Osteoporosis is a systemic skeletal disease characterized by reduced bone mass and disrupted bone architecture, resulting in increased bone fragility and fracture risk( 1 ). Osteoporosis is commonly referred to as a ‘silent disease’ because it remains asymptomatic until a fracture occurs. Such osteoporotic fractures are a major cause of morbidity in the elderly population( Reference Plawecki and Chapman-Novakofski 2 ). Low bone mineral density (BMD; i.e. number of standard deviations below the mean areal BMD for young adults, T<−2·5) is marker of osteoporosis and a predictor of low trauma fractures( Reference Hernlund, Svedbom and Ivergård 3 ).
The accumulation and loss of bone mineral mass is influenced by various factors such as age, sex, ethnicity, heredity, lifestyle (physical activity and smoking) and nutritional status (Ca, protein and vitamin D intakes)( Reference Cooper, Westlake and Harvey 4 ). Of these, diet is one of the most important modifiable determinants( Reference Kitchin and Morgan 5 ). However, most studies have focused on a single nutrient or food/food group to examine the effects on bone health. These common approaches have methodological and conceptual limitations. They can detect the effects of single nutrients or foods on bone health but cannot explain the interactions among nutrients and food items( Reference Movassagh and Vatanparast 6 ). We have comprehensively explained different approaches in evaluating the association between nutrient patterns (NP) and bone health elsewhere( Reference Kontogianni, Melistas and Yannakoulia 7 ).
NP analysis has emerged as an alternative approach to overcome the aforementioned limitations in nutritional epidemiology( Reference Hu 8 ). In this approach, statistical methods are used to examine the pattern of intake of multiple foods or nutrients and derive single-exposure variables, or NP( Reference Newby and Tucker 9 ). Such NP may provide an improved and more generalizable insight into diet–disease relationships( Reference Hu 8 ). Using the NP approach could facilitate the development of public health recommendations that are more convenient to follow( Reference Slattery 10 ). Moreover, the human diet includes a range of different food items and complex mixtures of nutrients. Therefore, applying traditional methods (focusing on single nutrients) cannot take into consideration the numerous intercorrelations and interactions among foods and nutrients( Reference Hu 8 ).
Although previous studies have examined the association of NP with BMD, the findings are conflicting. Some studies suggest an inverse relationship between an ‘unhealthy’ NP (characterized by red and processed meat, fats and sweets) and BMD, and a positive association between a ‘healthy’ NP (characterized by high consumption of fish, olive oil, fruits and vegetables, and low consumption of red meat and candy) and BMD( Reference Tucker, Chen and Hannan 11 – Reference McNaughton, Wattanapenpaiboon and Wark 14 ). Others have failed to show such associations( Reference Kontogianni, Melistas and Yannakoulia 7 , Reference Langsetmo, Poliquin and Hanley 15 ). For example, in a sample of 220 adult Greek women, adherence to a Mediterranean dietary pattern was not associated with indices of bone mass( Reference Kontogianni, Melistas and Yannakoulia 7 ). Another study of 1928 men and 4611 women from the Canadian Multicentre Osteoporosis Study found no consistent relationship between diet and BMD( Reference Langsetmo, Poliquin and Hanley 15 ). Given the inconsistencies across existing studies, we conducted the current study to identify NP associated with high or low BMD at different parts of the body (total femur, femoral neck, intertrochanter, trochanter, Ward’s triangle, total spine, lumbar spine L1 to L4), as well as risk of fractures, in a nationally representative sample of US adults. We hypothesized that higher adherence to the Western diet would be associated with a less favourable index of bone health (BMD) and greater likelihood of fractures, while adherence to a healthy diet would be related with favourable bone health and lower risk of fractures.
