Protective Effects of Dietary Carotenoids on Risk of Hip Fracture in Men: The Singapore Chinese Health Study



Experimental and epidemiologic data suggest that carotenoids in vegetables and fruits may benefit bone health due to their antioxidant properties. The relationship between dietary total and specific carotenoids, as well as vegetables and fruits, and risk of hip fracture was examined among Chinese in Singapore. We used data from the Singapore Chinese Health Study, a prospective cohort of 63,257 men and women who were of ages 45 to 74 years between 1993 and 1998. At recruitment, subjects were interviewed on lifestyle factors and medical history. Usual diet was measured using a validated food frequency questionnaire. During a mean follow-up of 9.9 years, we identified 1630 hip fracture incident cases. Among men, consumption of vegetables was associated with lower hip fracture risk. Similarly, dietary total carotenoids and specific carotenoids, α-carotene, β-carotene, and lutein/zeaxanthin were inversely associated with hip fracture risk. Compared to men in the lowest quartile of nutrient density, men in the highest quartile had statistically significant 26% to 39% risk reduction (all p for trend <0.05). When stratified by body mass index (BMI), the greatest protective effects of total vegetables and carotenoids were found in men with BMI <20 kg/m2 (p for trend ≤0.004). There was no association between dietary carotenoids or vegetables/fruits and hip fracture risk among women. This study suggests that adequate intake of vegetables may reduce risk of osteoporotic fractures among elderly men and that the antioxidant effects of carotenoids may counteract the mechanism of osteoporosis related to leanness. © 2014 American Society for Bone and Mineral Research.


Osteoporosis is an epidemic problem affecting both elderly men and women worldwide. Although the prevalence of osteoporosis and related fractures is higher in women than in men, men experience greater comorbidity and mortality after hip fracture than women.[1, 2] Studies have shown that osteoporosis in men is underrecognized and understudied epidemiologically and clinically.[3-5] Hence, understanding the risk factors and the underlying mechanism of hip fracture would help to develop a preventive strategy for osteoporotic fractures.

Evidence from in vitro and in vivo studies suggests that reactive oxygen species (ROS) may increase osteoclastogenesis[6, 7] and osteoclastic differentiation,[7, 8] or suppress osteoblastic differentiation.[9, 10] Therefore, increased oxidative stress may play an important role in age-related bone loss[11, 12] that leads to osteoporotic fractures. Dietary antioxidant carotenoids, primarily derived from fruits and fresh vegetables, have been reported to promote bone formation in experimental studies,[13, 14] and to protect against osteoporosis and related fractures in some but not all epidemiological studies.[15-20] Additionally, lycopene, a carotenoid, was shown to inhibit osteoclast formation in rat bone marrow cultures,[21] and was negatively related to bone resorption biomarkers and oxidative stress in postmenopausal women.[22, 23] Taken together, these data raised a hypothesis that antioxidant carotenoids may counteract the adverse effect of ROS in bone turnover, and therefore can potentially protect against osteoporotic fractures.

Leanness has been established as an independent risk factor of hip fracture.[24-26] Recently, several observational studies have reported that low body mass index (BMI) could be related to increased oxidative stress,[27-29] particularly, in men.[28] Hence, in the present study, we aimed to assess the association between dietary total and specific carotenoids, including α-carotene, β-carotene, β-cryptoxanthin, lycopene, and lutein/zeaxanthin, and the risk of hip fracture among men and women. In addition, we also evaluated whether BMI modified the effects of these dietary components on the risk of hip fracture in an Asian population, who generally have lower BMI than their Western counterparts.[30]

Subjects and Methods

Study population

This study was conducted in the Singapore Chinese Health Study, a community-based prospective cohort study to investigate diet, lifestyle factors, and risk of chronic diseases. We enrolled 63,257 middle-aged to older men (n = 27,959) and women (n = 35,298) between April 1993 and December 1998. The study participants were restricted to two major dialect groups in Singapore, the Hokkiens and the Cantonese, who originated from Fujian and Guangdong provinces in Southern China. During the enrollment period, all of our study participants were residents of government housing estates, where 86% of the Singapore population was housed. This study was approved by the Institutional Review Board at the National University of Singapore, and all enrolled subjects gave informed consent.

