FIGURE 3.

Cubic spline curve plot of the association between fish intake and serum testosterone levels. (A) total fish, (B) lean fish, and (C) fatty fish. Solid line represents the predicted levels of serum testosterone based on observed data of fish intake, and colored background represents its 95% confidence intervals. The model (Model 2) is adjusted for age (y, continuous), marital status (married or others), education (<10, 10–12, or ≥13 y), employment status (employed or unemployed), household income (<3, 3–5,9, or ≥6 million Japanese yen/y), BMI (kg/m², continuous), smoking status (never, former, or current), use of cholesterol- or triglyceride-lowering drugs (yes or no), alcohol consumption (nondrinkers; drinkers consuming <23, 23–45.9, or ≥46 g ethanol/d), sleep duration (<7, 7–7.9, or ≥8 h/d), leisure-time physical activity (metabolic equivalents h/wk, quartile), energy intake (kcal/d, continuous), cholesterol intake (mg/1000 kcal, continuous), zinc intake (mg/1000 kcal, continuous), and magnesium intake (mg/1000 kcal, continuous). In the separate analysis of lean and fatty fish, the model was further adjusted for the combined intake of salted fish and dried fish, small fish with bones, and canned tuna, as well as for lean or fatty fish for mutual adjustment.