Virtual Library

Start Your Search

Valentine Derangere

Author of

  • +

    EP1.01 - Advanced NSCLC (ID 150)

    • Event: WCLC 2019
    • Type: E-Poster Viewing in the Exhibit Hall
    • Track: Advanced NSCLC
    • Presentations: 1
    • Moderators:
    • Coordinates: 9/08/2019, 08:00 - 18:00, Exhibit Hall
    • +

      EP1.01-25 - An Interactive Interface for Prediction of Anti-PD-1 Efficacy in Lung Cancer Patients (ID 2460)

      08:00 - 18:00  |  Author(s): Valentine Derangere

      • Abstract


      The past decade has seen a new field of promising treatments for non-small cell lung cancer patients (NSCLC) with immune checkpoint inhibitors (ICI). Medicine community embarks on a biomarker race to identify the one-quarter of responders and potential hyper-progressors. The assessment of PD-L1 tumor expression by IHC is used to select responder patients and is considered as the gold standard biomarker in many studies but it does not predict the absence of anti-PD-1 efficacy. Recent review from Keenan TE et al. summaries all potential studied markers in literature as intratumoral T cell infiltration, tumor neoantigens or else mismatch repair deficiency. Despite this abundance of potential markers, recommendations only rely on high PD-L1 expression and more recently on high tumor mutational burden (TMB). In this study, we propose estimations of response probability based on the different data that may be available for clinicians according their molecular biology and material means.


      Based on a cohort of 100 patients with advanced NSCLC treated with nivolumab in second line of treatment, we developed an algorithm enabling the calculation of the probability of survival without progression at 6 or 12 months when treated with ICI. Using Cox proportional hazards multiple regression, we adjusted these three stages of information to estimate the probability of response of a patient based on the type of available data: only clinical data and/or exome analysis and/or RNA sequencing. Stability and predictive ability of these models where evaluated internally and externally through bootstrap procedure.


      Among the 100 patients, 90 had both somatic and constitutional exome sequencings available and 50 had an RNA sequencing. We built a main model based on clinical and pathological data easily available for clinicians. As mandatory criteria, the age, sex, performance status, tumor histology, routine mutational status and PD-L1 immunohistochimical expression were first included in the predictive model. In addition, RECIST response to previous chemotherapy line and antibiotics usage were added into the initial model as they were described significantly associated with respectively good and poor survival to ICI in previous publications. Additional criterion were obtained thanks to extensive DNA tumor analysis such as TMB, mutations in DNA repair pathways, the number of large deletions and intratumoral TCR clones. A third stage of data was added with CD8A and CD274 RNA expressions obtained by RNA sequencing. Thus, several models are available for clinicians to estimate the probability of survival without progression at 6 or 12 months for their patients according to the type of available data. An interactive on-line interface based on R Shiny enables the questioning of all these models for further predictions.


      Altogether, these data provide validated and more complex biomarkers according to the type of available data to predict probability of survival without progression to anti-PD-1 in patients with advanced NLCSC. Further predictions can be obtained thanks to a shiny R free and interactive interface available on-line.