Discrete Time Analysis of Mortality among Persons with Diabetes Using the Logit Model

• Activity Number: 428
• Type: Contributed
• Date/Time: Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Edward Tierney*+
• Companies: Centers for Disease Control and Prevention
• Keywords: Bayesian logistic regression ; mortality ; discrete time analysis ; BMI ; diabetes

Abstract

To assess the role of body mass index (BMI) in mortality, we modeled mortality by age, race, sex, and BMI for a sample of persons with diabetes from 1986 to 1994 with follow up through December 31, 1997. Data are from the National Health Interview Survey linked to the National Death Index and include 1,032 deaths among 28,783 person years of follow-up. We discretized the data by one year intervals from 19 to 85 years of age and used Bayesian logistic regression to model hazard rates. We used a cubic spline with five knots for age (as the time variable) and included terms for race, sex, and BMI. Odds ratios and 95% posterior intervals, compared to the normal BMI groups, are 2.3 (1.4, 3.3) for underweight, 0.89 (0.76, 1.0) for overweight, 0.85 (0.69, 1.0) for obese, and 0.95 (0.75, 1.2) for severely obese. Only the underweight group has significantly higher death rates than the normal BMI group.