Serum Retinoids and β-Carotene as Predictors of Hip and Other Fractures in Elderly Women


  • The authors have no conflict of interest.


There is debate about the possible deleterious effect of excessive vitamin A exposure on fracture risk. In this nested case control study in older women (312 cases and 934 controls), serum retinol, retinyl palmitate, and β-carotene were not associated with fracture risk, and there was no evidence of excess risk with multivitamin or cod liver oil supplementation.

Introduction: Recent studies have suggested that higher vitamin A intake may account for a component of fracture risk within the general population and that supplemental vitamin A may be harmful even within recommended limits. No studies have examined the relationship between biochemical retinol status and fracture in older women.

Materials and Methods: We examined serum retinol, retinyl palmitate, and β-carotene as predictors of incident hip and other fractures in a large prospective study of British women over the age of 75 years (n = 2606, 312 incident osteoporotic fractures, 92 incident hip fractures; mean follow-up duration, 3.7 years). Fasting blood samples (9:00-11:00 a.m.) were collected at baseline. Using a case-control design (three controls per case), serum retinol, retinyl palmitate, and β-carotene were assessed as univariate predictors of incident osteoporotic fracture or hip fracture. Baseline BMD at the total hip, age, 25(OH)D, serum β Crosslaps, bone-specific alkaline phosphatase, weight, height, and smoking were considered as covariates in a multivariate model.

Results: Serum retinol, retinyl palmitate, and β-carotene were not significant univariate predictors of either hip fracture or any fracture (all p > 0.05; Cox proportional hazards regression). For all osteoporotic fractures, the hazard ratio (HR) was 0.92 (95% CI, 0.81-1.05) per 1 SD increase in serum retinol. Risk of any osteoporotic fracture was slightly less in the highest quartile of serum retinol compared with the lowest quartile (HR, 0.85; 95% CI, 0.69-1.05; p = 0.132) There was a tendency for increased serum retinol to predict benefit rather than harm in terms of BMD (r = 0.09, p = 0.002). Multivitamin or cod liver oil supplementation was associated with a significantly lower risk of any fracture (HR, 0.76; 95% CI, 0.60-0.96; p = 0.021). In multivariate analysis, only age, total hip BMD, and weight were associated with fracture risk (p < 0.05).

Conclusions: We found no evidence to support any skeletal harm associated with increased serum indices of retinol exposure or modest retinol supplementation in this population.


Much recent attention has focused on the potential deleterious impact of vitamin A intake on fracture risk.(1–10) Numerous case studies have shown that overt vitamin A toxicity is associated with increased bone resorption and hypercalcemia.(11–13) Some studies have found that higher dietary vitamin A intake may account for a component of fracture risk within the general population and that supplemental vitamin A may be harmful even within recommended limits.(1–3) However, there is substantial lack of consensus, possibly because of differences in methodology or study population.(5,6,9,10,14) Dietary intake of vitamin A intake is difficult to assess and may be collinear with intake of other nutrients, and statistical adjustment for nutrient covariance may be problematical.(14) Some studies have included individuals who are very heterogeneous in terms of age or sex.(1,3) Other studies have been difficult to interpret because of a substantial time-lag between exposure assessment and fracture.(4) The effect of retinol status on hip fracture in particular has been less thoroughly studied.(4)

An apparent substantial relationship between dietary intake of vitamin A and hip fracture rate has been reported in two studies.(1,2) In the first of these studies,(1) women with intake of preformed vitamin A in the upper quintile of the population distribution (>2000 μg/day of vitamin A) had an 89% greater risk of hip fracture compared with those in the lowest quintile (<500 μg/day). Increased fracture risk was evident only for intake of vitamin A but not β-carotene. Although retinol supplements or multivitamins were the primary source of retinol intake in this study, current or previous supplement use was a weak and nonsignificant predictor of fracture. In the second study,(2) the risk of hip fracture was increased 2-fold in postmenopausal women with preformed retinol intake >1500 μg/day in comparison with those whose intake were <500 μg/day. In marked contrast, two recent prospective studies found no association between dietary vitamin A intake and risk of fracture.(5,6)