Baseline assessment was conducted through a face-to-face structured interview during the initial enrollment. Information was recorded by a trained interviewer using a structured questionnaire that included demographics, medical history, cigarette smoking, alcohol consumption, physical activity, and detailed menstrual and reproductive history (women only). Body weight and height at baseline were self-reported during the interview. There were 9781 cohort participants with unknown weight, 97 with unknown height, and 192 with both unknown weight and height. BMI is calculated as weight in kilograms divided by height in meters squared. For those with missing weight and/or height, BMI was calculated using imputed weight and/or height derived from the linear regression equation: Weight = y-intercept + gradient × height, where values for the y-intercept and gradient were derived from gender-specific weight-height regression lines obtained from all cohort participants with known heights and weights. This method of imputed BMI has been reported in detail.[31]

Dietary intake was recorded using a 165-item semiquantitative Food Frequency Questionnaire (FFQ), which incorporates common and distinct food items in Singapore. The FFQ had been validated previously using 24-hour recalls and re-administration of the FFQ among a subset of 810 participants.[32] The validation study by these two methods showed similar distributions, with most mean pairs for energy and nutrients within 10% of each other's values. The correlation coefficient by these two methods for each dietary component ranged between 0.24 and 0.79, which is comparable with previous validation study in diverse populations.[33] The dietary intake of each nutrient was derived based on the Singapore Food Composition Database, which has been described in detail elsewhere.[32] This database was developed specifically for this cohort study and listed approximately 100 nutritional and non-nutritional values per 100 g of the edible raw and cooked foods. The physical activity portion of the questionnaire was modeled after the European Prospective Investigation in Cancer (EPIC) study physical activity questionnaire, which has been shown to be valid and reproducible.[34] More information about the questionnaire used in this cohort can be found in the online Supplemental Material.

Case ascertainment

Hip fracture cases were identified via record linkage analysis with hospital discharge databases of the MediClaim System, which captures inpatient discharge information from all public and private hospitals in Singapore. All cases were verified manually by surgical records or medical records. As of December 31, 2010, only 47 subjects from this cohort were known to be lost to follow-up due to migration out of Singapore or for other reasons. After excluding 4 cases of traumatic fractures from road traffic accidents, and 1 case due to cancer metastasis in the femur, 1733 hip fracture cases were identified through record linkage. We excluded from statistical analysis of 103 prevalent cases of hip fracture, which occurred before enrollment to the cohort for these participants. Thus, 1630 fracture cases and 61,524 subjects without fractures were included in the final analysis. The causes and date of death of all cohort participants were ascertained through record linkage with the population-based Singapore Registry of Births and Deaths.

Statistical analysis

For each study subject, person-years were counted from the date of baseline interview to the date of hip fracture diagnosis, death, migration, or December 31, 2010, whichever occurred first. The Cox proportional hazards regression was used to assess the association between baseline dietary carotenoids, fruits and vegetables, and hip fracture risk by comparing higher quartiles to the lowest quartile.[35] Quartiles of each energy-adjusted food or nutrient were based on the distribution of both men and women combined in the whole cohort. The strength of the association between a dietary factor and hip fracture risk was estimated by hazard ratios (HRs) and their corresponding 95% confidence intervals (CI). To examine linear trend, ordinal values of the quartile of each dietary component was entered as a continuous variable in the Cox proportional hazards model. We did not identify any violation of the proportional hazard assumption or multicollinearity among the covariates that were entered in the models.

All models in the analyses included the following covariates: age (continuous), year at recruitment (1993–1995 and 1996–1998), dialect group (Hokkien, Cantonese), level of education (no formal education, primary school, secondary school or higher), BMI (kg/m2, continuous), smoking status (never smokers, ex-smokers, current smokers), moderate physical activity (none, 2–3 hours weekly, 4+ hours weekly), calcium (mg/1000 kcal/d), soy isoflavones (mg/1000 kcal/d), pyridoxine (mg/1000 kcal/d), total energy intake (kcal/d), menopause status (women only; yes, no), use of hormone replacement therapy at recruitment (women only; yes, no), and baseline self-reported physician-diagnosed history of diabetes mellitus (yes, no) and stroke (yes, no). Dialect group was a recruitment criterion. In addition, Cantonese men appeared to have lower hip fracture risk than the Hokkien men (p < 0.01), although there was no difference in hip fracture incidence in women regardless of dialect groups (p = 0.39). Hence, dialect group was also included as a covariate in our models. Adjustment for total energy intake in addition to the energy-adjusted dietary component (exposure of interest) is based on the multivariate nutrient density model.[35] We used the multivariate nutrient density model proposed by Willet and colleagues[35] to assess the diet-hip fracture associations in this study. The use of nutrient density, in which nutrient intake is divided by total energy intake, is an accepted measure for studying nutrient intake, and accounts for total energy intake in nutritional studies and epidemiologic analyses. As explained by Willet and colleagues,[35] the nutrient density method has several advantages: it can be calculated directly for an individual without the use of any statistical models; it is familiar to nutritionists as a measure of dietary composition; and it has been used in national dietary guidelines.

