The Nurses' Health Study was initiated in 1976 to study the relation between oral contraceptives and breast cancer. Subsequent funding was available to follow the cohort to address relations between cigarette smoking, postmenopausal hormones, hair dyes, and a range of cancers including breast, endometrial, ovarian, and lung cancer. The 121,700 participating nurses are followed up every 2 years via mail questionnaire to update exposure information to identify incident cancers and other illnesses. Follow-up through 1994 has achieved 90% response from living cohort members. Over the course of the study, additional exposures have been added and refined, including weight at age 18, current weight, height, waist and hip measurements and history of major voluntary weight loss among others. Our focus has been on the health effects of weight gain during middle age. The results relating to diabetes, coronary heart disease, certain types of cancer and total mortality are reviewed. Our primary analytic tools have been multiple logistic regression and Cox proportional hazards models. These methods allow for flexibility in defining the exposures of interest as well as determining their relative importance while controlling for key risk factors. Our models show that even moderate weight gain after age 18 increases risk of each condition. The benefits of physical activity include reduced risk of disease. Issues in the measurement and validation of weight and activity highlight the complexities that are inherent in observational studies addressing the health consequences of lifestyles and anthropometric variables. Our experience of working with repeated measures of body weight and recreational activity are described. Recreational activity has been ascertained in various ways on several questionnaires and may be subject to misclassification. For both weight and activity it may be the pattern of these values that is of importance (e.g. "weight cycling") for some outcomes or conditions. Our research in this area is ongoing. These issues regarding longitudinal measurement will never be completely resolved because weight and activity are intrinsically complex concepts. For these and other such variables, the primary solution is to minimize the problems associated with longitudinal studies. This is best accomplished by developing and maintaining a very strong study design/protocol, including: careful consideration of the sample frame and sample size; maintenance of a high response rate; and continuous monitoring and improvement of the survey/interview instrument(s).