Only one prospective study(4) has addressed exposure to vitamin A using biochemical markers of retinol intake in relation to fracture risk. This cohort study(4) followed 2322 middle-aged Swedish men for 20 years and reported a 64% increase in risk of any osteoporotic fracture for men in the top quintile of serum retinol relative to those in the middle quintile. The corresponding increase in risk was 250% for hip fracture, although there were only 10 cases of hip fracture in the reference (middle) quintile. The risk of hip fracture was not monotonic and was lowest in the designated reference quintile. It was not clear whether the choice of the middle quintile as reference was related to the a priori hypothesis. In this study, serum β-carotene was not related to risk of fracture. The long-term stability of vitamin A intake is uncertain, and only 4% of individuals in this study were taking retinol-containing supplements. In contrast, no association between serum concentrations of retinyl esters and BMD was observed in a large study of ∼6000 North American men and women in the NHANES III study.(7) Fasting serum retinyl esters are purported to be the most sensitive marker of vitamin A status.(15)

The aim of this study was to examine the relationship between vitamin A exposure, assessed by measures of serum retinol, retinyl palmitate, and β-carotene, and risk of incident osteoporotic fracture in a study of British women >75 years of age.(16)


Study sample

The study was a prospective nested case control study. Case and controls were identified from the placebo arm of a population-based Medical Research Council funded prospective study of hip fracture in elderly women.(16) This study was designed to test the effects of the bisphosphonate clodronate on hip fracture incidence in women >75 years of age within the general unselected population. A total of 35,000 women were contacted from health service registers and invited to participate; 5212 individuals contacted agreed to participate in the main study and an initial pilot study and were followed for a mean duration of 3.7 ± 1.2 (SD) years. Exclusion criteria included bilateral hip arthroplasty, hypocalcemia, leucopenia, impaired hepatic function, known malignancy, inflammatory bowel disease, impaired renal function, and use of a bisphosphonate or a calcium supplement at a dosage >500 mg/day. A total of 2606 women were allocated to the placebo arm. The sample was largely white (>99.8%). Twenty-one percent of participants had a prevalent osteoporotic nonvertebral fracture, and 15% had a known prevalent vertebral fracture. Recruitment began in November 1996 and was completed in 1999. The follow-up period ended in July 2002. Participants gave written informed consent, and the study was approved by South Sheffield Research Ethics Committee.

At enrollment, a fasting blood sample was collected between 9:00 and 11:00 a.m. Aliquots of serum were stored at −70°C and were protected from light. Baseline clinical assessment included a detailed medical and lifestyle history and anthropometry. BMD of the proximal femur was measured using DXA (Hologic QDR2000/QDR4500). Weight and height were measured using standard methods. Fractures were ascertained by access to several sources of information including medical records and home visits by nursing staff and were confirmed by radiology or surgical reports. Verification of fracture ascertainment was carried out in a subsample by comparison with incident fracture ICD codes at one local hospital. Ascertainment of fracture in this subsample was >98% for any fracture and 100% for hip fracture. Of nonhip incident fractures, 40% were of the forearm, 14% were clinical vertebral fractures, 11% were of the humerus, 9% were pelvic, and the remainder were other appendicular fractures (20%) and nonappendicular fractures (6%).

Selection of cases and controls was performed to allow two separate case control analyses: (1) individuals with an incident hip fracture during the follow-up period (case-patients) were compared with an unmatched random sample of individuals without hip fracture (controls) and (2) individuals with any osteoporotic fracture during the follow-up period (patients) were compared with an unmatched random sample of women without osteoporotic fracture (controls). Osteoporotic fracture was defined here as all fractures apart from those of the hands, feet, skull, and face. For each fracture case, we randomly sampled three control subjects from the remainder of the placebo arm. There were 332 incident osteoporotic fractures and 100 incident hip fractures during the follow-up period. There were 934 control subjects for the osteoporotic fracture analysis and a total of 273 control subjects for the hip fracture analysis. Blood samples were available for 312 osteoporotic fracture cases and 92 hip fracture cases.


