The parent compound for n-3 polyunsaturated fatty acids (PUFAs), α-linolenic acid (18:3, ALA), and n-6 PUFAs, linoleic acid (18:2, LA), competitively participate in a series of elongation and desaturation reactions, which yield the long-chain n-3 fatty acids, eicosapentaenoic acid (20:5, EPA) and docosahexaenoic acid (22:6, DHA), and the long-chain n-6 fatty acid, arachidonic acid (20:4, AA). These long-chain fatty acids serve as precursors in the production of eicosanoids and cytokines involved in bone modeling and remodeling, including prostaglandin E2 (PGE2) derived from AA (reviewed in Refs.1–6). This suggests that fat intake may be related to skeletal health. In fact, n-3 and n-6 fatty acids and their derivatives may influence bone through alternative pathways, including effects on peroxisome proliferator-activator receptor (PPAR) activation7–9 and the production of lipid mediators involved in resolving inflammation.10–12 N-3 fatty acids may also enhance osteoblast function,13 as well as enhance calcium transport14, 15 and reduce urinary calcium excretion.15–17
Protective effects of n-3 fatty acid intake and supplementation on bone mass,18–20 bone formation markers,21 and bone resorption markers,22 as well as lower (n-6):(n-3) ratio, have been observed in human studies. Furthermore, beneficial effects of ALA on bone resorption have been reported.23 Fish intake was recently shown to be associated with less postflight bone mineral density (BMD) loss in astronauts,24 and in young men (<18 years old), serum DHA concentration was positively associated with total body and spine BMD.25 Previous studies of fatty acids and fracture risk have been conflicting. In a recent case-control study of elderly patients, n-6 fatty acid intake was significantly associated with increased risk of low-energy fracture, but no relationship between n-3 fatty acids and fracture was observed.26 Fish or EPA + DHA intakes were not associated with hip fracture incidence in over 5000 older adult men and women.27 However, recently, total n-6 fatty acid intake was inversely associated with total fracture risk, and EPA + DHA intake was positively associated with total fracture risk among over 135,000 postmenopausal women enrolled in the Women's Health Initiative (WHI).28
These previous studies conducted in humans on the effects of PUFAs on BMD or fractures have been restricted to assessing dietary intake of fatty acids as the primary exposure of interest, which may not be reflective of actual fatty acid status. Thus, we evaluated the relationships between plasma phosphatidylcholine (PC) essential fatty acids (DHA, LA, and AA) and BMD of the femoral neck (FN) and hip fracture risk over 17-years of follow-up among older adult women and men in the Framingham Osteoporosis Study.
Materials and Methods
Subjects were drawn from the Framingham Heart Study (FHS), which began in 1948 with the goal of identifying risk factors for cardiovascular disease. At its inception, 5209 women and men, aged 28 to 62 years, were recruited from a random sample of two-thirds of the households in Framingham, MA, USA, to participate in the initial examination and were examined biennially thereafter for over 50 years.29 In 1988, the Framingham Osteoporosis Study was initiated as an ancillary study to the FHS and 1164 of the 1402 surviving members of the original cohort had BMD measurements taken at the 20th biennial examination (1988–1989). At this exam, 829 of the subjects with FN-BMD measurements also had measures of plasma PC fatty acids. We excluded subjects with a hip fracture prior to this time (n = 21), or with incident hip fracture due to trauma (n = 1). We further excluded subjects with missing covariate information for body mass index (BMI) (n = 23), and education status (n = 19). After exclusions, the final sample size (n = 765) for cross-sectional analyses of FN-BMD was comprised of 281 men and 484 women. These subjects were also followed for incident hip fracture from the date they had blood measures taken at exam 20 until December 31, 2005. The final sample size (n = 556) for longitudinal analyses of 4-year change in FN-BMD was comprised of 193 men and 363 women. This study was approved by the Institutional Review Boards for Human Research at Boston University, Hebrew Rehabilitation Center, and Tufts University.
