Modeling Prognostic Factors in Gastric Cancer with Binary and Continuous Variables

• Activity Number: 639
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Peipei Shi* and Yanzhi Hsu
• Companies: Eli Lilly and Company and Eli Lilly and Company
• Keywords: Cox regression ; binary variable ; prognostic factor ; gastric cancer

Abstract

Recently 1020 patients were enrolled in two phase 3 randomized double-blind global studies of ramucirumab in second-line advanced gastric cancer: REGARD and RAINBOW. To identify prognostic factors for overall survival, individual patient data were pooled, and 41 key baseline factors were examined. All 22 lab tests, reported as continuous values along with abnormality assessments, were analyzed in 3 ways: 1) binary variable cut by median; 2) best fitted variable form selected from continuous raw value and log-transformed value, as well as binary variable cut by median; 3) binary variables based on lab abnormality: high vs normal or low, low vs normal or high. The model starts with univariate Cox regressions to select covariates with p-value ?0.05, then a multivariable Cox regression was used to make stepwise selection with entry and exit p=0.01. All models were stratified by treatment and geographic region. The 3 approaches yielded similar results, while the binary approaches had considerable appealing for being easy to interpret and simple construction of prognostic index for practical use.