To study the BMI–hip fracture risk association and to assess the modifying effects of BMI on the diet–hip fracture risk associations, we only included participants with known body weight and height in these analyses. Hence, 52,792 cohort participants who reported body weight and height at baseline were included in the analysis including 1238 hip fracture cases in this subgroup. Subjects were stratified by BMI levels of <20, 20 to <25, and ≥25 kg/m2, which is based on the previous finding of BMI <20 kg/m2 being a risk factor for hip fracture from this cohort[26] and a meta-analysis of 12 population-based cohort studies[24]; and BMI ≥25 kg/m2 being the current cutoff for overweight proposed by the World Health Organization[36] which is applicable to Asian populations.[30] Heterogeneity of the diet–hip fracture risk associations among different BMI levels was tested using an interaction term (product between diet and BMI) included in the Cox model. Finally, we performed a sensitivity analysis for the diet–hip fracture risk association after excluding subjects who had reported extreme energy intakes (≤600 and ≥3000 kcal). All statistical analysis was conducted using SAS Version 9.2 (SAS Institute, Inc., Cary, NC, USA). All reported p values are two-sided. For the present analysis, we had an a priori hypothesis that total and specific carotenoids were protective against hip fracture risk. The p values were thus not adjusted for multiple testing; p < 0.05 was considered statistically significant.


Among the 1630 incident hip fracture cases, the time interval between cohort enrollment and hip fracture diagnosis was (mean ± SD) 9.9 ± 4.5 years. The mean age at fracture was 74.4 ± 7.5 years. Women accounted for 72.4% of all hip fractures and the age-standardized incidence rates of hip fractures in women (234/100,000 person-years) was twice that in men (123/100,000 person-years). In the whole cohort, compared to women, men were older in age, more educated, and had a greater proportion who were habitual smokers and alcohol drinkers. The mean BMI in the study population was 23 kg/m2. BMI levels were similar between men and women, as were physical activity and prevalence of diabetes mellitus and stroke (Table 1). Seventy-two percent of women were postmenopausal at enrollment, but only 3.7% of them reported use of hormone replacement therapy (Table 1). Regarding dietary intakes, according to the U.S. Recommended Daily Allowances (RDA) values,[37] both men and women in our cohort had adequate intake of protein. Intakes of vegetables were comparable between the two genders, although fruit intake was higher in men than in women. Overall, absolute dietary intake of total carotenoids for men was about 5% higher than women. Among all carotenoids that were examined, β-carotene had the highest correlation with total carotenoids, followed by lutein/zeaxanthin and lycopene. Among specific carotenoids, β-carotene was highly correlated with α-carotene and lutein. Total carotenoids were better correlated with vegetables than with fruits. The correlation between different types of vegetables and carotenoids reflect the dietary source of the carotenoids; green vegetables were best correlated with lutein/zeaxanthin, and yellow-orange vegetables with α-carotene (see Pearson's correlation coefficients between dietary intake of carotenoids and food groups/specific vegetable or fruit item in Supplemental Table 1).

Table 1. Baseline Characteristics of Men and Women Who Participated in the Singapore Chinese Health Study (n = 63,154)
CharacteristicsMen (n = 27,913)Women (n = 35,241)
  1. NA = not applicable.
Age at recruitment, years, mean ± SD59.6 ± 7.956.3 ± 8.0
Body mass index, mean ± SD23.0 ± 3.223.2 ± 3.3
<20 kg/m2(%)16.114.9
20–25 kg/m2(%)62.363.2
≥25 kg/m2(%)21.621.9
Level of education (%)
Secondary school or higher37.920.7
Smoking status (%)
Ever smokers58.18.8
Alcohol consumption (%)
Daily drinkers6.41.2
Physical activity (%)
0.5–3 hours per week15.412.7
≥4 hours per week9.37.4
Diabetes mellitus (%)8.79.2
Stroke (%)1.81.2
Postmenopausal (%)NA72.4
Hormone replacement use, yes (%)NA3.7
Dietary intake
Total energy intake (kcal), mean ± SD1749.7 ± 608.91399.3 ± 472.0
Total protein intake (g), mean ± SD65.2 ± 25.954.4 ± 21.5
Total vegetables (g/d), mean ± SD111.3 ± 65.6110.0 ± 61.9
Green vegetables (g/d), mean ± SD67.4 ± 41.666.3 ± 39.6
Yellow-orange vegetables (g/d), mean ± SD7.3 ± 8.48.5 ± 8.8
Cruciferous vegetables(g/d), mean ± SD44.1 ± 29.643.4 ± 28.3
Tomato products(g/d), mean ± SD7.7 ± 9.57.3 ± 9.1
Total fruit (g/d), mean ± SD212.7 ± 177.1194.4 ± 162.2
Total carotenoids (µg/d), mean ± SD5800.4 ± 3884.05541.6 ± 3495.8
α-carotene (µg/d), mean ± SD240.9 ± 275.8268.2 ± 287.9
β-carotene (µg/d), mean ± SD2125.6 ± 1514.32197.4 ± 1472.7
β-cryptoxanthin (µg/d), mean ± SD270.3 ± 343.0240.8 ± 327.6
Lycopene (µg/d), mean ± SD1307.0 ± 1707.6950.3 ± 1300.2
Lutein/zeaxanthin (µg/d), mean ± SD1856.6 ± 1143.01885.0 ± 1123.7

