Osteoporotic fracture is a significant cause of morbidity and mortality and is a challenging global health problem. Previous reports of the relation between vitamin A intake or blood retinol and risk of fracture were inconsistent. We searched Medline and Embase to assess the effects of vitamin A (or retinol or beta-carotene but not vitamin A metabolites) on risk of hip and total fracture. Only prospective studies were included. We pooled data with a random effects meta-analysis with adjusted relative risk (adj.RR) and 95% confidence interval (CI). We used Q statistic and I2 statistic to assess heterogeneity and Egger's test to assess publication bias. Eight vitamin A (or retinol or beta-carotene) intake studies (283,930 participants) and four blood retinol level prospective studies (8725 participants) were included. High intake of vitamin A and retinol were shown to increase risk of hip fracture (adj.RR [95% CI] = 1.29 [1.07, 1.57] and 1.40 [1.03, 1.91], respectively), whereas beta-carotene intake was not found to increase the risk of hip fracture (adj.RR [95% CI] = 0.82 [0.59, 1.14]). Both high or low level of blood retinol was shown to increase the risk of hip fracture (adj.RR [95% CI] = 1.87 [1.31, 2.65] and 1.56 [1.09, 2.22], respectively). The risk of total fracture does not differ significantly by level of vitamin A (or retinol) intake or by blood retinol level. Dose-response meta-analysis shows a U-shaped relationship between serum retinol level and hip fracture risk. Our meta-analysis suggests that blood retinol level is a double-edged sword for risk of hip fracture. To avoid the risk of hip fracture caused by too low or too high a level of retinol concentration, we suggest that intake of beta-carotene (a provitamin A), which should be converted to retinol in blood, may be better than intake of retinol from meat, which is directly absorbed into blood after intake. © 2014 American Society for Bone and Mineral Research.
Osteoporosis is one of the major global health problems. Osteoporotic fracture is a significant cause of morbidity and mortality, becoming a challenging global burden.[1-3] One projection indicated that the number of hip fractures will rise to 6.26 million worldwide by 2050.
Vitamin A, retinol, beta-carotene, and its metabolites are involved in bone metabolism. Some studies reported that excessive intakes of vitamin A had adverse skeletal effects of decreasing bone mineral density and increasing incidence of fractures.[5-7] However, findings by others were inconsistent.[8-11] The aim of this review is to evaluate the evidence from prospective studies on the relation between vitamin A (or retinol or beta-carotene) or blood (serum or plasma) levels of retinol and the risk of fracture (hip fracture and total fracture).
Materials and Methods
We first searched Medline and Embase on August 10, 2013. Key words used were “vitamin A,” “retinol,” “carotene,” “beta-carotene,” “carotenoid,” “fracture,” and “hip fracture.” The function of “related article” was also used for search. The reference of retrieved articles were manually searched to avoid initial miss. A track search was performed on October 16, 2013, to include the new studies published between August 10 and October 16.
Studies were included in this meta-analysis according to the following criteria: 1) designed as a prospective study; 2) the exposure of interest in vitamin A or retinol or beta-carotene intake or blood (serum or plasma) retinol level; 3) the primary outcome of interest in hip fracture or total fracture of whole body; 4) the relative risk (RR) estimates with 95% confidence intervals (CI) are reported or could be calculated by data reported. If the data are duplicated and reported in more than one study, only the study of the largest number of cases was included. All potential studies were reviewed independently for eligibility by two authors (AMW and CQH), and any disagreement was discussed and resolved with the third independent author (WFN).
Two authors (NFT and ZKL) independently extracted data for analysis, and the third author checked the consistency between them. A standard data extracted form was used, including the first author's last name, publication year, country where the study was performed, study period, sample size (cases and controls or cohort size), the sex and age of participants, measure and range of exposure, variables adjusted for analysis, and RR estimates with corresponding 95% CIs for each category of vitamin A, retinol, and beta-carotene intake or blood (serum or plasma) retinol levels. If there were two or more RRs of different potential confounders, we extracted the RRs that reflected the greatest degree of control for potential confounders. If necessary, the primary authors were contacted to retrieve additional information. The study quality was assessed by using the nine-star Newcastle-Ottawa Scale.
