Confounding Bias Correction for Longitudinal Data

• Activity Number: 139
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yuanzhang Li*+ and Robert Yolken and David Niebuhr
• Companies: Walter Reed Army Institute of Research and Johns Hopkins School of Medicine and Walter Reed Army Institute of Research
• Keywords: bias correction ; longitudinal ; case control ; simulation ; confounding ; schizophrenia

Abstract

Studying change from baseline measure in longitudinal data promises to lead to early identify a disease which would benefit both patients and society. Categorization of the first measurement may confound the values of repeated measures and generate bias. We developed modeling strategies for case-control longitudinal studies to estimate and correct the bias. We use this approach in a military dataset to evaluate the association of antibody risk of developing schizophrenia. Cases and matched controls were grouped into categories by 50th and 75th percentile of the first sample of cases. The bias generated from such confounding is corrected by simulation to develop potentially unbiased estimation. Seven antibody agents were studied. This proposed approach can aid investigators to identify risk factor in the pre-clinical period, and it can be extended to other longitudinal studies.