Table 2 shows the relations between dietary vegetables, fruits and carotenoids, and hip fracture risk for men and women separately. There was a statistically significant, dose-dependent inverse relationship with hip fracture risk for total, green or yellow-orange vegetables in men (p for trend ≤0.002), but not in women. Men in the highest quartile of total vegetables had approximately 40% lower in hip fracture risk than men in the lowest quartile. Consumption of tomato products or fruits was not associated with hip fracture risk in either gender. Similarly, statistically significant inverse associations between nutrient densities of total or specific carotenoids, including α-carotene, β-carotene, and lutein/zeaxanthin, and risk of hip fracture were seen in men (p for trend <0.05), but not in women. The strongest association was observed in dietary β-carotene. Compared with men in the lowest quartile, men in the third or fourth quartile of β-carotene experienced a statistically significant 37% decrease in risk of hip fracture (p ≤ 0.005). Nutrient density of β-cryptoxanthin or lycopene was not associated with hip fracture risk in either men or women. The results were essentially the same using gender-specific cut points for the definition of quartiles for the food and nutrient variables included in this study (data not shown). As expected, the consumption of total vegetables and carotenoids were highly correlated in our cohort, with a Pearson's correlation coefficient of 0.84. We hypothesized that the protective effects of vegetables were mediated via their carotenoids. When we put both intake levels of vegetables and specific carotenoid, such as β-carotene, in the same model, there was a loss of statistical significance due to the high collinearity of both variables. However, the association remained stronger for β-carotene (p for trend = 0.081) than for vegetables (p for trend = 0.27), supporting our hypothesis that the protective effect of vegetables was likely to be mediated by carotenoids such as β-carotene. Finally, we reanalyzed the data by treating food/nutrient densities as continuous variables, and the results were essentially the same. The p values for the association between dietary carotenoid and hip fracture risk were 0.0055 for men and 0.35 for women.

Table 2. Dietary Intake of Vegetables, Fruits, and Carotenoids in Relation to Hip Fracture Incidence for Men (n =27,913) and Women (n = 35,241)
 Men (quartile of energy-adjusted intake)Women (quartile of energy-adjusted intake)
  1. HRs were adjusted for age at recruitment (years), year of recruitment (1993–1995, 1995–1998), dialect group (Hokkien, Cantonese), body mass index (kg/m2), level of education in categories (no formal education, primary school, secondary school or higher), total energy intake (kcal/d), smoking status (never smokers, ex-smokers, current smokers), physical activity (none, 2–3 hours weekly, ≥4 hours weekly), calcium (quartiles, mg/1000 kcal/d), soy isoflavones (quartiles, mg/1000 kcal/d), vitamin B6 (quartiles, mg/1000 kcal/d), menopausal status (women only; yes, no), use of hormone replacement therapy at recruitment (women only; yes, no), and baseline physician-diagnosed history of diabetes (yes, no) and stroke (yes, no).
  2. HR = hazard ratios; CI = confidence interval.
Total vegetables
Cases1861218954 285294326275 
HR (95% CI)1.000.87 (0.69–1.10)0.74 (0.56–0.96)0.59 (0.42–0.81)0.00061.000.94 (0.80–1.11)1.04 (0.88–1.23)0.89 (0.74–1.07)0.424
Green vegetables
Cases1801218960 286289302303 
HR (95% CI)1.000.88 (0.70–1.12)0.76 (0.58–0.99)0.64 (0.47–0.88)0.0021.000.98 (0.83–1.16)1.04 (0.88–1.23)1.02 (0.85–1.21)0.712
Yellow-orange vegetables
Cases1701299952 281308300291 
HR (95% CI)1.000.93 (0.74–1.17)0.85 (0.66–1.09)0.56 (0.41–0.77)0.00081.000.96 (0.82–1.13)0.88 (0.74–1.03)0.90 (0.76–1.07)0.143
Tomato products
Cases13113111078 345291296248 
HR (95% CI)1.001.03 (0.80–1.31)1.01 (0.78–1.31)0.84 (0.63–1.12)0.3041.001.00 (0.85–1.17)1.01 (0.86–1.18)0.90 (0.76–1.06)0.272
Total fruits
Cases1301299893 389282251258 
HR (95% CI)1.001.10 (0.85–1.41)0.91 (0.69–1.22)0.96 (0.70–1.31)0.5331.000.94 (0.80–1.10)0.93 (0.79–1.11)0.97 (0.81–1.17)0.677
Total carotenoids
Cases1771229358 329310288253 
HR (95% CI)1.000.85 (0.67–1.08)0.80 (0.61–1.04)0.63 (0.45–0.88)0.0061.000.97 (0.83–1.14)0.98 (0.83–1.16)0.95 (0.79–1.15)0.642
Cases16611710760 273303309295 
HR (95% CI)1.000.83 (0.65–1.05)0.88 (0.68–1.12)0.61 (0.45–0.83)0.0051.000.91 (0.77–1.07)0.97 (0.82–1.15)0.97 (0.81–1.15)0.927
Cases1831347657 282319291288 
HR (95% CI)1.000.96 (0.76–1.20)0.63 (0.48–0.83)0.63 (0.46–0.87)0.00031.001.01 (0.86–1.19)0.96 (0.81–1.14)0.97 (0.81–1.17)0.616
Cases114123110103 376266231307 
HR (95% CI)1.001.12 (0.86–1.45)1.07 (0.81–1.41)0.95 (0.71–1.27)0.7161.000.99 (0.84–1.16)0.96 (0.80–1.14)1.18 (1.00–1.39)0.103
Cases1661329260 277285286332 
HR (95% CI)1.001.04 (0.82–1.32)0.87 (0.67–1.13)0.74 (0.54–1.02)0.0491.000.95 (0.80–1.12)0.95 (0.80–1.13)1.03 (0.87–1.23)0.673
Cases12412911661 457341212170 
HR (95% CI)1.001.15 (0.89–1.47)1.04 (0.80–1.35)0.91 (0.68–1.24)0.5381.001.08 (0.94–1.25)0.87 (0.73–1.02)0.88 (0.73–1.07)0.07