Methods
Population
Data from the National Health and Nutrition Examination Surveys (NHANES) conducted between 2005 and 2010 were used for the present study. NHANES is a repeated cross-sectional survey conducted on an ongoing basis by the US National Center for Health Statistics, applying protocols and procedures described in detail previously( Reference Kassebaum, Kyu and Zoeckler 16 ). All methods were performed in compliance with the Declaration of Helsinki on ethical standards for research involving human subjects( Reference Kassebaum, Kyu and Zoeckler 16 ). NHANES uses a complex, multistage and stratified sampling design to select a sample representative of the civilian and non-institutionalized resident population of the USA. The National Center for Health Statistics Research Ethics Review Board approved the protocol and all participants provided informed consent( Reference Mazidi, Michos and Banach 17 ). Demographic information and interviews were collected using questionnaires administered during home visits, while trained personnel conducted a physical examination and biological sample collection in mobile examination units. The NHANES mobile examination centres are equipped with QDR 4500A fan-beam dual-energy X-ray absorptiometry densitometers (Hologic, Inc.) to measure BMD of the anterior–posterior lumbar spine and proximal femur. Details on measuring the BMD and protocols for the corresponding quality controls can be found elsewhere( Reference Cogswell, Looker and Pfeiffer 18 ). For fractures, we used data from self-reported hip, wrist (representing distal radius/ulna) and spine fractures( Reference Cogswell, Looker and Pfeiffer 18 ). A trained phlebotomist drew a blood specimen from the participant’s antecubital vein. Detailed information on the measurement of C-reactive protein (CRP) concentrations are available elsewhere( Reference Mazidi, Kengne and Mikhailidis 19 , Reference Mazidi, Gao and Vatanparast 20 ).
Dietary intake was assessed via 24 h recall by a trained interviewer, during the mobile examination centre visit, by use of a computer-assisted dietary interview system with standardized probes, i.e. the US Department of Agriculture Automated Multiple-Pass Method( Reference Ahluwalia, Dwyer and Terry 21 ). Briefly, the type and quantity of all foods and beverages consumed in a single 24 h period before the dietary interview (from midnight to midnight) were collected via the Automated Multiple-Pass Method, which is designed to enhance complete and accurate data collection while reducing the respondent burden( Reference Ahluwalia, Dwyer and Terry 21 ).
For the present analysis, three survey cycles (i.e. 2005–2006, 2007–2008 and 2009–2010) were combined to produce estimates with greater precision and smaller sampling error. The analytical sample was limited to adults aged ≥18 years. After excluding pregnant and lactating (n 795) respondents, as well as those with missing information on the variables of interest (n 1325), the final analytical sample included 18 318 respondents from NHANES 2005–2010.
Statistical analysis
We analysed the data in compliance with prescribed guidelines for analysis of the complex NHANES data set, taking account of the masked variance and utilizing the proposed weighting methodology( Reference Mazidi, Kengne and Mikhailidis 22 ). Factor analysis with orthogonal transformation (Varimax procedure) was applied to derive NP based on the nutrients. We used principal component factor analysis with Varimax orthogonal transformation to generate principal components representative of NP based on the highest correlation coefficients between the nutrients constituting each principal component( Reference Mazidi and Kengne 23 ). All the necessary prerequisites of principal component analysis, including linearity, Kaiser–Meyer–Olkin measure of 0·88 and significant Bartlett’s test of sphericity (P<0·001), were met. We then used regression methods to calculate the factor scores of each NP for each study participant( Reference Mazidi and Kengne 23 ). Factors were retained for further analysis based on their natural interpretation and eigenvalues on the scree test( Reference Stanhope 24 , Reference Khayyatzadeh, Moohebati and Mazidi 25 ). We computed the factor score for each NP by summing up intakes of nutrients weighted by their factor loadings (see online supplementary material, Supplemental Table 1). Each participant received a factor score for each identified pattern. Simple linear dose–response relationships are unlikely to be found in nutritional epidemiology( Reference Willett 26 ). To avoid issues with departure from a normal distribution and accordingly the distortion of regression coefficients by the extreme values, NP variables were categorized using population-specific quartiles of each NP before inclusion in regression models. The bottom or first quartile of each NP was then used as the reference category in all regression analyses. We computed means of BMD adjusted for age, sex, race/ethnicity, physical activity, smoking and CRP across the quartiles of each NP using ANCOVA. Categorical demographic variables were compared by using ANOVA and χ 2 tests, respectively. Fully adjusted multivariable linear regression models (adjusted for age, sex, race/ethnicity, physical activity, smoking and CRP) were used to determine the association of each NP score with BMD. Results were analysed using the complex sample module of the statistical software package IBM SPSS Statistics version 22.0. Sample weights were applied to account for unequal probabilities of selection, non-response bias and oversampling.
