CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Sensitivity analysis: E-value based on direct adjusted survival probabilities Authors:  Zhang Mei-Jie - Medical College of Wisconsin (United States) [presenting]
Cheng Zheng - University of Nebraska Medical Center (United States)
Zhenhuan Hu - Medical College of Wisconsin (United States)
Abstract: The estimated treatment effect from an observational study may be biased, and potential unmeasured confounding bias could be a central limitation of observational studies. Sensitivity analysis is useful in assessing the robustness of the estimated treatment effect based on a statistical analysis of observational data. Recently, a new sensitivity analysis method called E-value has been proposed, which measures potential inference of unmeasured confounding in observational studies. E-value is defined as the minimum strength of association of an unmeasured confounder would need to explain away a treatment-outcome association. A common approach to testing for differences in survival rates between two therapies while adjusting for prognostic factors is to compare the direct adjusted survival curves. The direct adjusted survival probabilities are estimated by taking the average of the individual predicted survival curves over the entire study cohort, which control for the possible imbalance of patient characteristics between treatment groups. We suggest a sensitivity analysis using an E-value for comparing two direct adjusted survival probabilities. A real stem cell transplant data example illustrates the practical utility of the proposed method.