BMI was inversely related to risk of hip fracture in men (Table 3). Compared to men with BMI <20 kg/m2, men with BMI 20 to <25, and BMI ≥25 kg/m2 had hip fracture risk reduced by 18% and 33%, respectively (p for trend = 0.016). There was a 9% reduction of hip fracture risk for women with BMI ≥25 kg/m2 compared with women with BMI <20 kg/m2, but the difference was not statistically significant between two genders (p for interaction = 0.137) (Table 3). When BMI was analyzed on a continuous scale, for each unit increase in BMI, risk of hip fracture was reduced by 5.5% (95% CI, 0.92–0.98) in men, and only by 1.1% (95% CI, 0.97–1.01) in women. Further adjustment for dietary protein did not materially change the effect size of this BMI–hip fracture association.

Table 3. BMI in Relation to Hip Fracture Incidence in Men and Women
 Men (n = 23,918)Women (n = 28,874)
CasesHRa95% CICasesHRb95% CI
  1. Only participants with reported BMI (n = 52,792) are included in this table.
  2. aHRs were adjusted for age at recruitment (years), year of recruitment (1993–1995, 1995–1998), dialect group (Hokkien, Cantonese), level of education in categories (no formal education, primary, secondary or higher), smoking status (never smokers, ex-smokers, current smokers), total energy intake (kcal/d), physical activity (none, 2–3 hours weekly, ≥4 hours weekly), calcium (quartiles, mg/1000 kcal/d), soy isoflavones (quartiles, mg/1000 kcal/d), vitamin B6 (quartiles, mg/1000 kcal/d), and baseline physician-diagnosed history of diabetes (yes, no) and stroke (yes, no).
  3. bHR adjusted for parameters in footnote a, plus menopausal status (women only; yes, no) and use of hormone replacement therapy at recruitment (women only; yes, no).
  4. BMI = body mass index; HR = hazard ratio; CI, confidence interval.
BMI (kg/m2)      
<20851.00 1641.00 
20 to <252110.820.64–1.064680.930.78–1.11
p for trend 0.016  0.384 

Figure 1 shows the associations between total vegetables/carotenoids and hip fracture risk stratified by BMI for men. Among men with BMI <20 kg/m2, compared with the first quartile, HRs (95% CIs) of hip fracture for the second, third, and fourth quartile of total vegetable consumption were 0.60 (95% CI, 0.35–1.03), 0.49 (95% CI, 0.26–0.92), and 0.22 (95% CI, 0.08–0.64), respectively (p for trend = 0.0008). A null association was observed between total vegetables and hip fracture risk in men with BMI ≥25 kg/m2. The difference in the protective effects of total vegetables by BMI categories was statistically significant (p for interaction = 0.011). Similarly, the associations between total carotenoids and reduced hip fracture risk in men by BMI categories were also statistically different (p for interaction = 0.022). Using total vegetables/carotenoids as a continuous variable for the diet–hip fracture association in each BMI stratum yielded essentially the same results, p = 0.0002 and p = 0.0086 for total vegetables and total carotenoids, respectively, for men with BMI <20 kg/m2.

Figure 1.