Results
The analytical sample comprised 18 318 participants, of whom 47·0 % (n 8607) were men. The mean age was 45·8 years in the overall sample and did not vary significantly between men and women (P=0·126). The White (non-Hispanic) population comprised the majority (69·4 %) of the participants. Furthermore, 56·1 % (n 8759) of the participants were married and 19·8 % were current smokers (23·9 % of the men and 16·7 % of the women). Overall, fewer participants engaged in a vigorous physical activity (5·2 %) than in little or no physical activity (24·0 %). As can be seen from Table 1, there were significant differences across quartiles of each NP with respect to race/ethnicity, education and sex distribution (P<0·001 for all comparisons). For each NP, participants in the upper higher quartile were younger than those in the bottom quartile (P<0·001 for all comparisons). Furthermore, participants in the upper quartile of the first NP had a higher level of CRP compared with those in the bottom quartile (0·45 v. 0·36 mg/dl, P<0·001), while participants in the upper quartile of the second and third NP had lower levels of CRP compared with those in the bottom quartile (both P<0·001; Table 1).
Table 1 Demographic characteristics of participants across quartiles of nutrient patterns (NP): US adults aged ≥18 years (n 18 318), National Health and Nutrition Examination Survey (NHANES) 2005–2010
Data are presented as percentages or as means with their standard errors where noted. ANOVA or the χ 2 test was applied.
The principal component method uncovered three NP altogether explaining 55·9 % of the variance in consumption of dietary nutrients. The first NP was essentially representative of a diet high in carbohydrates and sugar, total fat and saturated fat (called the ‘high-energy’ NP); the second NP was high in vitamins, minerals and fibre (‘nutrient-dense’ NP) and the third NP was mainly representative of high dietary PUFA and MUFA (‘healthy fat’ NP).
The age, sex, race/ethnicity, physical activity, smoking and CRP-adjusted means of BMD for different sites across quartiles of each NP are shown in Table 2. BMD at the total femur, femoral neck, trochanter and intertrochanter decreased significantly with increasing quartile of the high-energy NP (P<0·001 for all). BMD at the total femur, femoral neck and intertrochanter increased significantly across increasing quartiles of the nutrient-dense and healthy fat NP (all P<0·001). The profile of the associations was similar in age, sex, race/ethnicity, physical activity, smoking, CRP, BMI and hormone replacement therapy-adjusted linear regression models examining the continuous associations of NP with BMD. Indeed, there was a significant inverse association between the high-energy NP and total femur (β=−0·029), femoral neck (β=−0·025) and trochanter BMD (β=−0·034; all P<0·001). On the other hand, there was a significant positive association between the nutrient-dense and healthy fat NP and BMD at the total femur, femoral neck, trochanter and intertrochanter (all P<0·001; Table 2). No significant interactions were found between NP (all interaction P>0·153).
Table 2 Adjusted mean bone mineral density (BMD) across quartiles of nutrient patterns (NP) among US adults aged ≥18 years (n 18 318), National Health and Nutrition Examination Survey (NHANES) 2005–2010
Data are presented as means with their standard errors. ANCOVA was applied to calculate means adjusted for age, sex, race/ethnicity, physical activity, smoking, C-reactive protein, BMI and hormone replacement therapy. Adjusted linear regression was used to investigate the associations. Bold indicates significant differences across quartiles of a given NP.
Percentage of fractures in the hip, wrist and spine were 1·4, 8·9 and 2·3 %, respectively. The rate of fracture by quartile of each NP is shown in Table 1. In logistic regression models adjusted for age, sex, race/ethnicity, physical activity, smoking, CRP, BMI and hormone replacement therapy, there was no significant variation across quartile of NP in the odds of fractures in the hip, wrist and spine (Table 3).
Table 3 Adjusted odds from logistic regression for the association of nutrient patterns (NP) with prevalent fractures among US adults aged ≥18 years (n 18 318), National Health and Nutrition Examination Survey (NHANES) 2005–2010
Ref., reference category.
Data are presented as odds ratios and 95 % confidence intervals. Models were adjusted for age, sex, race/ethnicity, physical activity, smoking, C-reactive protein, BMI and hormone replacement therapy.
