Cox proportional hazards analysis was used to determine the independent association between CRP levels at baseline and incident type 2 diabetes. The CRP levels were categorized into quartiles based on the concentrations measured in all participants included in this analysis. Variables investigated for possible confounding included age, survey, BMI, smoking status (never smoker, former smoker, current smoker), education (<12 years of schooling, ≥12 years of schooling), leisure time physical activity level (active, not active), alcohol intake (0, >0 and <40, or ≥40 g/d), parental history of diabetes mellitus (negative, unknown, positive), history of myocardial infarction, history of angina, systolic and diastolic blood pressure, actual hypertension, TC level, HDL-C level, and the TC/HDL-C ratio. Since the number of events was small, we used a forward stepping procedure to keep the number of variables to a minimum. A 5% change of the hazard ratio (HR) for any quartile of CRP was used as the criteria for inclusion in the model. Age and survey were forced into the model from the beginning, and at each step the variable that changed the value of the HR most, when added to the other variables already in the model, was also included. Among blood pressure variables, we included only systolic blood pressure in the multivariable modeling process, since it had the strongest confounding effect in the age- and survey-adjusted model. For the same reason, the TC/HDL-C ratio was used. For all covariables used in continuous form in multivariable modeling, departures from linearity were initially checked by the use of categorical indicator variables. P<.05 was regarded as statistically significant. All analyses were performed with SAS statistical software (version 6.12; SAS Institute Inc, Cary, NC).