Plasma PC fatty acid measurements
Plasma measures were derived from the PC fraction of plasma lipids. Plasma phospholipid measures of DHA have been shown to correlate (r = 0.42) with dietary DHA measures in cross-sectional data.30 The phospholipid fraction in serum has also been demonstrated to respond to dietary LA intervention over a 3-year time period in elderly men.31 The laboratory techniques that were used to derive and measure plasma fatty acids at exam 20 of the Framingham Heart Study have been described in detail elsewhere.32–35 Briefly, lipids were extracted from stored blood samples and phospholipids were subsequently isolated and separated by high-performance liquid chromatography (HPLC). The phosphatidylcholine fraction was then re-extracted and methylated and individual fatty acids were separated and quantified by gas chromatography. Plasma PC fatty acid concentrations were expressed as the percent of total fatty acids for DHA, LA, and AA. EPA was not analyzed because EPA concentrations for the majority of subjects (∼75%) were equal to zero.
FN-BMD (g/cm2) of the right proximal hip was measured at baseline (exam 20) using a Lunar dual-photon absorptiometer (DP3; Lunar Corporation, Madison, WI, USA) and again at 4-year follow-up (exam 22) using dual X-ray absorptiometry (DPX-L; Lunar Corporation), as described.36, 37 Cross-calibration performed between the two machines showed a small, consistent shift in hip BMD values and, thus, published correction equations were used to adjust exam 20 BMD values to make them more comparable to exam 22 values.38 For longitudinal analyses, 4-year change in BMD was calculated as BMD at exam 22–BMD at exam 20.
Assessment of hip fracture
Hip fractures were reported by participants at each biennial exam upon interview, beginning at exam 18 (1984). For those unable to attend exams, hip fractures were reported by telephone interview. Occurrence of hip fractures were further identified through systematic review of medical records of hospitalizations and deaths, and were confirmed by reviewing medical records and radiographic and operative reports, as described.39 Incident hip fracture was defined as first time fracture of the proximal femur that occurred following the date of the blood measures through the follow-up period until December 31, 2005.
SAS statistical software (version 9.1; SAS Institute, Cary, NC, USA) was used to perform all statistical analyses. Two-sided p < 0.05 was considered statistically significant for all analyses. For BMD analyses, we tested for effect modification by sex by first including an interaction term in analyses on the combined sample. Due to an interaction between sex and plasma PC DHA concentrations (p = 0.05) in longitudinal FN-BMD analyses, measures of baseline FN-BMD, and 4-year change in FN-BMD were regressed onto plasma PC fatty acid concentrations (DHA, LA, and AA) modeled both continuously and using quintiles, separately for women and men using the general linear models (GLM) procedure in SAS, adjusting for multiple comparisons with Dunnett's adjustment. In addition to obtaining BMD least squares means by quintile of plasma PC fatty acid concentration, we also performed a test for linear trend across the quintiles by creating a continuous variable set equal to the median plasma PC fatty acid value for each quintile and regressing BMD measures on this continuous variable.
For hip fracture risk analyses, we also tested for effect modification by sex. There were no interactions between sex and plasma PC fatty acid concentrations, thus Cox proportional-hazards regression was used to estimate hazard ratios (HR) and 95% confidence intervals (CIs) continuously and categorically by quintile of the plasma PC fatty acid exposure variables in the combined sample. Tests for linear trend across quintiles were also performed. The proportional hazards assumption was satisfied for each plasma PC fatty acid exposure variable and covariate as deemed evident by p > 0.05 for all time-dependent covariates tested.
Differences in potential confounding variables measured at baseline were also compared across quintiles of plasma PC fatty acid concentration in women and men separately. These variables included BMI (weight/height2, in kg/m2), physical activity measured by the Framingham physical activity index (PAI), total energy intake (kcal/d), protein intake (g/d), total (dietary plus supplemental) calcium intake (mg/d), and total (dietary plus supplemental) vitamin D intake (IU/d). Measures of BMI were available for the full sample (n = 484 women and n = 281 men), whereas measures of physical activity were available for nearly the full sample (n = 479 women and n = 280 men). Dietary measures, which were assessed at baseline by Willett food frequency questionnaire,40 were only available on a subset of the sample (n = 424 women and n = 247 men). BMI was the only potential confounding variable found to be significantly associated with any of the plasma PC fatty acids (LA). Thus, models were first adjusted for a standard set of covariates (age, education status, estrogen use in women, and baseline FN-BMD for longitudinal BMD analyses), and then additionally adjusted for BMI. In the combined sample of women and men in the analysis of hip fracture, sex and estrogen use were adjusted by creating a categorical variable (0, 1, and 2), such that: 0 = men (referent group); 1 = never or former estrogen using women; and 2 = current estrogen using women.