Hazard ratio (95% CI) for hip fracture by quartiles (Q1–Q4) of dietary intake of total vegetables (A) or total carotenoids (B) in men (n = 23,918), according to levels of body mass index. (A) Total vegetables in relation to hip fracture risk. Test for trend, p = 0.004, 0.87, and 0.68 for BMI <20, BMI = 20 to <25, and BMI ≥25 kg/m2, respectively. (B) Total carotenoids in relation to hip fracture risk. Test for trend, p = 0.0008, 0.22, and 0.82 for BMI <20, BMI = 20 to <25, and BMI ≥25 kg/m2, respectively. Model was adjusted for age at recruitment, year of recruitment, dialect group, level of education in categories, smoking status, physical activity, total energy intake, calcium, soy isoflavones, vitamin B6, baseline physician-diagnosed history of diabetes and stroke. CI = confidence interval; BMI = body mass index.

For vegetable subtypes, the inverse association between dietary green or yellow-orange vegetables and hip fracture risk was present in men with BMI <20 kg/m2, but generally attenuated in men of the higher BMI categories. For specific carotenoids, α-carotene, β-carotene or lutein/zeaxanthin, the reduction in hip fracture risk was again the greatest in men with BMI <20 kg/m2 (Table 4). On the other hand, the associations between fruits, vegetables, and carotenoids (total and specific) and hip fracture risk were generally null across all strata of BMI among women (Supplemental Table 2). Further adjustment for dietary protein did not materially alter any results described (data not shown). In addition, we also conducted sensitivity analyses excluding those with extreme caloric intake, and the findings were materially unchanged (data not shown).

Table 4. Dietary Intake of Vegetables and Carotenoids in Relation to Hip Fracture Incidence for Men (n = 23,918), According to Levels of BMI
 BMI <20 kg/m220≤ BMI <25 kg/m2BMI ≥25 kg/m2
Cases (n = 85)HR95% CICases (n = 211)HR95% CICases (n = 70)HR95% CI
  1. HRs were adjusted for age at recruitment, year of recruitment, dialect group, level of education, smoking status, physical activity, total energy intake, calcium, soy isoflavones, vitamin B6, and baseline physician-diagnosed history of diabetes and stroke.
  2. BMI = body mass index; HR = hazard ratio; CI, confidence interval; Q1–Q4 = quartile 1 through quartile 4.
Green vegetables (g/1000 kcal/d)
Q1 (<28)451.00 761.00 161.00 
Q2 (28–40)270.970.59–1.59510.900.62–1.29251.720.91–3.26
Q4 (≥55)40.230.08–0.66380.910.60–1.39121.290.58–2.86
p for trend 0.0004  0.67  0.604 
p for interaction    0.004    
Yellow-orange vegetables (g/1000 kcal/d)
Q1(<1.8)421.00 651.00 281.00 
Q2 (1.8–3.7)230.770.46–1.30651.140.81–1.62190.760.42–1.36
Q3 (3.7–6.8)110.440.22–0.86501.060.72–1.54180.860.47–1.57
Q4 (≥6.8)90.490.23–1.04310.810.52–1.2550.270.10–0.70
p for trend 0.009  0.42  0.02 
p for interaction    0.752    
α-Carotene (µg/1000 kcal/d)
Q1 (<59.8)411.00 651.00 231.00 
Q2 (59.8–114.4)200.660.39–1.14570.960.67–1.34210.960.52–1.74
Q3 (114.4–212.3)160.650.36–1.18531.040.72–1.51160.810.42–1.57
Q4 (≥212.3)80.410.19–0.89360.870.57–1.34100.600.28–1.29
p for trend 0.02  0.70  0.18 
p for interaction    0.406    
β-Carotene (µg/1000 kcal/d)
Q1 (<850.4)481.00 691.00 241.00 
Q2 (850.4–1235.4)210.690.41–1.18661.200.85–1.70200.940.51–1.73
Q3 (1235.4–1772.4)110.420.21–0.84390.820.54–1.25160.920.47–1.80
Q4 (≥1772.4)50.270.10–0.70371.030.66–1.61100.680.31–1.53
p for trend 0.0008  0.68  0.41 
p for interaction    0.054    
Lutein/zeaxanthin (µg/1000 kcal/d)
Q1 (<781.8)411.00 671.00 181.00 
Q2 (781.8–1104.2)281.070.65–1.75621.220.85–1.74221.480.78–2.80
Q3 (1104.2–1544.1)100.460.22–0.93441.000.68–1.49171.320.66–2.63
Q4 (≥1544.1)60.400.16–0.97381.100.71–1.70131.380.64–3.01
p for trend 0.009  0.87  0.45 
p for interaction    0.015    
Lycopene (µg/1000 kcal/d)
Q1 (<191.9)241.00 591.00 111.00 
Q2 (191.9–451.9)291.530.87–2.67520.970.66–1.41231.810.87–3.75
Q3 (451.9–858.4)170.940.50–1.77621.110.77–1.61211.520.72–3.23
Q4 (≥858.4)151.280.65–2.56380.820.53–1.28151.260.54–2.91
p for trend 0.85  0.64  0.80 
p for interaction    0.844    


The present study shows an inverse association for hip fracture risk in men with dietary total vegetables, green vegetables, and yellow-orange vegetables as well as with total carotenoids, α-carotene, β-carotene, and lutein/zeaxanthin. In addition, BMI modified the protective effect of total vegetables/carotenoids against hip fracture risk, with an apparent protective effect for lean men (<20 kg/m2), but not for their heavier counterparts.

