than the NLST or NELSON trials. This study evaluates the performance of different risk thresholds for the selection of individuals for lung cancer screening utilising the LLP~v2~ model.
    
    Method:
    The performance of the LLP~v2~ risk model to predict 5-year LC incidence was evaluated in ever-smokers from the PLCO. The sensitivity (the proportion of LC in the total population that occur within those selected for screening), specificity (the proportion of individuals excluded from screening which do not develop LC), and proportion of individuals eligible for screening was assessed across a wide range of risk thresholds. In addition, the trade-off between the sensitivity and the proportion of individuals eligible for screening was assessed.
    
    Result:
    Applying low risk thresholds yielded high sensitivities, at the cost of low specificities and higher ratios of persons eligible per LC included within the eligible population. For example, a LLP~v2~ risk threshold of 1.0% would yield a sensitivity of 91.5% at the cost of a specificity of 37.2% and a ratio of 39 eligible individuals per LC. In contrast, a LLP~v2~ risk threshold of 5.0% would only yield a sensitivity of 36.5%, but had a specificity of 88.8% and a ratio of 18 eligible individuals per LC. A LLP~v2~ risk threshold of 2.03% yielded a similar sensitivity, but higher specificity and a more favourable ratio of eligible individuals per LC compared to the NLST criteria. LLP~v2~ risk thresholds between 2.0-3.0% may provide an advantageous balance between sensitivity and the ratio of eligible individuals per LC.
    
    Conclusion:
    The level of the risk threshold applied to select individuals for screening has an inverse relationship between the efficacy and efficiency of LC screening. Implementing LC screening programs which use risk prediction models to determine screening eligibility require further assessment of the trade-off between these aspects with regards to the long-term benefits, harms and cost-effectiveness to ascertain the optimal risk threshold.