PC concentrations of LA and AA were normally distributed, while PC DHA was positively skewed for both women and men. A square root transformation of PC DHA improved the normality and homoscedasticity to a greater extent than a natural log transformation, thus we applied a square root transformation to PC DHA prior to performing regression models in continuous analyses. Results from continuous and categorical analyses were similar, thus for ease of interpretation, we present only results from the categorical analyses.
Mean values (± SDs) and percentages for subject characteristics at baseline (exam 20) are presented in Table 1. The mean age for women and men was ∼75 years. Mean BMI values were slightly lower in women than in men (26.5 ± 5.0 versus 27.2 ± 4.02). Proportionately fewer women than men were past smokers (45.0% versus 63.1%), but the proportion of current smokers did not appear different (10.8% versus 9.7%). Few women reported current estrogen use (4.13%). More than two-thirds of both women and men were at least high school graduates. Mean plasma PC fatty acid concentrations, expressed as a percent of total fatty acids, were slightly higher in women than in men for DHA (3.68% ± 1.17% versus 3.44% ± 1.07%), but lower in women than in men for LA (24.6% ± 3.0% versus 25.1% ± 3.1%). Mean baseline FN-BMD was also lower in women than in men (0.722 ± 0.111 g/cm2 versus 0.874 ± 0.137 g/cm2) and women, on average, lost more FN-BMD over 4 years than men (–0.024 ± 0.054 g/cm2 versus –0.019 ± 0.071 g/cm2) (data not shown). The mean follow-up period and SEM for analyses of hip fracture risk for women and men were 11.6 ± 4.94 years (4230 ± 1808 person-years) and 9.98 ± 5.41 years (3646 ± 1976 person-years), respectively. Women sustained 75 hip fractures during the follow-up period, while men sustained 17 hip fractures, corresponding to a crude incidence rate of 17.7 and 4.66 hip fractures per 1000 person-years, respectively (data not shown).
Table 1. Subject Characteristics at Baseline Exam 20
BMI = body mass index (weight/height; kg/m2); NA = not applicable; PC = phosphatidylcholine; DHA = docosahexaenoic acid; FA = fatty acid; LA = linoleic acid; AA = arachidonic acid; FN-BMD = femoral neck bone mineral density.
Based on subjects with complete covariate and FN-BMD measurements at baseline exam 20, unless otherwise indicated.
No associations between potential confounding variables and the PC fatty acid measures were observed, with the exception of an association between BMI and plasma PC LA concentrations (Table 2). Subjects in the highest quintile of plasma LA had significantly lower BMI than those in the lowest quintile in both women and men, as well as in the combined sample (data not shown).
Table 2. Baseline Differences in BMI According to Quintile of Plasma PC Fatty Acid Concentration in Women and Men
Mean ± SE
Mean ± SE
Mean ± SE
Sample size varied according to variable availability: n = 484 for women and n = 281 for men. Q1 of plasma PC fatty acid concentration is the reference group for each potential confounding variable.
BMI = body mass index (weight/height; kg/m2); PC = phosphatidylcholine; DHA = docosahexaenoic acid; LA = linoleic acid; AA = arachidonic acid.
26.4 ± 0.50
27.5 ± 0.51
26.2 ± 0.51
26.9 ± 0.51
26.4 ± 0.50
26.8 ± 0.51
27.1 ± 0.51
26.9 ± 0.50
26.5 ± 0.51
26.4 ± 0.50
26.0 ± 0.50
26.4 ± 0.51
25.8 ± 0.85
25.5 ± 0.51
26.6 ± 0.51
26.5 ± 0.53
28.6 ± 0.51
26.6 ± 0.53
26.9 ± 0.54
28.3 ± 0.51
27.7 ± 0.54
27.8 ± 0.54
27.1 ± 0.51
26.7 ± 0.54
27.1 ± 0.53
27.4 ± 0.51
27.7 ± 0.53
28.0 ± 0.54
24.8 ± 0.51
27.5 ± 0.53
Associations between plasma PC essential fatty acids and BMD
Cross-sectionally, there were no significant associations between plasma PC DHA or plasma PC AA and FN-BMD among women (Table 3). Women with the highest plasma PC LA concentrations had lower mean FN-BMD than women with the lowest concentrations in the model adjusted only for age, education, and estrogen use (Q5 versus Q1: p = 0.03; p-trend = 0.02), but this association was no longer significant after adjustment for BMI (Q5 versus Q1: p = 0.20; p-trend = 0.12). Among men, those with higher plasma PC AA tended to have higher mean FN-BMD than those with the lowest PC AA (p-trend = 0.06). There were no significant cross-sectional associations between plasma PC DHA or PC LA and FN-BMD in men.
