The Cox proportional hazards model32 was used to assess the effects of physical activity on the risk of CHD and type 2 diabetes incidence. Physical activity was fitted as a categorical variable for the 5 groups (none, occasional, light, moderate, and moderately vigorous/vigorous). In some of the analyses, tests for linear trend for physical activity were assessed by assigning quantitative values (1-5) for the 5 groups of physical activity and fitting physical activity as a continuous variable rather than as categorical variables. In the adjustment, age; BMI; heart rate; levels of γ-glutamyltransferase (GGT), urate, triglyceride, and HDL cholesterol; and diastolic blood pressure were fitted as continuous variables. Alcohol intake (5 levels: none, occasional, light, moderate, and heavy), smoking (5 levels: never, ex-smoker, and 1-19, 20, and ≥21 cigarettes per day), and social class (7 groups: I, II, III nonmanual, III manual, IV, V, and armed forces) were fitted as categorical variables. Direct standardization was used to obtain age-adjusted rates per 1000 person-years by using the study population as the standard. To determine the possible mediating factors, each biological or metabolic risk factor was fitted in turn to the multivariate model, which included age, BMI, smoking, social class, alcohol intake, and preexisting CHD. The validity of the proportional hazards assumption for these models was assessed by fitting a time-dependent explanatory factor X = X(t), where X(t) = log(t) × physical activity levels.33 At each event time, subjects still alive just before each event time would have their X value changed accordingly. The test for trend over time in the hazard ratio was not statistically significant for any level of physical activity. Logistic regression was used to assess the odds of having elevated levels of the metabolic and biological factors, adjusting for confounders.