Person-years of follow-up were calculated for each participant from the date of the baseline interview to the date of the first fracture, death, cancer diagnosis, or last contact (ie, the date of the second follow-up interview for those who completed 2 follow-up surveys or the date of the first follow-up interview for those who completed only 1 follow-up survey), whichever came first. To better estimate usual dietary intake, we used the mean of intake values reported on the baseline and first follow-up FFQs in the analysis of outcomes that occurred after the first follow-up survey. For outcomes that occurred between the baseline and first follow-up surveys, only information from the baseline FFQ was used. There were 22 566 study participants (92.5%) whose intake values were estimated based on the baseline and first follow-up FFQs and 1837 (7.5%) whose values were based on the baseline FFQ only. Study participants were classified into 5 categories according to quintiles of soy protein or isoflavone intake, with the lowest quintile serving as the reference group. Incidence rates were calculated by dividing the number of events by the person-years of follow-up in each category. The Cox proportional hazards model was used to compute relative risks (the rate ratios of each specific quintile vs the lowest quintile), and to adjust for potential confounding variables. Variables adjusted for in the multivariate analyses included age, cigarette smoking, alcohol consumption, body mass index (calculated as weight in kilograms divided by the square of height in meters), regular exercise, history of diabetes mellitus, education, family income, season of recruitment, and intake of total calories, calcium, nonsoy protein, fruits, and vegetables. Additional adjustments for multivitamin use, tea drinking, type of exercise (including tai chi, folk dance, and walking or jogging), occupation, marital status, and other factors related to domestic violence, such as husband’s education, occupation, and alcohol drinking, did not appreciably alter the results; these variables were, therefore, not included in the final model. Tests for linear trend were performed by using the median intake values for each category of soy protein or isoflavone and modeling them as continuous variables. Further analyses stratified by years since menopause were conducted to assess possible effect modification. Statistical analyses were performed using SAS statistical software, version 9.1 (SAS Institute Inc, Cary, NC). All tests were 2-sided.