Table 3. Cross-Sectional FN-BMD ± SE According to Quintile of Plasma PC Fatty Acid Concentrations in Women and Men
Quintiles of BMD (g/cm2)
Model 1: adjusted for age, education, and estrogen use in women; Model 2: adjusted for age, education, estrogen use in women, and BMI; n = 484 women and n = 281 men.
FN-BMD = femoral neck bone mineral density; PC = phosphatidylcholine; DHA = docosahexaenoic acid; LA = linoleic acid; AA = arachidonic acid.
Longitudinally, we observed an interaction between sex and plasma PC DHA (p = 0.05) in relation to 4-year change in BMD. Unexpectedly, greater concentrations of PC DHA were associated with more bone loss of over 4-years (Q5 versus Q1: p = 0.09, p-trend = 0.04) (Fig. 1). In contrast, men with the highest PC DHA maintained FN-BMD over 4 years relative to men with the lower PC DHA, who lost FN-BMD, but this association did not reach significance (p-trend = 0.15). Because the relationship between PC DHA and FN-BMD change did not appear linear in men, we examined the possibility of a threshold effect by combining Q1–Q4 in a post hoc analysis (Table 4). Men in the highest quintile of PC DHA maintained FN-BMD over 4 years relative to men in the lower four quintiles, who lost FN-BMD (p = 0.01). This threshold effect was not observed in women (Q1–Q4 versus Q5: p = 0.22). There were no significant longitudinal associations between either plasma PC LA or plasma PC AA and 4-year change in FN-BMD in women or men (data not shown).
Table 4. Threshold Effects of Plasma PC DHA on Mean 4-Year Change in FN-BMD ± SE
Adjusted for age, education, estrogen use in women, BMI, and baseline FN-BMD.
PC DHA = phosphatidylcholine docosahexaenoic acid; FN-BMD = femoral neck bone mineral density; BMI = body mass index (weight/height; kg/m2).
Mean ± SD.
Women (n = 363)
3.23 ± 0.69
−0.022 ± 0.003
5.57 ± 0.99
−0.031 ± 0.006
Men (n = 193)
3.18 ± 0.70
−0.025 ± 0.006
5.17 ± 0.76
0.007 ± 0.012
Associations between plasma PC essential fatty acids and hip fracture
Analyses conducted in the combined sample of men and women are presented (Fig. 2). No significant associations were observed between plasma PC DHA and hip fracture in the combined sample (p-trend = 0.60). Subjects in the highest quintile of plasma PC LA had over twice the risk of hip fracture than subjects in the lowest quintile (Q5 versus Q1: HR = 2.17, 95% CI = 1.10–4.28, p-trend = 0.05) in the model adjusted only for age, education, and estrogen use in women (data not shown). However, additional adjustment for BMI attenuated the magnitude and significance of this association (Q5 versus Q1: HR = 1.88, 95% CI = 0.94–3.76, p-trend = 0.14). High plasma PC AA was associated with a reduced risk of hip fracture (p-trend = 0.03) in the full model adjusted for age, education, estrogen use in women, and BMI. Subjects in the highest quintile of plasma PC AA had a 51% lower risk of hip fracture than subjects in the lowest quintile (Q5 versus Q1: HR = 0.49, 95% CI = 0.24–1.00).