Study-specific RR estimates were combined using a random-effects model, which considers both between-study variation and within-study. In a preview of the included studies, the included blood retinol studies consistently showed a lower RR at the category of about 2 to 2.4 umol/L;[13, 14] therefore, to acquire a more accurate result, RRs and 95% CIs of both highest and lowest versus this middle lower RR's category were pooled for synthesis separately. Dose-response meta-analysis was also used to analyze the relation of blood retinol level and risk of fractures (total fracture and hip fracture). The method of dose-response meta-analysis was according to Orsini and colleagues, whereas the methods of random-effects meta regression models were according to Greenland and colleagues.[15-17] We found that in vitamin A (retinol or beta-carotene) intake studies using the lowest category of vitamin A or retinol or beta-carotene as reference, RR is not obviously decreased at the middle category; therefore, we pooled the RRs and 95% CIs only of the highest versus lowest category for synthesis.
Q and I2 statistics were used to evaluate the statistical heterogeneity. Sensitivity analysis involved removing one study and evaluating whether the rest results would be markedly affected. Each study involved in the meta-analysis was deleted each time to reflect the influence of the individual data set on the pooled RRs. Potential publication bias was evaluated by the method of Egger's regression asymmetry test. All statistical tests were performed with the STATA software (version 12.0; StataCorp, College Station, TX, USA).
The selection of literature for included studies is shown in Fig. 1. A total of 574 potential records were identified from the databases, and 563 studies were excluded. Eight vitamin A (or retinol or beta-carotene) studies[8-10, 20-24] and three blood retinol studies[13, 14, 25] were identified at first. One blood retinol study available online was added by track research on October 16, 2013.
The characteristics of the included vitamin A (retinol or beta-carotene) intake studies are shown in Table 1; blood retinol studies are shown in Table 2. The total number of participants is 283,930, with 3693 hip fracture cases and 3487 total fracture cases in vitamin A (retinol or beta-carotene) intake studies, and 8725 with 348 hip fracture cases and 701 total fracture cases in blood retinol studies.
|Source||No. of participants||Location/period||Sex||Age (years)||No. of cases||Measure/range of Exposure (ug/d)||Study qualitya||Adjustment for covariates|
|Melhus 1998||1120||Sweden||F||40–76||247 HF||Retinol:||6||Physical activity in leisure time, BMI, smoking, HRT use, diabetes, former athletic activity, cortisone use, menopausal status and menopausal age, previous osteoporotic fractures, energy intake|
|Q5: > 1500|
|Feskanich 2002||72,337||United States||F||34–77||603 HF||Vitamin A:||7||Age, follow-up cycle, BMI, HRT use, smoking, leisure-time activity per week, thiazide diuretics use, calcium intake, protein, vitamin D, vitamin K, alcohol, caffeine|
|1980–1998||Q1: < 1250|
|Q5: ≥ 3000|
|Q1: < 500|
|Q5: ≥ 2000|
|Q1: < 500|
|Q5: ≥ 2000|
|Lim 2004||34,703||United States||F||55–69||525 HF||Vitamin A:||7||Age, BMI, diabetes, cirrhosis, past irregular menstrual duration, thyrotropic medication, sedative medication, steroid, antiepileptic medication, diuretic medication, education, alcohol, energy intake|