This study population had comparable vegetable intake with other Western populations,[17, 38] although intake of carotenoids was lower.[16] This is possibly due to relatively low prevalence of supplement use in this cohort (only about 5% of men and 8% of women reported use of oral supplements). More than three-quarters of total carotenoids in this population were sourced from dietary intake of total vegetables, and another 10% from fruit. Green vegetables provided 62% and 80% of β-carotene and lutein/zeaxanthin, respectively; 72% of α-carotene was provided by yellow-orange vegetables; fruits contributed 99% of β-cryptoxanthin; melons and tomato products contributed 47% and 29% of lycopene, respectively.

Although several cross-sectional studies have reported a positive association between vegetables and reduced hip fracture risk,[39-41] most of the results in prospective cohort or intervention studies failed to reach statistical significance.[38, 42, 43] A recent systematic review also concluded no clear beneficial effects of fruit and vegetables on bone health in postmenopausal women over 45 years of age.[44] This concurs with our null findings among women, for whom the protective effects of carotenoids or vegetables were weakened and became nonstatistically significant after further adjustment for other lifestyle and dietary covariates. This suggests that other lifestyle factors, particularly other dietary nutrients, may confound the association between carotenoids/vegetables and hip fracture risk for women. We have previously reported dietary isoflavones[26] and pyridoxine[45] as significant protective factors for hip fracture among women only. We postulate that the plausible mechanisms for these two nutrients may be related to their effects on estrogen, which plays a critical and dominant role in bone turnover in women.[46] Thereby, it is possible that the effect of oxidative stress on bone health may be less pronounced in postmenopausal women, which may explain the null association between carotenoids/vegetables and hip fracture risk for women. In the Framingham Osteoporosis Study, a statistically significant protective effect of total carotenoids on hip fracture risk was reported when men and women were combined in the analysis.[16] Nevertheless, our finding of an inverse association between vegetable/carotenoid and hip fracture risk in men is generally consistent with other findings from the Framingham study that also supported a stronger protective role of carotenoids in men. In the Framingham Study, dietary vegetables reduced longitudinal bone loss at the trochanter and Ward's area in men, but not in women.[41] The Framingham study also reported that more apparent protective effects from multiple carotenoids, including total carotenoids, β-carotene, lycopene, and lutein/zeaxanthin, were significantly related to trochanter BMD in men but less so in women.[17] Furthermore, testosterone may increase the bioavailability of carotenoids.[47] Our observed protective effect of carotenoids on hip fracture risk in men only could reflect the male hormonal effect on carotenoid uptake and biological effect.

We propose that the benefits of carotenoids on bone might be via their counteractive effects against oxidative stress, which has been shown to play a possible etiologic role in age-related bone loss and consequent osteoporotic fractures[12] by increasing osteoclastogenesis and stimulating osteoclastic differentiation[7, 8] through the receptor activator of NF-κB ligand (RANKL) expression and signaling mechanism. Besides the antioxidant function of carotenoids, other mechanisms of non-antioxidant functions can also be responsible for their roles in bone health. In vitro and animal studies have demonstrated that the carotenoid family may exert dual anticatabolic and proanabolic activities in bone by antagonizing the same NF-κB signal transduction pathway to suppress osteoclast differentiation and promote osteoblast mineralization.[48] Several in vitro studies have also reported that β-carotene, lycopene, and β-cryptoxanthin may have direct stimulatory effects in the proliferation, differentiation, and alkaline phosphatase activity of osteoblasts.[13, 49, 50]

In addition to the high carotenoid content in vegetables, another possible mechanism for the beneficial effects on bones includes the alkaline nature of vegetables. In particular, potassium in vegetables may neutralize excess metabolic acid to maintain an acid-base homeostasis[51, 52] and to promote calcium balance in bones.[53] However, in the current study, dietary potassium was not associated with hip fracture risk in either gender.[45] Further adjustment for potassium did not change the effect estimate for the relationship between carotenoids and hip fracture risk (data not shown). In our study, there was a null association between dietary vitamin C and hip fracture risk in both genders (data not shown). Although the Framingham Osteoporosis study showed that the use of supplementary vitamin C was associated with reduced fracture risk, there was no association with dietary vitamin C.[54] Furthermore, the National Health and Nutrition Examination Survey (NHANES) III study also showed no effects of dietary or serum ascorbic acid measures on bone mineral density and risk of self-reported fracture.[55]