To our knowledge, this is the first investigation of the association between plasma concentrations of PUFAs, BMD, and incident hip fracture. Men with the highest concentrations of PC DHA had higher BMD relative to men in the lower four quintiles in post hoc analyses, but in women we observed that higher concentrations of PC DHA were associated with greater amounts of bone loss over 4 years. This finding is difficult to explain, as baseline measures of BMI, physical activity, and potential dietary confounding variables (energy intake, protein intake, dietary and supplemental calcium) were not different between quartiles of PC DHA in women. Although we are unaware of any other studies reporting this association between plasma concentrations of DHA and bone density measures, we previously observed significant interactions between dietary EPA + DHA and AA in relation to FN-BMD measures in both women and men, such that such that AA was positively associated with BMD in women with high EPA + DHA intakes (but not in women with low intakes), and AA was negatively associated with BMD in men with the lowest intake of EPA + DHA (but EPA + DHA appeared protective in men at each level of AA intake).41 One possible explanation for these findings may be that AA served as a protective factor against high EPA + DHA intakes in the women. Interestingly, an unexpected detrimental effect of dietary long-chain n-3 fatty acids in relation to total fracture risk was observed in the WHI.28 It is curious that these negative effects have been observed in two populations of postmenopausal women. Other than a spurious association, one explanation for the detrimental effect of PC DHA may be related to lipid peroxidation. Postmenopausal women have been noted to have higher levels of lipid peroxidation than premenopausal women42 and estrogen is recognized to have both antioxidant and pro-oxidant actions.43 The highly unsaturated hydrocarbon chain of DHA may increase the susceptibility of this fatty acid to lipid peroxidation in postmenopausal women experiencing decreased concentrations of circulating estrogens. In one animal study, a DHA-enhanced diet resulted in significantly higher phospholipid peroxidation in plasma and liver of male rats than rats fed a control diet.44 In another study, aged (100-week-old) female Wistar rats fed a DHA-supplemented diet had significantly higher concentrations of a marker of oxidative DNA damage (8-hydroxydeoxyguanosine) in bone marrow than did young rats (aged 10 weeks) or controls.45 Oxidized lipids have recently been shown to enhance receptor activator of NF-κB ligand (RANKL) production,46 which may affect osteoclastogenesis and bone resorption.
We further observed a detrimental association between plasma PC LA concentrations and baseline FN-BMD in women, as well as an increased risk of hip fracture in persons with the highest plasma PC LA in the combined sample of women and men. Adjustment for BMI attenuated these associations. Detrimental associations between dietary LA intakes and measures of BMD or fracture have been observed in other studies. Weiss and colleagues19 observed that a higher ratio of total dietary n-6:n-3 fatty acids and of LA:ALA was associated with lower hip BMD among 1532 women and men enrolled in the Rancho Bernardo Study, and increased intake of total PUFA was also found to be associated with bone loss at the FN among 891 women.47 A case-control study of hospitalized elderly patients (n = 334) found a significantly increased risk of low-energy osteoporotic fracture among subjects in the highest quartile of n-6 intake, relative to those in the lowest quartile.26 In contrast, recent findings from the WHI showed that total dietary n-6 fatty acid intake (LA comprised the most abundant n-6 fatty acid) was associated with a slightly decreased risk of total fracture, although no association was observed with hip fracture.28 Each of the aforementioned studies investigating dietary LA or PUFA intake accounted for energy intake, weight, or BMI through varied model specifications. In our analyses of plasma PC LA concentrations, BMI confounded the detrimental associations we observed; BMI was inversely associated with PC LA concentrations in both women and men, and in the full models, was associated with baseline FN-BMD in women (p < 0.01, data not shown) and with reduced risk of hip fracture in the combined sample of women and men (HR = 0.94, 95% CI = 0.89–0.99, data not shown). LA is the predominant PUFA in typical Western diets, due to its presence in several vegetable oils, including safflower, sunflower, soybean, and corn oil, in addition to its presence in foods made with these oils. It may be speculated that women and men in our study with the highest plasma PC LA concentrations are healthier in general, and engage in heart-healthy behaviors such as consuming PUFA-rich vegetable oils and maintaining lower BMI.
Still, LA may have weak effects on BMD, independent of a relationship with BMI or weight. In a previous study by our group, women with the highest intakes of LA tended to lose more FN-BMD over 4 years than those with the lower intakes (p-trend = 0.06), after adjustment for energy, BMI, and height.41 The hypothesis that AA derived from LA may exert negative effects on bone by increasing the production of proinflammatory eicosanoids and cytokines, and by interfering with n-3 fatty acid metabolism and subsequent production of less or anti-inflammatory eicosanoids and cytokines, has been described elsewhere.5, 6 Watkins and colleagues1, 2 further described the dual role of the eicosanoid PGE2, a derivative of AA, as a potent stimulator of bone resorption, and stimulator of bone formation at low concentrations. The biosynthesis of AA from LA has been shown by some to be limited,48 thus it should be considered whether LA itself may have effects on bone, in addition to AA production or competition with n-3 fatty acid metabolism. LA has been shown to activate NF-κB in endothelial cells,49, 50 which may suggest a role of LA in osteoclast formation if this finding can be replicated in bone cell lines.