All samples from cases and controls were analyzed in a randomized order by staff blinded to case-control status. Serum β-carotene and retinoids (retinol and retinyl palmitate) were measured by high-performance liquid chromatography as described elsewhere.(17) All measurements were carried out using a single batch of reagents for each analyte for the hip fracture and all fracture analyses. The between-batch reproducibility (CV) for retinol was 5.1%. Consistency of measurement with other laboratories was confirmed by participation in the UK NEQAS external quality control scheme. The chromatographic peak for β-carotene was too small to allow precise quantitation in 10% of individuals. In the case of retinyl palmitate, the chromatographic peak was large enough to allow precise quantitation in 25% of individuals, reflecting excess retinol stores in these individuals. Although retinyl palmitate is only quantifiable in a proportion of individuals, elevated concentrations of this ester are generally presumed to be the best index of retinol overload.(7,18) Plasma retinol was easily quantitated in all individuals.

Serum β Crosslaps (sβCTX), a biochemical marker of bone resorption, was measured by immunoassay with electrochemical detection (Elecsys 2010 immunoanalyser; Roche Diagnostics, Mannheim, Germany). The interassay (within-day) analytical CV was <8.1% over the reference interval. Bone-specific alkaline phosphatase (Bone ALP), a biochemical marker of bone formation, was measured by immunoassay (Beckman Access; Beckman Coulter, Fullerton, CA, USA). 25(OH)D was measured by radioimmunoassay after solvent extraction (IDS, Tyne and Wear, UK).


We considered the following potential confounders: age, serum 25(OH)D, βCTX, Bone ALP, total hip BMD, weight, height, smoking, exercise exposure, milk consumption, and multivitamin or cod liver oil supplementation.

Information on past smoking habits (ever/never), current smoking (yes/no), current daily milk consumption (never/occasionally/1-2 glasses per day/3-4 glasses per day/5 or more glasses per day), and childhood, adulthood, and current exercise exposure (never/occasionally/<1 hour per week/1-2 h per week/>2 h per week). Individuals reporting taking a multivitamin supplement or cod liver oil were assumed to be taking supplemental vitamin A for purposes of a secondary analysis. Because the study was not designed in advance to examine retinol intake, the exact composition of “multivitamins” was not recorded. However, practically all multivitamin supplements available in the United Kingdom contain retinol at a dose of 800 μg/day, and cod liver oil contains ∼100 μg of retinol per gram of oil.

Statistical analysis

Serum retinol, βCTX, Bone ALP, and 25(OH)D had skewed distributions and were log-transformed before analysis. Height, weight, and total hip BMD were normally distributed. Serum retinyl palmitate and β-carotene were analyzed as dichotomous variables (upper quartile versus the rest of the population).

Women were first compared according to case-control status for each predictor or covariate by t-test or Mann-Whitney U-test as appropriate. Correlation between serum retinol and other variables was performed using Pearson correlation analysis.

Cox proportional hazards regression models (univariate and multivariate) were used to estimate the hazard ratio for osteoporotic fracture or hip fracture with 95% CIs. Serum retinol was evaluated as a continuous variable (per SD increase) and as a categorical variable based on quartiles of the control population using the lowest quartile as reference. All tests were two-sided, and p < 0.05 was considered significant.

All variables were also examined in a multivariate stepwise model. Only those variables with a preselected entry criterion of p < 0.1 in a univariate model were entered into a multivariate model. The multivariate model used a backward elimination procedure at p < 0.05 with total BMD entered first. Statistical analysis was performed using NCSS 6.0.21 (Kaysville, UT, USA).


Participant characteristics and univariate analysis

Baseline characteristics of cases and controls for all osteoporotic fractures are shown in Table 1. Compared with control subjects, cases were slightly older (p = 0.002), had lower body weight (p < 0.001), and had lower hip BMD (p < 0.001). Baseline characteristics of cases and controls for incident hip fracture are shown in Table 2. Compared with control subjects, cases were also slightly older (p < 0.001), had lower body weight (p < 0.001), had lower hip BMD (p < 0.001), and were shorter (p < 0.05). Indices of vitamin A status did not differ significantly between cases and controls for all osteoporotic fractures or for hip fracture.

