Abstract P091: A Comparison of Analytical Methods to Evaluate the Association Between Diet and Long-Term Weight Gain in Three Prospective Cohorts
Background: Because long-term weight gain typically occurs insidiously (~1 lb/y) it is very difficult to study in RCTs and prospective cohorts provide crucial evidence on its key contributors. Most prior studies have evaluated how baseline diet, rather than change in diet that may be more physiologically relevant, relates to future weight gain. Aim: To evaluate and compare different methodological approaches for investigating how diet relates to long-term weight gain. Methods: Participants from 3 separate cohorts, the Nurses Health Study (NHS, n=50,422), Nurses Health Study II (NHS II, n=47,898), and the Health Professionals Follow-up Study (HPFS, n=22,557), free of obesity and chronic diseases at baseline, were included and followed for up to 20 y. Lifestyle, health status, and weight were assessed by questionnaires every 2 y, and diet by validated FFQ every 4 y. We assessed 3 different analytic approaches, including relations of 1) baseline diet at the start of each 4 y with weight change in the next 4 y; 2) 4-y change in diet with weight change in the same 4 y; and 3) 4-y change in diet with lagged weight change in the next 4 y. We compared these approaches evaluating consistency across cohorts, magnitudes of associations, and biological plausibility of findings. Results: Across the three methods, consistent, robust, and biologically plausible associations were only seen between changes in diet and changes in weight in the same 4 y (Figure). Findings evaluating baseline diet and lagged dietary change were less consistent across cohorts, far smaller in magnitude, and often not biologically plausible, suggesting presence of both bias and misclassification of the true relevant dietary metric. Conclusions: The methods used to analyze dietary habits and long-term weight gain are crucial. The most robust, biologically relevant, and consistent findings are seen when evaluating dietary change and weight change in discrete periods.