We further observed a protective association between plasma PC AA and baseline FN-BMD in men, in addition to reduced risk of hip fracture observed with high concentrations of plasma PC AA in the combined sample of women and men. Dietary supplementation with AA has been shown to elevate bone mass in piglets,51, 52 and maternal umbilical cord AA has been related to bone mass in healthy full-term infants in humans.53 No studies, to our knowledge, have examined associations between plasma AA and either bone mass or fracture risk in adult humans. However, we previously observed protective effects of dietary AA in relation to hip fracture in men,54 and interactive effects between dietary AA and EPA + DHA in relation to FN-BMD measures in men and women.41 We noted that the protective effects of AA may be due, in part, to protein intake, because AA is found in protein-rich animal food sources, including eggs and meats. Energy-adjusted intakes of AA and protein were highly correlated among the men in our previous study (r = 0.62), but in this study of plasma PC AA, there were no significant differences in protein intake between quintiles of plasma PC AA in either women or men.
Other mechanisms may be responsible for protective effects of AA on bone. For example, AA and its derivatives, in addition to other essential fatty acids, have also been demonstrated to serve as ligands for both PPAR-α and PPAR-γ.7 Activation of PPAR-α has been found to upregulate nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, alpha (IκBα), a protein that inhibits the function of NF-κB by retaining it in the cytoplasm in an inactive form.8, 55 PPAR-γ activation has also been shown to inhibit tumor necrosis factor (TNF)-α, interleukin (IL)-1, and IL-6 production56, 57 and 12d–prostaglandin J2 (delta 12-PGJ2) (a derivative of AA) has been shown to stimulate collagen synthesis in human osteoblast cells.58, 59 Pro-resolving lipid mediators biosynthesized from n-6 AA, termed lipoxins (LXA4 and LXB4), may be involved in the downregulation of inflammatory cytokine gene regulation (reviewed in Serhan and colleagues60). Aspirin has been identified to trigger lipoxin synthesis from AA by interacting with COX-2, which shifts the enzyme's activity away from generating proinflammatory eicosanoids in favor of generating aspirin-triggered 15-epi-lipoxin A4 (ATL).61 Overexpression of 15-LOX, which redirects PGE2 from leukotriene synthesis to LXA4 synthesis, in addition to topical gingival application of an ATL analog, reduced bone loss in rabbits with periodontitis.62
Our study has some limitations. The availability of plasma PC fatty acid measures at only baseline limits our ability to assess how changes to plasma PC fatty acid concentrations over time may affect bone outcomes. We also were not able to adjust for use of nonsteroidal anti-inflammatory drugs, such as aspirin, which may be involved in generating the production of pro-resolving lipid mediators, lipoxins and resolvins.11, 63 Nonetheless, plasma measures of n-3 and n-6 essential fatty acids are generally considered good biomarkers of dietary intake because these fatty acids cannot be synthesized in the body from carbohydrate.64 We limited our analyses to FN-BMD and hip fracture because of all osteoporotic fractures, hip fractures in particular are associated with significant morbidity and mortality,65, 66 and in our sample are fully adjudicated using medical records in the vast majority of cases. However, our findings should be replicated at other bone and fracture sites.
Taken together, these results suggest protective effects of plasma PC AA on FN-BMD in men, in addition to protective effects of plasma PC AA on hip fracture risk in women and men. Our results also suggest that detrimental effects of plasma PC LA on BMD or hip fracture risk may be confounded by BMI. The differential effect of plasma PC DHA on BMD in elderly women and men observed in our study needs further clarification. If these results are confirmed in other studies, examination of interactive effects between DHA and antioxidant nutrients, such as vitamins C and E, may be warranted. Future investigations on relationships between dietary or plasma fatty acids and BMD or fracture would benefit from adjusting for confounding effects of energy intake, BMI, weight, or weight change, with prudence.
All authors state that they have no conflicts of interest.
This study was supported by a grant from the National Institute of Arthritis Musculoskeletal and Skin Diseases and the National Institute on Aging R01 AR/AG 41398 and the National Heart, Lung, and Blood Institute (NHLBI) Framingham Heart Study (National Institutes of Health/NHLBI contract N01-HC-25195, Bethesda, MD), Framingham, MA, USA.
Authors' roles: The authors' responsibilities were as follows: EKF, KLT, and DPK designed the study; EKF analyzed the data, composed the manuscript, and was responsible for the final content; KLT provided study oversight; KLT, DPK, RR, EJS, and LAC provided essential materials; and KLT, DPK, RR, and EJS provided critical revision of the manuscript.