The NHS was started in 1976 when 121,700 married female registered nurses aged 30–55 years received a questionnaire on health status and potential risk factors for major chronic diseases. Ever since, participants were sent questionnaires biennially and the response rates have been ~90% (8
). The NHS was approved by the institutional review board of the Brigham and Women's Hospital (Boston, MA).
For the present analysis, we used the return of the 1984 questionnaire, which was completed by 97,510 participants, as baseline because information for family history was first collected in 1982 and the first extended food frequency questionnaire (FFQ) was administered in 1984. In this 1984 questionnaire, complete dietary information was available for 81,757 women. Women with a history of type 2 diabetes at baseline (n = 329) were excluded. Furthermore, we excluded participants with cancer (except for nonmelanoma skin cancer) or cardiovascular diseases at baseline (n = 4,934), because the diagnosis of these diseases could have interfered with self-reports of lifestyle and family history. Women with incident types of diabetes other than type 2 diabetes or unconfirmed type 2 diabetes were also excluded (n = 1,480). Moreover, women with missing information for diabetes status, family history of diabetes, date of birth, weight, or height were excluded (n = 1,787). As a result, 73,227 women remained for the present analysis. There were no differences between the total study population (n = 81,757) and the population included for this analysis (n = 73,227) for baseline characteristics, except for slight differences in mean age (50.6 and 50.2 years), the proportion of participants with obesity (13.3 and 12.3%), mean weight change since age 18 years (28.3 and 27.6 kg), and the proportion of participants reporting to drink no alcohol (31.5 and 30.6%) or >10 g of alcohol/day (23.6 and 24.0%).
Assessment of family history of diabetes and parental body size
Women were asked to report whether any of their first-degree family members (father, mother, and/or siblings) ever had diabetes in the questionnaires mailed in 1982, 1988, and 1992. No questions about the type of diabetes in family members were included.
To assess obesity in the parents, a series of pictograms of body shape were included in the 1988 questionnaire (9
). The pictograms estimate relative obesity, with values ranging from 1 (very lean) to 9 (very obese). Women were asked to choose the pictogram that best described the body shape of their natural mother and their natural father at age 50 years. Previous research has shown that these pictograms can provide a reasonably accurate estimate of measured BMI of the parents 15 years in the past (r
= 0.74 for mothers and 0.63 for fathers) (9
). According to the chosen image, both father and mother were categorized as nonobese (image 1–5) or obese (image 6–9).
Assessment of lifestyle, socioeconomic status, race, and adiposity
Dietary information was collected using a semiquantitative FFQ included in the questionnaires mailed to the participants every 2–4 years. The reported portions were converted to gram weights per serving and intakes of nutrients were computed by multiplying the frequency of consumption by the nutrient content in grams. The FFQ has been validated against biomarkers and diet records (10
). Physical activity was assessed every 2–4 years, asking about the time spent on vigorous and moderate physical activity on an average day in the last month. Information on cigarette smoking was assessed every 2 years. Information on the highest educational degree of the participant and her husband, father's and mother's occupation, and race was assessed in the 1992 questionnaire.
Height was assessed in the baseline questionnaire and body weight every 2 years. BMI was calculated as weight in kilograms divided by the square of height in meters. In 1980, women were asked to report their weight at age 18 years, which we used to calculate BMI at age 18 and weight change since age 18. In 1986, 1996, and 2000, waist and hip circumferences were assessed by questionnaires that included measurement instructions and a tape measure. The validity of self-reported weight and circumference measurements was assessed in 140 participants from the NHS aged 41–65 years. Pearson correlations between self-reported and technician-measured weight, waist circumference, and hip circumference were high (r
= 0.97, 0.89, and 0.84, respectively) (12
Ascertainment of type 2 diabetes
Women who reported having diabetes in the biennial questionnaires were mailed a supplemental questionnaire to gather more information on diagnosis and treatment. The National Diabetes Data Group criteria (13
) were used to confirm diagnosis of diabetes according to 1
) an elevated glucose concentration (fasting plasma glucose of ≥7.8 mmol/l, random plasma glucose of ≥11.1 mmol/l, or plasma glucose ≥11.1 mmol/l after an oral glucose load) and at least one symptom related to diabetes (excessive thirst, polyuria, weight loss, or hunger); 2
) no symptoms, but elevated glucose concentrations on two occasions; or 3
) treatment with insulin or oral hypoglycemic medication. For cases of type 2 diabetes identified after 1998, the American Diabetes Association criteria (14
) were used, lowering the cutoff point for fasting plasma glucose concentrations to 7.0 mmol/l. In a subsample, type 2 diabetes that was confirmed by the supplementary questionnaire was consistent with medical record reviews by an endocrinologist in 98% of the women (15
Person-time for each participant was calculated from the date of return of the 1984 questionnaire to the date of diagnosis of type 2 diabetes, death, or June 2004, whichever came first. Cox proportional hazards analysis was used to estimate the relative risk (RR) for type 2 diabetes according to family history. To evaluate the contribution of different covariates on the association between family history and diabetes risk, we used multivariate models with the following adjustments: 1) age; 2) age, socioeconomic status (SES) (occupation of the father, occupation of the mother, husband's highest degree, and participant's highest degree), and race; 3) age, SES, race, and lifestyle (smoking, physical activity, polyunsaturated/saturated fat intake ratio and intakes of coffee, alcohol, fruit, vegetables, sugar-sweetened beverages, whole grains, red meat, trans-fat, and total energy); and 4) age, SES, race, lifestyle, and BMI. Additional analyses were performed to evaluate whether inclusion of waist circumference or a combination of BMI at age 18 and weight change since age 18 in the statistical model would alter the results. In addition, we conducted a sensitivity analysis evaluating the effect of blood glucose testing on the results. Since 1998, women were asked biennially to report whether they had been screened for fasting blood glucose in the previous 2 years. We repeated our analysis for follow-up cycles from 1996 to 2004 restricted to women who reported to have been tested for blood glucose during a 2-year follow-up period.
All covariates were updated during follow-up whenever new information was obtained. For dietary variables we used cumulative updating to reduce within-person variation (16
). The proportion of the association explained by including different covariates in the model was calculated based on the change in regression coefficients, using the method of Lin et al. (17
) to calculate the 95% CI.
All P values were two-tailed, and P < 0.05 was considered to be statistically significant. Inclusion of multiplicative interaction terms between time and family history in the multivariate models suggested that the proportional hazards assumption was not violated. The SAS statistical program (version 9.1; SAS Institute, Cary, NC) was used for all analyses.