|1985–1997||Q1: < 7055 IU|
|Q5: ≥ 19893 IU|
|Q1: < 1405 IU|
|Q5: ≥ 7002 IU|
|Rejnmark 2004||1141||Denmark||F||45–58||163 TF||Retinol:||7||Age, years, weight, diet, and lifestyle, use of medicine, thyretoxicosis, BMD|
|5-year FU||Q1: < 600|
|Q5: ≥ 1500|
|Key 2007||34,696||United Kingdom||F: 26,749||20–89||1898 TF||Retinol:||6||Age, smoking, other nutrient intakes, alcohol, BMI, walking, cycling, vigorous exercise, other exercise, physical activity at work, marital status, HRT use|
|Average 5.2 years FU||M: 7947||Q1: < 200|
|Q5: ≥ 1000|
|Sahni 2009||929||United States||F/M||67–95||100 HF||B-carotene (average):||8||Sex and estrogen use, age, BMI, height, total energy intake, physical activity index, alcohol, smoking, calcium, vitamin D, potassium|
|Caire 2009||75,747||United States||F||50–79||588 HF||Vitamin A:||7||Age, energy, vitamin K, protein, alcohol, caffeine, smoking, BMI, HRT use, total METs per week, ethnic group and region, vitamin D, calcium|
|1993–2005||1040 TF||Q1: < 5055|
|Q5: ≥ 7508|
|Q1: < 474|
|Q5: ≥ 1426|
|Dai 2013||63,257||Singapore||F: 35,298||45–74||1630 HF||Beta-carotene: Q1: < 850.4||6||Age, year, dialect group, BMI, education, total energy intake, smoking, physical activity, calcium, soy isoflavones, vitamin B6, menopausal status, HRT use, diabetes, stroke|
|1993–2010||M: 27,959||Q4: ≥ 1772.4|
|Source||No. of participants||Location/period||Sex||Age (years)||No. of cases||Measure/range of exposure (umol/L)||Study qualitya||Adjustment for covariates|
|Michaelsson 2003||2322||Sweden||M||49–51||84 HF||Serum retinol:||8||Age, weight, height, serum beta carotene, calcium, albumin values, smoking, marital status, socioeconomic class, physical activity at work, leisure physical activity, and alcohol consumption|
|1970–2001||266 TF||Q1: <1.95|
|Opotowsky 2004||2799||United States||F||50–74||172 HF||Q1: <1.61||8||Age, weight, serum albumin, serum cholesterol, alcohol, recreational and nonrecreational physical activity, HRT use, previous fracture, dietary calcium intake, race|
|Q5: ≥ 2.56|
|Barker 2005||2606||United Kingdom||F||>75||92 HF||Serum retinol:||8||Age, height, total hip BMD, 25(OH)D, beta-CTX (ng/mL), bone ALP|
|1996–2002||312 TF||HF: Q1: <1.58|
|Q4: ≥ 2.34|
|TF: Q1: <1.66|
|Q4: ≥ 2.42|
|Ambrosini 2013||998||Australia||F: 335||15–80||123 TF||Plasma retinol:||7||Age, sex, medications, previous fracture, smoking status|
|1990–2007||M: 663||T1: 0.3–2.8|
Vitamin A (retinol or beta-carotene) intake and fracture risk
Only one study concerns the relation between beta-carotene and risk of total fracture and cannot reach a meta-analysis. Other results of meta-analyses are shown in Fig. 2. The present meta-analysis of highest versus lowest category shows that the adjusted relative risk (adj.RR) of vitamin A and retinol intake for total fracture is 0.96 (0.90, 1.03) and 0.97 (0.92, 1.03), respectively. High intake of vitamin A and retinol has been found to increase risk of hip fracture (adj.RR [95% CI) = 1.87 [1.31, 2.65] and 1.56 [1.09, 2.22], respectively), whereas high beta-carotene intake does not increase the risk of hip fracture (adj.RR [95% CI]) = 0.82 [0.59, 1.14]). Moderate heterogeneity was observed by studies of retinol intake for risk of hip fracture (I2 = 64.0%, p = 0.04) and beta-carotene intake for risk of hip fracture (I2 = 78.2%, p = 0.003).
