The participants were selected from a community-based cohort of the Korean Genome and Epidemiology Study. The study design has been described in detail previously [17]. In brief, the cohort initially included 10,038 participants aged 40–69 years. Baseline examinations were performed between 2001 and 2003, and follow-up examinations are conducted biennially. The total follow up rates of participants were 85.7% and 74.9% respectively, at the 2 and 4-year follow-up examinations. Participants included residents of a rural area (Ansung) and an industrialized area (Ansan). Of the 10,038 baseline participants, 2,317 postmenopausal women completed the food frequency questionnaire at the baseline survey. The speed of sound (SOS, m/s) at their radius or tibia had been measured at the baseline survey, primarily follow up survey (2-years) or secondary follow-up survey (4-years). A total of 744 participants were excluded for following reasons: no dietary data (n = 93); those who reported daily energy intake was lower or higher than 2 standard deviations (SD) (< 419.2 kcal; n = 5 or > 3,341.1 kcal; n = 111) [18]; early menopause before 40 years of age (n = 194); bilateral ovariectomy (n = 25); on thyroid drug use (n = 93); on hormone therapy (n = 26); or on insulin therapy (n = 23); osteoporosis subjects at both sites at baseline survey (n = 103), no quantitative ultrasound (QUS) data at baseline (n = 52), or no follow up data (n = 19). Finally, we used the data of 1,573 subjects for the analysis (1,487 subjects had normal values for the radius, whereas 1,212 subjects had normal values for the tibia. Among them, 1,126 subjects had normal values at both sites). Informed consent was obtained from all study participants. The data was provided with bioresources from the National Biobank of Korea, the Centers for Disease Control and Prevention, Republic of Korea (4845-301, 4851-302 and -307). The study protocol was approved by the Eulji University institutional review board (EUIRB2015-33).

Dietary intake was assessed by the use of a validated 103 food-item, semi-quantitative food frequency questionnaire (SQFFQ) which administered at the baseline survey. The procedure used to design the SQFFQ is described in detail elsewhere [19]. The frequency of servings was classified into nine categories (never or seldom, once a month, 2–3 times a month, 1–2 times a week, 3–4 times a week, 5–6 times a week, once a day, twice a day, and three times or more a day). The portion size of each food item was classified into three categories (small, medium, and large). This SQFFQ has been validated using 12-day diet record data in the Korean population. The correlation of two methods ranged between 0.23 (vitamin A) and 0.64 (carbohydrate). The median for all nutrients was 0.39 [20]. Nutrient intakes were calculated for each participant using the food composition tables of the Korean Nutrition Society [21]. This FFQ contains dairy foods as follows: milk, yogurt, cheese, and ice cream. Dairy product intake was calculated as the combined intake of milk and yogurt.

Height and weight were measured by trained staff using a scale and a wall-mounted extensometer to the nearest 0.1 cm and 0.1 kg, respectively. Body mass index (BMI) was calculated as the weight in kilograms (kg)/squared of height in meters (m²). Sound velocity, usually expressed in meters per second and termed SOS, was measured using the QUS at the distal radius and the mid-tibia (Omnisense 7000s, Sunlight Medical Ltd. Israel). The measurement sites were the distal radius and the mid-tibia of the non-dominant arm and leg, respectively. The measurement was performed three times, and their average was taken as the final value as described previously [2223]. QUS provides the T-score value of each measurement site [24]. Hans et al. [25] reported that the optimal discrimination threshold of T-score was −2.5 when using Omnisense. Therefore, we defined osteoporosis as T-score ≤ −2.5 following the World Health Organization criteria [26]. The osteoporosis incidence defined individuals who were not diagnosed with osteoporosis at the baseline but subsequently became osteoporotic in the follow-up study.

General characteristics, socioeconomic status, and lifestyle data were collected by an interviewer-administered questionnaire. Data regarding baseline age, age at menopause and residential area [Ansung (rural) or Ansan (urban)] were collected. Educational level was categorized as graduate elementary school, middle school, high school, or college or higher degree. Income status data were collected as amount per month and categorized as < $500, $ 500–$1,500 and ≥ $ 1,500. As for health behavior, smoking was categorized as non/ex-smoker or current smoker due to the small number of ex-smoker. Drinking information was collected then categorized as non/ex-drinker or current drinker due to the small number of ex-drinker. Exercise was categorized as yes or no.

Descriptive statistics such as mean values and standard error were used to summarize continuous variables, and frequencies were expressed as percentages. Categorical variables were tested by the Chi-square test and continuous variables were tested by the t-test. Hazard ratios (HRs) and 95% confidence intervals (CIs) for osteoporosis incidence risk associated with frequencies of dairy products, milk, and yogurt were estimated using the Cox proportional hazards regression model. The study was initiated in June 2001, and the participants were followed-up until November 2006. The entry time was the data at recruitment and the exit time was defined as the date at osteoporosis diagnosis or date of attended first or second follow-up survey. Person-years were calculated from the date of baseline examination to the date of an event or loss to follow up. The multivariate models adjusted for age, BMI, residual area, education level, income status, smoking, drinking, physicals activity, and total energy intake. To assess whether there was any linear association between milk, yogurt or dairy product intake and osteoporosis, tests for linear trend were performed by treating the median value of each exposure intake category as a continuous variable. The mean differences in SOS values in the radius and tibia according to the intake frequencies of dairy products, milk and yogurt were analyzed using analysis of covariance (ANCOVA). Age-adjusted values of baseline and 2^nd follow up SOS and age and baseline SOS adjusted % change were used for ANCOVA. Statistical significance was accepted at P < 0.05 in the two-tailed test. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

During the 4-year of follow-up, we had identified 273 newly diagnosed osteoporosis cases (54/1,000 person-years) in the radius and 407 (101/1,000 person-years) in the tibia. The median follow up years was 3.4 years (range 1.5–4.5 years) of radius incidence and 3.3 years (range 1.4–4.3 years) of tibia incidence. The general characteristics of the study population according to radius and tibia osteoporosis status are presented in Table 1. Women in the osteoporosis group were more likely to be older than those in the normal group. The osteoporosis group was more likely to have higher BMI, less education, lower household income, and lived more in a rural area. Health-related variables such as smoking and physical activity did not differ between the two groups. The radius osteoporosis group had a significantly lower intake of milk, yogurt, and dairy products than the normal group.

Table 2 shows the age-adjusted and multivariate-adjusted HR of having osteoporosis of the radius and tibia according to the frequency of dairy products, milk, and yogurt. We found that frequent consumption of dairy products was associated with a lower incidence of radius osteoporosis (HR = 0.51, 95% CI: 0.33–0.80 for 1 time/day vs. non-consumer; P for trend = 0.0027). Similarly, frequent consumption of milk and yogurt reduced the risk of radius osteoporosis (milk: HR = 0.60, 95% CI: 0.42–0.87 for >5–6 times/week vs. non-consumer; P for trend = 0.0130, yogurt: HR = 0.51, 95% CI = 0.30–0.85 for >5–6 times/week vs. non-consumer; P for trend = 0.0167). However, the consumption of dairy products, milk, and yogurt showed no reduction in osteoporosis risk in the tibia.

The SOS values and the % change in SOS values according to dairy products, milk, and yogurt intake frequencies are reported in Table 3. The SOS value in the second follow up, and % change in SOS did not show any differences according to dairy product intake frequency. However, subjects in the highest frequency (>1 / day) of the dairy product group had a higher tibia SOS value than those in the other groups (3,877.6 ± 8.4 m/sec at the baseline, 3,777.3 ± 9.6 m/sec at the second follow up).