Author of

• 17556

  +

  P2.03a - Poster Session with Presenters Present (ID 464)

  • Event: WCLC 2016
  • Type: Poster Presenters Present
  • Track: Advanced NSCLC
  • Presentations: 1
  • Moderators:
  • Coordinates: 12/06/2016, 14:30 - 15:45, Hall B (Poster Area)

  • 17611

    +

    P2.03a-055 - Predicting Risk of Chemotherapy-Induced Severe Neutropenia in Lung Patients: A Pooled Analysis of US Cooperative Group Trials (ID 3975)

    14:30 - 14:30  |  Author(s): X. Cao
    • Abstract

    Background:
    Neutropenia is the most serious hematologic toxicity associated with the use of chemotherapy. Severe neutropenia (SN) may result in dose delays and/or reductions, and the use of growth colony stimulating factors (CSFs) increases the cost of therapy. Lyman et al. (2011) published a risk model to predict individual risk of neutropenia in patients receiving chemotherapy for multiple types of cancer. The Lyman model (LM) has not been validated by external datasets. We investigated the LM with a large external lung cancer dataset based on clinical criteria of SN and investigated new risk prediction models for SN.
    
    Methods:
    Stage IIIA/IIIB/IV non-small cell lung cancer (NSCLC) and extensive small cell lung cancer (SCLC) chemotherapy phase II/III trials completed in 1990-2012 were assembled from U.S. cancer cooperative groups. SN was defined as any neutropenic complications grade ≥ 3 according to CTCAE. A risk score was calculated as a weighted sum of regression coefficients of the LM for all patients in the database. The performance of risk models was evaluated by the area under the ROC curve (AUC) with a good model defined as AUC ≥ 0.7. To develop new risk models, a random split was used to divide the database into training cohort (2/3) and testing cohort (1/3). Multivariable logistic regression models with stepwise selection and lasso selection (Tibshirani, 1996) were built in training cohort and validated in testing cohort. Candidate predictors included patient-level and treatment-level variables. The patients with complete data were used for validation and all patients, including those with imputed predictors, were used to develop new risk models.
    
    Results:
    Eighty seven trials with 14,829 patients were included. The LM had a good performance in SCLC patients (AUC=0.86), but it had poor performance in NSCLC patients (AUC=0.47), and an overall unsatisfactory performance in all patients (AUC=0.56). The stepwise model had superior performance than the lasso model (AUC: 0.84 vs. 0.76) in training, while the lasso model had smaller shrinkage in testing. A parsimonious model, based on histology, prior chemo, platinum-based, taxanes, gemcitabine, CSFs, age as continuous variable, relative dose intensity, and white blood cell (WBC), performed slightly worse (AUC=0.71) in testing than the stepwise model and the lasso model.
    
    Conclusion:
    The U.S. cooperative group data failed to validate the LM in predicting the risk for severe neutropenia in lung cancer patients receiving chemotherapy. The parsimonious model involving nine predictors showed good performance in predicting severe neutropenia. Prospective validation is warranted.