Table Table 1.. Participant Characteristics of Controls and Cases With any Incident Osteoporotic Fracture (Means and 95% CIs)
original image
Table Table 2.. Participant Characteristics of Controls and Cases With Incident Hip Fracture (Means and 95% CIs)
original image

Multivitamin supplements or cod liver oil were used by 39.4% of the entire study sample. Use of supplements was positively associated with serum retinol. The geometric mean (95% CI) for serum retinol was 2.07 μM; (range, 2.02-2.12 μM;) in supplement users versus 1.95 μM; (range, 1.90-1.99 μM;) in nonusers (p < 0.001, t-test). Mean 25(OH)D was also greater in supplement users compared with nonusers. The geometric mean (95% CI) for 25(OH)D was 47.2 nM (range, 45.6-48.8 nM) in supplement users compared with 38.1 nM (range, 37.0-39.2 nM) in nonusers (p < 0.001, t-test). Total hip BMD, sβCTX, and Bone ALP were not associated with supplement use.

Table 3 shows the correlation between serum retinol and a number of continuous covariates. Serum retinol was weakly but positively correlated with total hip BMD (r = 0.09, p < 0.01) and 25(OH)D (r = 0.12, p < 0.001). Serum retinol was not associated with any other baseline variables (milk drinking, exercise as a child, exercise as a young adult, past history of smoking) except for current smoking. Serum retinol was lower in current smokers compared with nonsmokers. The geometric mean (95% CI) for serum retinol was 2.02 μM; (range, 1.98-2.06 μM;) in nonsmokers versus 1.87 μM; (range, 1.74-2.00 μM;) in smokers (p = 0.041, t-test).

Table Table 3.. Pearson Correlation Coefficient of Serum Retinol With Other Variables at Baseline
original image

In univariate Cox regression models, serum retinol, serum retinyl palmitate, and serum β-carotene were not associated with risk of any osteoporotic fracture or hip fracture (Table 4). The upper 95% confidence limit of the hazard ratio for serum retinol and fracture was close to unity. Individuals taking vitamin A supplements had a 24% lower risk of any fracture (hazard ratio [HR], 0.76; 95% CI, 0.60-0.95; p = 0.021). Hip fracture was not associated with supplementation (HR, 0.86; 95% CI, 0.56-1.33; p = 0.51). Fracture risk was clearly associated with age, weight, and hip BMD.

Table Table 4.. Univariate HRs for Risk of any Osteoporotic Fracture or Hip Fracture
original image

In view of previous reports(3) suggesting different effects of dietary and supplemental retinol on skeletal health, the hazard associated with plasma retinol was also assessed in supplement users and nonusers separately. Serum retinol was not predictive of any osteoporotic fracture (HR, 1.02; 95% CI, 0.82-1.28; p = 0.826) or of hip fracture (HR, 1.37; 95% CI, 0.88-2.14; p = 0.164) among supplement users. Among supplement nonusers, serum retinol was not significantly predictive of any osteoporotic fracture, but higher plasma retinol tended to predict benefit rather than harm (HR, 0.89; 95% CI, 0.76-1.04; p = 0.146) or hip fracture (HR, 0.83; 95% CI, 0.61-1.13; p = 0.233).

The HRs for incident osteoporotic and hip fracture in relation to quartile of serum retinol are given in Table 5. The risk of fracture was similar in the upper quartiles relative to the lowest (reference) quartile. The risk of any osteoporotic fracture was slightly lower in the upper two quartiles, but this was not statistically significant.

Table Table 5.. Unadjusted HRs for Incident Osteoporotic and Hip Fracture by Quartile of Serum Retinol
original image

Multivariate analysis

Variables with a preselected entry criterion of p < 0.1 in a univariate model were entered into an initial multivariate model, followed by backward elimination at p < 0.05 with total BMD entered first. The only variables that were statistically significant in the multivariate model were age (p < 0.05) and total hip BMD (p < 0.001) in any osteoporotic fracture analysis and age (p < 0.01), total hip BMD (p < 0.001), and weight (p < 0.1) in hip fracture analysis (Table 6).

Table Table 6.. Multivariate Stepwise Cox Regression Model of Incident Osteoporotic Fracture or Hip Fracture
original image


Conflicting results have been observed in studies relating retinol intake or indices of retinol status to fracture risk.(14) These prospective data do not support the hypothesis of risk associated with increased retinol or with supplement use in a population-based cohort of elderly British women. Risk of hip or other osteoporotic fracture was not associated with serum retinol, serum retinyl palmitate, or supplement use.

