Joint Multivariate Response Modeling for Repeated BMI Measures and Single Measures of Adult Cardiovascular Disease Risk Factors

• Activity Number: 87
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Leah Li*+
• Companies: University College London
• Keywords: BMI trajectories ; CVD risk factors ; joint multivariate modeling ; life-course ; cohort study ; Covariance matrix

Abstract

Studies of the association between developmental trajectories and adult health require methods relating developmental measures from different ages to single measures of health outcomes. We present a joint multivariate response model to a longitudinal response variable BMI and adult SBP and HDL-C. We adopt a linear spline model for repeated BMI measures to allow for distinct childhood and adult curves and separate models for SBP and HDL-C. The models are fitted simultaneously assuming the joint distribution of random coefficients. We use the 1958 British Birth Cohort(17,000), whose BMI was recorded at six ages and SBP and HDL-C were measured at 45y. Results show that rate of BMI gain in adulthood has a stronger association with SBP and HDL-C than rate of BMI growth in childhood. For SBP, the estimated correlation for rate of adult BMI gain is 0.27(95% CI: 0.23,0.32) in males and 0.35 (0.31, 0.38) in females, compared to 0.22(0.16,0.27) and 0.18(0.11,0.25) respectively for rate of childhood growth. Similar patterns are found for HDL-C, but in the opposite direction. Joint multivariate response modeling has important applications in life-course epidemiology.