Multiple Analytical Approaches in the Presence of Missing Data: A Case Study of Esperanza Y Vida Cancer Screening Intervention

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jessica Hersh*+ and Zoran Bursac and D. Keith Williams and Linda Thelemaque and Lina Jandorf and Deborah O. Erwin
• Companies: University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and Mount Sinai School of Medicine and Mount Sinai School of Medicine and Roswell Park Cancer Institute
• Keywords: Missing data ; Imputation ; Logistic regression

Abstract

Missing data is a problem in all research. In longitudinal studies, this problem arises when subjects do not participate in follow-up. Using Esperanza y Vida (EyV), a 3-site, community based intervention, designed to increase cancer screening rates among Latinas as our motivating example, the 3 outcomes of Pap test (PAP), clinical breast exam (CBE), and mammogram (MAM) screening compliance are evaluated at the 2 month follow-up, among women non-adherent at baseline. Overall 78% of participants had complete follow-up data. Three analytical approaches are compared to determine whether there was a significant treatment effect for the 3 outcomes. The 1st approach is complete case analysis (CCA), the 2nd one assumes non-compliance among those with missing follow-up data (last value carried forward - LVCF), and the 3rd one involves logistic regression model based, multiple imputation of the outcomes (LRMI). All analyses except univariate LVCF indicate non-significant treatment effect. Significance of univariate LVCF is likely due to differences between subjects who participated in follow-up and those who did not and because of a reduction in variance when missing values are assigned 0.