A safe maximum amount of retinol cannot be identified from current studies. In the European Union, most vitamin A supplements contain 800 μg RE (the EU labeling RDA). The proportion of individuals consuming greater supplemental doses may differ between countries. Multivitamin supplements or cod liver oil were used by 39.4% of the study sample. This is similar to the proportion of women 75-84 years of age taking unprescribed multivitamin supplements including cod liver oil in the last British national diet survey of older people and has increased over recent decades.(19,20) The prevalence of retinol supplementation in this population of older British women is substantially greater than that recorded among Swedish men.(4)

In the United States, up to 10% of individuals in the general population may consume retinol supplements containing >1000 μg RE/day.(21) The National Diet and Nutritional Survey(19) reported that total vitamin A intake from food and supplements of British women over age 65 is 179% of the reference nutrient intake, which is 600 μg retinol equivalents/day.(22) Intake of vitamin A in older people in Great Britain is apparently somewhat less than that observed in the United States,(1) perhaps because of the different dietary instruments used to assess intake, diversity in food fortification practices, and differences in supplement use. Nevertheless about one-quarter of British women >65 years of age reported intake >1000 μg/d, and 8% had intakes >2000 μg RE/day of retinol equivalents.

Our results do not support the results of a prospective study of Swedish men,(4) which indicated that serum retinol in the highest quintile of the male population distribution (>2.64 μM;) was associated with a 64% increase in risk of any fracture and a 250% increase in risk of hip fracture, relative to the midquintile of men with serum retinol levels between 2.17 and 2.36 μM;. In this Swedish study,(4) the duration of follow-up extended for 30 years after baseline assessment of plasma retinol.

Plasma retinol seems to increase with age and is greater in men than in women.(23) However, the discrepancy between these studies is unlikely to arise from differences in the distributions of serum retinol in Swedish men relative to British women. In the Swedish study,(4) the risk of fracture increased substantially above the 80th percentile of serum retinol (>2.64 μM;), whereas in this study, the 80th percentile value for serum retinol was only slightly lower (2.51 μM;). The median serum retinol in our study was 2.01 μM;, which was 10% lower than the median of 2.26 μM reported in the Swedish study.(4) There is no a priori reason to think that plasma retinol would be much higher in Sweden than in the United Kingdom, because reported dietary retinol intakes in the two countries are very similar,(14) and use of supplements was only 4% in the Swedish study (versus 39% in our study).

The distribution of serum retinol in our sample was similar to that reported for the same population group in the National Health and Nutrition Examination Survey in the United States.(21) Direct comparison of serum retinol levels between studies may be problematic because of substantial interlaboratory variation in retinol assays.(24)

The relationship between serum retinol and the concentration of retinoids in bone is uncertain. Short-term administration of supplemental vitamin A or retinyl palmitate has a small but significant effect on serum retinol.(15,25) Serum retinol may not necessarily reflect subtle or short-term differences in vitamin A intake. In this study, serum retinol was higher in multivitamin supplement users, which supports the value of serum retinol as a biologic marker of vitamin A intake. The lack of association between serum retinol and risk of fracture found in this study is in keeping with the results of a cross-sectional study of 250 North American postmenopausal women, which found no association between serum retinol and radial BMD or fracture history.(8) More than one-third of this study population consumed a vitamin A supplement, and 8% had retinol intake >2000 μg retinol equivalents per day.

Retinyl esters usually comprise <5% of total serum vitamin A. If vitamin A intake is excessive, serum retinyl esters are markedly increased. The ratio of retinyl esters to total serum retinol may be a good marker of vitamin A excess.(15) However, there are few data on the time-course of response to altered retinol intake. In this study, we observed no association between serum retinyl palmitate and risk of fracture. This result concurs with the findings of a large cross-sectional study, which investigated the relationship between BMD and the serum concentration of retinyl esters in North American men and women.(7)

Previously reported relationships between dietary retinol intake and various measures of osteoporotic risk do not necessarily imply that these associations are causal. Other contributory factors in individuals with high or low retinol status could confound interpretation of the studies. Vitamin A intake is collinear with intake of several other nutrients (including calcium, protein, vitamin D, and vitamin K) and lifestyle factors (smoking and physical activity), which may influence skeletal health. For example, the correlation with dietary vitamin D intake was around r = 0.88 in one Icelandic study,(9) whereas the Nurses Health Study(1) reported a similar correlation (r = 0.76). In this study, we observed a weaker but positive correlation between serum 25(OH)D and serum retinol (r = 0.11, p < 0.001).

