Premium
“Causal” effects of dieting on obesity: what is the impact of adjustment for time dependent confounding by prior weight?
Author(s) -
Boone Janne E,
GordonLarsen Penny,
Field Alison E,
Adair Linda S
Publication year - 2008
Publication title -
the faseb journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.709
H-Index - 277
eISSN - 1530-6860
pISSN - 0892-6638
DOI - 10.1096/fasebj.22.1_supplement.299.1
Subject(s) - dieting , confounding , obesity , medicine , longitudinal study , underweight , body mass index , demography , weight gain , weight loss , endocrinology , overweight , body weight , pathology , sociology
Prior analyses show associations between dieting behaviors and weight gain, but confounding by prior body weight is possible. In longitudinal analysis, body weight may be a time dependent confounder: a confounder (associated with current dieting and obesity) and a mediator (in the causal pathway between prior and current dieting). Unlike traditional multivariate adjustment (MVA), marginal structural models (MSM) can control for time dependent confounding. To illustrate, we compared longitudinal effects of self‐reported dieting on incident obesity (BMI=30 kg/m2 or BMI=95th percentile CDC/NCHS; self‐reported height and weight) estimated by MSM versus MVA using National Longitudinal Study of Adolescent Health data (Wave I: 1995, ages 11–21; II: 1996; III: 2001; total N=7407), adjusting for sociodemographics and physical activities and excluding underweight and obese adolescents. MVA and MSM estimates of dieting at Waves I and II on incident Wave III obesity were positive, with stronger associations estimated by MSM [OR (95% CI): Wave I: 3.54 (2.07, 6.07); Wave II: 1.90 (1.19, 3.03)] than MVA [OR (95% CI): Wave I: 1.49 (1.10, 2.04); Wave II: 1.05 (0.84, 1.32)]. These results show that the effect of dieting on obesity is sensitive to the method used to adjust for prior body weight, and suggests that weight may be a time dependent confounder that requires appropriate analytical methods. Funding: R01‐041375