Discussion
Findings from the present study revealed that BMD at different sites of the proximal femur was inversely associated with a diet consisting highly of carbohydrates and sugar, total fat and saturated fat, and directly associated with a diet comprising vitamins, minerals and fibre. No association was found between NP and fractures, a finding to be interpreted in the context of a considerably low number of fractures.
In agreement with our results, several previous studies have demonstrated the direct association of fruits and vegetables (main sources of fibre, minerals and vitamins in the diet) with bone health( Reference Tucker, Chen and Hannan 11 , Reference Okubo, Sasaki and Horiguchi 12 , Reference McNaughton, Wattanapenpaiboon and Wark 14 , Reference Langsetmo, Poliquin and Hanley 15 ). The intake of fruits and vegetables in combination with fish was associated with high BMD in Japanese female farmers( Reference Okubo, Sasaki and Horiguchi 12 ). The existing Mediterranean diet score (MDS) was shown to be associated with high BMD( Reference Rivas, Romero and Mariscal-Arcas 27 ). Studies on the MDS and fracture risk have reported both unfavourable( Reference Feart, Lorrain and Coupez 28 ) and favourable( Reference Benetou, Orfanos and Pettersson-Kymmer 29 ) associations. The Mediterranean diet is known for its high content of cereals, legumes, fruits, nuts, vegetables, oils and fish, as well as low content of dairy and meat products. Although none of the patterns in our analyses of NHANES data exactly represents the Mediterranean diet, there are similarities with the nutrient-dense NP observed in our study, considering its high factor loadings for vitamins, minerals and fibre. Fruits and vegetables may act on bones by providing base buffers against excess dietary metabolic acids, increasing osteoclast activity that promotes bone resorption( Reference Arnett and Dempster 30 ) and balancing the pH of body fluids to the physiological range( Reference Bushinsky 31 ). In the Women’s Health Initiative study, a low-fat and high-fruit, -vegetable and -grain diet intervention slightly decreased the risk of multiple falls and marginally lowered hip BMD in postmenopausal women( Reference McTiernan, Wactawski-Wende and Wu 32 ). However, Pedone et al. reported no relationship between dietary acid load and BMD in elderly women( Reference Pedone, Napoli and Pozzilli 33 ). In a prospective study among 15 572 adults included in the China Health and Nutrition Survey, two dietary (traditional and modern) and two nutrient (plant- and animal-sourced) patterns were identified. During follow-up, participants in the upper tertile of the modern dietary and animal-sourced patterns had a 34 % (hazard ratio=1·34; 95 % CI 1·06, 1·71) and 37 % (hazard ratio=1·37; 95 % CI 1·08, 1·72) increase in fracture risk compared with those in the first tertile, respectively( Reference Mangano, Sahni and Kiel 34 ). In another study evaluating the association between NP and BMD among elderly Australians, animal- and plant-sourced patterns were not associated with BMD, while a mixed-source pattern had protective effect on BMD loss( Reference Melaku, Gill and Taylor 35 ).
After adjustment for CRP we found that there was still a significant relationship between BMD and NP, suggesting an independent role of diet on bone health. Studies have reported evidence on the role of inflammatory factors on bone remodelling. For example, in a study using NHANES III data (2807 females aged 65 years or older), there was an association between CRP and BMD( Reference Ganesan, Teklehaimanot and Tran 36 ). Moreover, in another prospective study (1872 community-dwelling women), fracture risk increased with increasing CRP levels( Reference Ishii, Cauley and Greendale 37 ).
Diets rich in minerals and vitamins are positively associated with BMD( Reference Whisner and Castillo 38 ). The nutrient-dense pattern in our study represented minerals and vitamins that are beneficial for the fundamental architecture of bone( Reference Nordin 39 , Reference Tang, Eslick and Nowson 40 ). The favourable effects of Ca and dairy food on bone health have been well established( Reference Nordin 39 , Reference Tang, Eslick and Nowson 40 ). Minerals found in fruits and vegetables might contribute to healthy bone remodelling via a favourable impact on osteoblastic and osteoclastic activity. K and Mg may contribute to the acid–base balance in the body, as discussed before, and prevent bone loss. K could also increase the retention of Ca in the kidneys, independently of its role in the alkaline state of the body. Mg is also necessary for Ca metabolism. Vitamin C may affect bone health through its antioxidant properties, which suppress osteoclast activity( Reference Gabbay, Bohren and Morello 41 ). It also acts as a cofactor for osteoblast differentiation and collagen formation( Reference Gabbay, Bohren and Morello 41 ). Carotenoids and other antioxidants also affect bone health by reducing oxidative stress( Reference Gabbay, Bohren and Morello 41 ). Vitamin K is involved in bone matrix formation, where mineralization happens( Reference Fusaro, Mereu and Aghi 42 ).