Blood retinol level and fracture risk
By previewing the included studies of blood retinol level and fracture risk, the lowest RR is more likely at the middle category. We pooled RRs and 95% CIs of both the highest and lowest versus this middle lower RR category for synthesis (Fig. 3). The results showed the pooled RRs (95% CIs) of total fracture risk of the highest and lowest versus this middle lower RR category are 1.22 (0.84, 1.78) and 1.13 (0.91, 1.41). For hip fracture, the pooled RRs (95% CIs) are 1.87 (1.31, 2.65) and 1.56 (1.09, 2.22), respectively. Only moderate heterogeneity was found in studies of high-level blood retinol for risk of total fracture (I2 = 56.7%, p = 0.099).
We then assessed the dose-response relationship between blood retinol levels and relative risk of hip and total fracture. The results showed a U-shaped relationship between serum retinol level and hip fracture risk (Fig. 4A). The RR is decreased before the knot of 2.14 umol/L retinol concentration, and then the curve goes up with the increase of retinol concentration. For blood retinol levels and RR of total fracture, the curve is flatter (Fig. 4B). The result is consistent with the forest plot of Fig. 3.
Sensitivity analysis and publication bias
The results suggest that the influence of each individual data set to the pooled RRs is not significant. The Egger's test showed no evidence of publication bias for vitamin A or retinol intake for total fracture (p = 1.000; Egger's p value is an error—this p value is from Begg's test; p = 0.397); vitamin A or retinol or beta-carotene intake for hip fracture (p = 0.312; p = 0.338; p = 0.406); highest or lowest blood level for total fracture (p = 0.796; p = 0.611); and highest or lowest blood level for hip fracture (p = 0.618; p = 0.632).
Vitamin A includes vitamin A1 (retinol) and vitamin A2 (dehydroretinol). The provitamin A carotenoids (mainly beta-carotene) could be converted to vitamin A in the body; therefore, the relation between beta-carotene intake and risk of fracture is included in the present study.
Our meta-analysis of prospective studies suggests that there is no evidence that high intake of vitamin A or retinol will increase the risk of total fracture; however, they will increase the risk of hip fracture. Beta-carotene mainly from plant-based foods such as fruits and vegetables was not found to increase the risk of hip fracture in present meta-analysis. Therefore, we propose that the high intake of vitamin A that increases the risk of hip fracture is mainly contributed by retinol, not beta-carotene.
The reasons may be that retinol from meat (fish, liver, poultry, or dairy foods) could be directly absorbed into the blood. One of the bioactive products of retinol is retinoic acid, which stimulates the formation of osteoclasts and inhibits the activity of osteoblasts. Thus, a high dose of retinol will increase the risk of hip fracture. Sugiura and colleagues reported that people with high serum beta-carotene had less developed osteoporosis than those with a low level of serum beta-carotene. In their study, osteoporosis was defined as a person whose T-score (which shows how a subject's BMD compares with the young adult mean) is less than 70%, according to the guidelines on the management of osteoporosis by the Japan Osteoporosis Society. Our result showed the RR of hip fracture of the highest beta-carotene intake category compared with the lowest category is 0.82 (0.59, 1.14), without increasing the risk of hip fracture.
Unlike retinol, which is directly absorbed into blood, beta-carotene is one of provitamin A. It will be converted to beta-apo-carotenals and retinoids in the body and the process is adjusted by enzyme; therefore, if the retinoid concentration in blood is sufficient for metabolism, the convert process will be suppressed by a feedback mechanism. This may be the reason why the high-level intake of beta-carotene does not increase the risk of hip fracture, whereas the high-level intake of retinol does.
