Sunday, October 23, 2011
Grand Ballroom AB (Hyatt Regency Chicago)
Poster Board # 58
(MET) Quantitative Methods and Theoretical Developments

Stephanie L. Bailey, PhD1, Kunnambath Ramadas, MD2, Catherine Sauvaget, PhD3 and Jeremy D. Goldhaber-Fiebert, PhD1, (1)Stanford University, Stanford, CA, (2)Regional Cancer Centre, Trivandrum, India, (3)International Agency for Research on Cancer, Lyon, France

Purpose: Developing countries like India face a growing prevalence of obesity even as substantial populations of underweight individuals persist. Because both obese and underweight individuals face distinct and elevated health risks, we developed microsimulation and calibration procedures appropriate for considering future obesity and underweight-related health in India.

Methods: We constructed 42 individual-level microsimulation models to represent cohorts of urban and rural men and women in India’s states. Each model tracks body mass index (BMI) and BMI changes for individuals over their lifetimes, exposing them to age-, sex-, urban/rural-, state-, and BMI-specific risks of death estimated from life tables and published studies. For projections, individuals begin as adults in 2005-6 based on the 3rd National Family and Health Survey (NFHS) (total n=59,405). Distributions of BMI change rates were estimated from the longitudinal Trivandrum Oral Cancer Study (n=43,055) conditioning on age, sex, and current BMI. We calibrated the BMI change rates for each model based on changes in state-, urban/rural-specific population BMI distributions observed between the 2nd and 3rd NFHS rounds (1998-99 and 2005-6). We compared the performance of Simulated Annealing and Nelder-Mead search algorithms for conducting the calibrations. Outcomes include projections of age- and sex-specific mean BMI, prevalences of underweight (BMI<18.5) and obesity (BMI>27.5) and related mortality patterns.

Results: Both calibration search algorithms achieved good fits for all models, with no model differing from the corresponding NFHS data on state- and urban/rural-specific 5th through 95th percentiles of BMI by more than 2%. Because Nelder-Mead ran approximately 15 times faster than Simulated Annealing, using multiple Nelder-Mead searches to avoid local optima was still faster than using Simulated Annealing. Current 50 year-olds in India have a mean BMI of 22.6, with 9.1% underweight and 21.2% obese. When current 35 year-olds become 50, their mean BMI is projected to be higher (23.2), with a slight increase in underweight individuals (10.1%) and a larger increase in obese individuals (26.8%). Future underweight 50 year-olds account for 23% of deaths.

Conclusions: Given India’s increasing obesity prevalence and health risks from being either obese or underweight, microsimulation and calibration methods aid in capturing complex, policy-relevant trends. While obesity is a growing problem, underweight individuals are projected to continue to represent a substantial and disproportionately high share of deaths in India.