FIGURE 1.

Pooled results for the relationships between changes in coffee habits (1 serving per day) and absolute weight change (kg) within each 4-y period: stratified by age or BMI. Pooled results are derived from the results of NHS, NHS II, and HPFS using inverse-variance-weighted random-effects meta-analysis. All models were multivariable adjusted: intercept of 4-y time period, age, BMI, daily hours of sleep (≤6 h/d, >6 ∼ ≤7 h/d, >7 ∼ ≤8 h/d, >8 h/d), change in physical activity (MET-h per week), change in smoking status (never-never, current-past, past-current, never-current, past-past, current-current), changes in sugar-sweetened beverages, fruit juice, water, tea (all variables in servings/d), socioeconomic status, change in alcohol intake (g/d), adherence to AHEI-2010 ex alcohol, cooked and fried potatoes, other fried food, sweets and desserts and change in total energy intake. In NHS and NHS II: sedentary time (≤1 h/wk, 2–5 h/wk, 6–20 h/wk, 21–40 h/wk, 41+ h/wk); in HPFS change in sedentary time (> 4h less, 1.5h ∼ 4h less, 0 ∼ 1.5h less, no change, up to 1.5h more, 1.5 ∼ 4h more, > 4h).

The error bars indicate 95% CIs.