We identified a pattern characterized by high consumption of carbohydrates, sugar, total fat and saturated fat, the high-energy NP, which is similar to ones identified in other studies( Reference Tucker, Chen and Hannan 11 , Reference McNaughton, Wattanapenpaiboon and Wark 14 , Reference Langsetmo, Poliquin and Hanley 15 ). McNaughton and co-workers( Reference McNaughton, Wattanapenpaiboon and Wark 14 ) identified a ‘high-energy nutrient-poor pattern’ characterized by high intakes of refined cereals, soft drinks, fried potatoes, processed meat, beer, chocolate, confectionery and added sugar, and low consumption of vegetables, fruits and wholegrain cereals, which was significantly inversely associated with total body BMD. Similarly, for the ‘candy pattern’ observed by Tucker et al.( Reference Tucker, Chen and Hannan 11 ) in the Framingham Cohort Study, a diet rich in refined foods and lacking in nutrient-dense foods may be detrimental to bone health in men. Indeed, participants in the ‘candy’ cluster had a lower BMD in comparison with individuals in the ‘fruit, vegetables and cereal’ and ‘alcohol’ clusters( Reference Tucker, Chen and Hannan 11 ). Furthermore, a recent study from Brazil found that a ‘sweet foods, coffee and tea pattern’ was inversely correlated with BMD( Reference De França, Camargo and Lazaretti-Castro 43 ). Additionally, like the high-energy NP in our study, high intakes of fat and saturated fat showed a borderline inverse association with BMD among Japanese women( Reference Okubo, Sasaki and Horiguchi 12 ).
Fish and seafood are rich in PUFA, especially n-3 fatty acids, which are known to have an anti-inflammatory impact that benefits bone( Reference Mazidi, Shivappa and Wirth 44 ). We found the healthy fat NP to be associated with BMD in adults. A systematic review of ten randomized controlled trials revealed that n-3 and n-6 fatty acids when combined with Ca or dairy products had a significant impact on bone measures in some but not all trials( Reference Orchard, Pan and Cheek 45 ). The association between n-3 fatty acids and bone biomarkers and BMD could be explained by their anti-inflammatory effect, although more studies are needed to clarify the potential mechanisms.
There are several limitations to the present study. First, the results from our cross-sectional study, although nationally representative, cannot demonstrate a causal relationship. Second, although our analysis included known potential variables that can affect bone health in terms of environmental and genetic factors, residual confounding variables may still exist. We used a wide age range in adulthood that might not be the best approach in evaluating the association between NP and BMD. In an ideal situation, when sample size allows, women after menopause should be studied separately controlling for other potential covariates such as hormone replacement therapy. The statistical power was low to reliably investigate the association between NP and fracture. Moreover, fracture cases were assessed based on self-reports.
The study has several strengths. We had a sample selected randomly from the general US population. Therefore, the results obtained from this nationally representative sample can be extrapolated to the entire population. In addition, a large number of participants aged 18–80 years, the use of standardized procedures and the inclusion of both men and women are other important strengths.
A healthy nutrient-dense NP, characterized by high intakes of minerals, vitamins and fibre, may benefit BMD independent of potential confounding factors. In contrast, adherence to a high-energy NP characterized by high consumption of total and saturated fats, carbohydrates and sugar may pose a risk for low BMD.
Acknowledgements
Financial support: This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. M.M. was supported by a TWAS studentship of the Chinese Academy of Sciences. Conflict of interest: None. Authorship: M.M. analysed and interpreted the data and drafted the manuscript. A.P.K. conceived of and designed the study, and carried out critical revision of the manuscript. H.V. conceived of and designed the study, and carried out critical revision of the manuscript. Ethics of human subject participation: NHANES was performed in compliance with the Declaration of Helsinki and all protocols involving human subjects were approved by the National Center for Health Statistics Research Ethics Review Board. Informed consent was obtained from all participants.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1368980018000939