It has been suggested that there might be direct antagonistic interactions between vitamins A and D in terms of skeletal health.(26) While supplement users had markedly higher levels of 25(OH)D and retinol, serum 25(OH)D was not a predictor of fracture in this population. Thus, it would seem that, whereas supplement users have a lower risk of fracture, this is not caused by vitamin D status.

Other studies have found that the apparent relationship of BMD with retinol intake differs depending on supplement use.(3) Similarly, we have also observed that increased serum retinol is associated with slightly but nonsignificantly lower fracture risk only in supplement nonusers. Other studies have also reported a positive rather than a negative association between retinol status and BMD in postmenopausal women.(10) Studies in rats have shown that supplementation with vitamin A is associated with an improvement in BMD(27) or has no effect on mineralization.(28)

The lower fracture rate risk observed in supplement users in our study, as well as the apparent deleterious effects observed in other studies, may not be a direct effect of the supplement. In the United Kingdom, supplement use is associated with social class, nonsmoking, and physical activity.(20) Supplement users also report eating more fruit and vegetables and oily fish and are more likely to suffer from musculo-skeletal disorders such as arthritis.(20) Similarly, in the United States, supplement use is associated with ethnic origin, age, educational status, physical activity, fruit intake, dietary fiber, low body weight, salt intake, and caffeine intake.(29,30) The relative importance of these confounders may differ in different populations.

Data from the prospective Nurses Health Study(1) involving 72,337 postmenopausal women and 18 years of follow-up support a detrimental effect of dietary retinol intake on hip fracture risk. Dietary retinol and preformed retinol and supplement use were estimated using food frequency questionnaires on five occasions throughout the study. Individuals with high vitamin A intake tended to have higher calcium, protein, and vitamin K intake, lower alcohol intake, were less likely to smoke, were more physically active, and took multivitamin, vitamin A, or β-carotene supplements.(1) Sixty-seven percent of those in the highest quintile of vitamin A intake took a multivitamin supplement in comparison with 17% in the lowest quintile.(1) Adequate statistical correction for these confounders may not be possible.

Only a small proportion of dietary β-carotene is converted to vitamin A (estimates range from one-sixth to one-twelfth). It is likely that serum β-carotene predominately reflects fruit and vegetable intake. Some observational studies have reported that a high intake of fruit and vegetables is associated with a reduced fracture risk.(31) We did not find an association between serum β-carotene and risk of fracture, in accordance with other studies of serum β-carotene and fracture risk(4) and studies of dietary β-carotene intake in relation to fracture(1) and BMD.(3)

The Rancho Bernardo study(3) assessed vitamin A intake by food frequency questionnaire in >1000 elderly men and women in California. Vitamin A intake was assessed as preformed retinol from food and supplements but excluded intake of preformed retinol. It was concluded that there is an “inverse U-shaped” relationship between retinol intake and BMD change and that supplement use affects the relationship between retinol intake and BMD. For example, among female supplement users, a higher retinol intake was associated with a lower bone mineral mass and greater bone mineral loss over 4 years, whereas among female nonsupplement users, a high retinol intake was associated with a greater bone mass and lower bone loss.

The tendency toward higher fracture risk in those with the lowest retinol intake or biochemical retinol status in our study and in previous studies(3,10,32) may relate to the well-described role of retinol binding protein as a negative acute phase protein,(23) and therefore, to the effect of general ill-health on fracture risk, rather than any direct effect of retinol status.

In summary, we found no association between biochemical retinol status or supplement use and either hip fracture or any osteoporotic fracture. If anything, vitamin A supplement users had a lower risk of fracture. There was no evidence of a “U-shaped” dose response in our data. Stratified analysis did not show that individuals in the highest quartile of serum retinol or those with highest retinyl palmitate had an increased risk of fracture. We suggest that there is not sufficient evidence to support the elimination of retinol supplements from multivitamin supplements or restriction of intake of dietary preformed retinol or β-carotene on the basis of skeletal risk.


This work was supported by research grants from the Food Standards Agency and the Medical Research Council. We thank Jenny Cliffe, Linda Reaney, Ann Hinch, Carol McGurk, Julie Porter, Jo Boreham, Matthew Potter, Marilyn Hill, and Anthony Colwell.