Pooled analysis of blood retinol level and risk of hip fracture also proves the importance of maintaining a reasonable concentration of blood retinol. Too low or too high a blood retinol level will increase the risk of hip fracture. From the figure of the dose-response meta-analysis of blood retinol levels and risk of fracture, we find the lowest RR of hip fracture is the knot of blood retinol concentration: 2.14 umol/L. Because the accurate dose value is not available when adj.RR = 1.0, we estimate the optimal blood retinol concentration by measurement and calculation of Fig. 4A. We found that in the range of 1.81 to 2.53 umol/L, the adj.RR value is below the line of RR = 1, and in the range of 1.99 to 2.31 umol/L, the upper line of the 95% CI was below the line of RR = 1.0. Therefore, we suggest that 1.99 to 2.31 umol/L of blood retinol is the optimal concentration, and 1.81 to 2.53 umol/L is the suboptimal concentration. Meta-analysis of retinol intake studies and blood retinol studies consistently find that high retinol has an adverse effect of hip fracture risk.
However, meta-analysis results of beta-carotine intake studies and blood retinol studies are inconsistent. This evidence supports that beta-carotine (one of provitamin A) converted to retinoids is adjusted by a feedback mechanism, and the high intake of beta-carotine will not increase the blood retinol level by the feedback mechanism, so there's no increase in the risk of hip fracture. Our meta-analysis has limitations common to individual studies, such as being unable to solve problems with confounding factors that are inherent in the included studies, as well as being an unblinded, nonrandomized control study. However, this analysis has many strengths, including the large number of participants, long duration of follow-up, and most individual studies are well powered. Our quantitative assessment is based on prospective studies, which overcome the shortcomings of recall or selection bias in retrospective studies. In addition, in meta-analyses of retinol intake studies and blood retinol studies that consistently find that high retinol has adverse effects on hip fracture risk, the evidence is convincing.
Another limitation of the present study is that the dose-response meta-analysis of vitamin A intake and risk of fracture cannot be performed. In some people who have a lower intake of retinol from meat (fish, liver, poultry, or dairy foods) but a high intake of beta-carotene from vegetables and fruits, the blood retinol may not be low because the retinol is converted from provitamin A carotenoids (mainly beta-carotene) in the body. Some people may have a lower intake of beta-carotene but a higher intake of retinol from meat. Only Feskanich and colleagues reported all of vitamin A, retinol, and beta-carotene intake; others studies were only observing retinol or beta-carotene. Blood retinol concentration will be influenced by retinol or beta-carotene intake but not be recorded. Therefore, it is worthless to do the dose-response meta-analysis of retinol or beta-carotene intake separately. However, the blood retinol level studies are measured by the concentration of serum retinol directly, avoiding the influence of isolated observation of retinol or beta-carotene intake, thus the dose-response meta-analysis of blood retinol levels and risk of fracture is valuable.
In 2005, Jackson and Sheenhan systematically reviewed the relationship of vitamin A and fracture based on six studies, without meta-analysis. Their review only indicated the potential risks associated with excess vitamin A intake. To the best of our knowledge, this is the first meta-analysis of the relationship between vitamin A and fracture risk based on prospective studies, and a quantitative assessment of the relationship between blood retinol level and risk of both hip and total fracture.
The clinical implication of this study is that blood retinol level is a double-edged sword for risk of hip fracture. To avoid the risk of hip fracture caused by too low or too high a level of retinol concentration, we suggest that intake of beta-carotene (a provitamin A), which is mainly from plant-based foods and should be converted (through a process adjusted by a feedback mechanism) to retinol in the blood, may be better than retinol from meat (fish, liver, poultry, or dairy foods), which will be directly absorbed into the blood after intake.
All authors state that they have no conflicts of interest.
This work is funded by National Natural Science Foundation of China (81372014, 81371988); Department of Health of Zhejiang Province, Backbone of Talent Project (2012RCB037); and Department of Science and Technology of Wenzhou, Wenzhou Science and Technology Project (Y20120073). The funders had no role in the design, execution, and writing of the study.
Authors' roles: AMW, CQH, and YLC contributed to the conception and design of the study. ZKL, ZYH, XHZ, and WFN contributed to the analysis and interpretation of data. AMW, XYW, and NFT contributed to the drafting of the article. All authors revised the manuscript critically for important intellectual content and gave final approval of the version to be published.