The Nurses’ Health Study began in 1976, when 121,700 female nurses 30 to 55 years of age responded to a questionnaire regarding medical, lifestyle, and other health-related information.16 Since then, questionnaires have been sent biennially to update this information and identify newly diagnosed cases of various diseases. Diet was first assessed in 1980. For the current analysis, we excluded women with previously diagnosed diabetes, cancer, or cardiovascular diseases at base line and those who left more than 10 items blank on the 1980 dietary questionnaire or had implausibly low or high scores for total intake of food or energy (less than 500 or more than 3500 kcal per day). After these exclusions, the analysis included 84,941 women. The follow-up rate with respect to the incidence of diabetes in the overall cohort was 97 percent of the total potential person-years of follow-up. The study was approved by the institutional review board of Brigham and Women’s Hospital in Boston; completion of the self-administered questionnaire was considered to imply informed consent.
Assessment of Diet
In 1980, we assessed diet using a 61-item, semiquantitative food-frequency questionnaire.17 An expanded dietary questionnaire including approximately 120 items was used to update the information about diet in 1984, 1986, and 1990.18 We asked how often, on average, a participant had consumed a particular amount of a specific type of food during the previous year. The intake of nutrients was computed by multiplying the frequency of consumption of each unit of food by its nutrient content. Questions about the consumption of beer, wine, and liquor were included in each questionnaire. The reproducibility and validity of the food-frequency questionnaires have been described in detail previously.18,19
Assessment of Nondietary Factors
Every two years, we update participants’ smoking status (never smoked, former smoker, or current smoker, including the number of cigarettes smoked per day), menopausal status and use or nonuse of postmenopausal hormone therapy, and body weight. Reported weights have been highly correlated with measured weights (r=0.96).20 The presence or absence of a family history of diabetes (in first-degree relatives) was assessed in 1982 and 1988. Information about physical activity was first obtained in 1980 and was updated in 1982, 1986, 1988, and 1992 with the use of a validated questionnaire.6 We estimated the amount of time per week spent in moderate-to-vigorous activities (including brisk walking) requiring the expenditure of 3 MET or more per hour.6
Definition of the Low-Risk Group
The criteria we used to define a low-risk group according to dietary and lifestyle variables were similar to those used in previous analyses of coronary disease.21 In terms of the body-mass index (the weight in kilograms divided by the square of the height in meters), low risk was defined as a value of less than 25.0, the standard cutoff point for the classification of overweight.22 We did not include waist or hip circumferences in the analyses because they were first assessed in 1986 and because a high body-mass index was a much stronger predictor of diabetes in this cohort.23
In terms of physical activity, low risk was defined as an average of at least one half-hour per day of vigorous or moderate activity, including brisk walking, in keeping with published guidelines.24,25 In terms of cigarette smoking, low risk was defined as no current smoking, and in terms of alcohol use, low risk was defined as an average of 5 g or more of alcohol per day (about half a drink or more per day). Because few women in this cohort drank heavily (1.2 percent reported drinking more than 45 g of alcohol per day), we did not define an upper limit for alcohol consumption, although clearly such a limit would be necessary in order to establish public health guidelines.
Previous studies have found that a reduced risk of type 2 diabetes is associated with a higher intake of cereal fiber11,12,26 and polyunsaturated fat27 and that an increased risk is associated with a higher intake of trans fat (formed during the partial hydrogenation of vegetable oils)27 and a higher glycemic load (which reflects the effect of diet on the blood glucose level).11,12 Therefore, a low-risk diet was defined as a diet low in trans fat and glycemic load and high in cereal fiber, with a high ratio of polyunsaturated to saturated fat. For each dietary factor, we assigned each woman a score between one and five, corresponding to her quintile of intake, with five representing the lowest-risk quintile, and summed her quintile values for the four nutrients. Participants with composite dietary scores in the highest 40 percent among the women in the study were considered to be in the lowest risk category in terms of diet.
Ascertainment of Cases of Diabetes
A supplementary questionnaire regarding symptoms, diagnostic tests, and hypoglycemic therapy was mailed to women who reported having received a diagnosis of diabetes. A case of diabetes was considered to be confirmed if at least one of the following was reported on the supplementary questionnaire: classic symptoms plus a plasma glucose concentration of at least 140 mg per deciliter (7.8 mmol per liter) in the fasting state or a randomly measured plasma glucose concentration of at least 200 mg per deciliter (11.1 mmol per liter); at least two elevated plasma glucose concentrations on different occasions (a concentration of at least 140 mg per deciliter in the fasting state, a randomly measured concentration of at least 200 mg per deciliter, or a concentration of at least 200 mg per deciliter two or more hours after the initiation of oral glucose-tolerance testing) in the absence of symptoms; or treatment with hypoglycemic medication (insulin or an oral hypoglycemic agent). Our criteria for the classification of diabetes are consistent with those proposed by the National Diabetes Data Group.28 The validity of this questionnaire has been verified in a subsample of our study population.5 The diagnostic criteria for type 2 diabetes changed in June 1996, and a fasting glucose concentration of 126 mg per deciliter is now considered the threshold for a diagnosis of diabetes.29 We used the earlier criteria because all the cases in our cohort were diagnosed before June 1996.
The duration of follow-up was calculated as the interval between the return of the 1980 questionnaire and the diagnosis of type 2 diabetes, death, or June 1, 1996. Relative risks were calculated by dividing the incidence of diabetes among women in the low-risk group by the incidence among the remaining women. To adjust for multiple risk factors, we used pooled logistic regression with two-year intervals,30 which is approximately equivalent to Cox regression for time-dependent covariates. In all models, we simultaneously included terms for age, time (eight periods), presence or absence of a family history of diabetes, menopausal status, and use or nonuse of postmenopausal hormone therapy. In the initial analyses, we calculated the relative risks and 95 percent confidence intervals31 for the different categories of each variable that was included in the low-risk profile, adjusting for age, time, presence or absence of a family history of diabetes, menopausal status, and use or nonuse of postmenopausal hormone therapy but not for the other components of the low-risk profile. We then examined the combined low-risk group, defined as women in the low-risk category for each variable, with all other women as the comparison group.
We calculated the population attributable risk,31,32 an estimate of the percentage of cases of type 2 diabetes in this population that would theoretically not have occurred if all women had been in the low-risk group, assuming a causal relation between the risk factors and type 2 diabetes. We also conducted analyses stratified according to the presence or absence of a family history of diabetes and according to the body-mass index. Within each stratum, we compared the women in the low-risk category with all the other women.
To obtain the best estimate of long-term dietary intake and physical activity, we used the cumulative-update method,33,34 which takes the average of all previous data. For variables unrelated to diet and exercise, we used the most recent information; the body-mass index and smoking status were updated every two years, and the information about alcohol intake was updated in 1984, 1986, and 1990.