The novel finding of the current study is that the reduction in hip fracture risk associated with vegetables and carotenoids was most pronounced in lean men with BMI <20 kg/m2. BMI is an established independent risk factor of osteoporosis from accumulative evidence,[24, 25] and appears to influence risk of hip fracture similarly in both genders.[24] The proposed mechanism of a higher BMI as a protective factor against osteoporotic fractures has been attributed to greater weight-bearing in the loading bone sites for both genders,[56] and increased endogenous estrogen produced in adipose tissues for postmenopausal women.[57] However, our findings suggest that BMI had a greater influence on hip fracture development in men than in women. Low BMI has been associated with increased oxidative damage indexed by 8-hydroxy-2′-deoxyguanosine,[27-29] particularly in men[28]; the molecule 8-hydroxy-2′-deoxyguanosine is also considered a reliable biomarker for the measurement of systemic oxidative stress.[58] Following this, we speculate that lean men possibly have higher oxidative stress in bones leading to hip fracture, and the antioxidant effects of carotenoids may counteract this mechanism of osteoporosis (for example, the RANKL pathway) related to leanness. Another possible explanation is that because adipose tissue is the major storage site for carotenoids,[59] the effects of insufficient carotenoid intake could be expected to be more apparent in lean subjects as a result of reduced storage and bioavailability.

The weaker association between BMI and hip fracture risk in women may be due to a possible nondifferential misclassification of BMI, because some of the postmenopausal women in our cohort may have experienced weight gain after recruitment during menopause.[60] In addition, it is plausible that the deleterious effect of a rapid decline of estrogen outweighs that of leanness related to oxidative stress on hip fracture development among postmenopausal women. If this is true, this is another possible explanation for the null association between vegetables/carotenoids and hip fracture risk in lean women <20 kg/m2. Taken together, our findings support the “estrogen-centric” role in postmenopausal osteoporosis and related fractures.[26, 61]

To our best knowledge, the current study is the first study to examine dietary carotenoids and hip fracture risk in an Asian population. The strength of this study is the large number of incident hip fracture cases identified from a population-based prospective cohort with a long follow-up time. Because supplementation was relatively uncommon in this study population, nutrient consumption computed from dietary intake using the food frequency questionnaire was considered valid and relevant. Another strength is the presumed lack of recall bias in exposure data because they were obtained prior to disease diagnosis. Singapore is a small city-state with a system for easy access to specialized medical care. Since practically all hip fracture cases will seek medical attention immediately and be hospitalized, our case ascertainment through linkage with the comprehensive, nationwide hospital database can be considered complete. Finally, we included all established and other possible risk factors for hip fracture as covariates in our regression-based risk models to minimize the likelihood of spurious associations resulting from insufficient control of confounding. Limitations of this study include the possible misclassification of dietary intake from the use of a food-frequency questionnaire. However, such misclassifications are nondifferential in nature, resulting in an underestimation (as opposed to an overestimation) of the true relative risk. Although the use of self-reported body weight and height could be prone to nondifferential misclassification and thus lead to underestimation of the BMI-hip fracture risk association, self-report of body weight has been shown to be highly valid across many populations[62] and specifically in Asians. A review of 64 studies suggested the difference between the self-reported and objectively measured BMI was generally null or slightly underestimated (less than 1 kg/m2) for those with BMI <30 kg/m2,[63] and 97% of our study population had BMI <30 kg/m2. Another limitation of this study is the lack of measurements on BMD at baseline, which may confound the relationship between BMI and risk of hip fracture, because BMD is closely related to body weight.[64] Finally, we did not include any biological measurement of oxidative stress in our study, and were therefore unable to confirm our hypothesis.

In conclusion, the findings of the present study suggest that dietary vegetables and carotenoids exert a strong protection against hip fracture, especially in lean elderly men. Our epidemiologic findings are supported by experimental data that have provided the mechanistic pathway in the role of carotenoids in regulating the biology of osteoblasts and osteoclasts. Future research in mechanistic and intervention studies is warranted to evaluate the effect of carotenoids in relation to osteoporotic fractures among elderly.


All authors state that they have no conflicts of interest.


This study was supported by the National Medical Research Council, Singapore (NMRC/EDG/0011/2007) and U.S. National Institutes of Health (NCI RO1 CA55069, R35 CA53890, R01 CA80205, and R01 CA144034). We thank Siew-Hong Low of the National University of Singapore for supervising the field work of the Singapore Chinese Health Study. We also thank the Ministry of Health in Singapore for assistance with the identification of hip fracture cases and mortality via database linkages. Finally, we acknowledge the founding, long-standing Principal Investigator of the Singapore Chinese Health Study: Mimi C Yu.

Authors' roles: Study design: ZD and WPK. Study conduct and data collection: WPK, LWA, and JMY. Data analysis: ZD and RWW. Data interpretation: ZD, YLL, JMY, and WPK. Drafting manuscript: ZD and WPK. Revising manuscript content and approving final version of manuscript: all authors. WPK takes responsibility for the integrity of the data analysis.