'Bootstrapping' by Variables: Using Resampling to Test the Robustness of Index Variables - an Application to Health Assessments

• Activity Number: 137
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Arnold Mitnitski*+ and Melissa Andrew and Kenneth Rockwood
• Companies: Dalhousie University and Dalhousie University and Dalhousie University
• Keywords: Sub-sampling ; bootstrapping ; index variable ; health status ; variable selection ; frailty index

Abstract

We describe a novel technique of bootstrap-inspired re-sampling of the individual variables used in the construction of an index variable. With this re-sampling within the index, multiple replicates of the index are generated, each consisting of sub-samples of variables from the original complete index. Each replicate index is then analyzed in relation to the dependent variables of choice, allowing for an assessment of both the robustness of the index, and of its sensitivity to inclusion/exclusion of individual variables. We apply this technique to several large data-sets containing up to 100 dichotomized health variables (deficits) to create a frailty index as the ratio of the number of deficits present in the individual to the total number of deficits available. Using sub-sampling from the constituent variables, multiple indices were created by generating a sample with replication from the original data-set. The replicate indices were used in a Kaplan-Meier survival analysis, stratified by quartile of the index. Survival plots of the sub-sampled indices showed little overlap between the strata, suggesting that the index is robust to the specification of individual variables.