U-Shaped Association Between Serum 25-Hydroxyvitamin D and Fracture Risk in Older Men: Results From the Prospective Population-Based CHAMP Study



The aim of this population-based, prospective, observational study was to examine the relationship between serum levels of 25-hydroxyvitamin D (25OHD) and fracture risk in a cohort of 1662 community-dwelling men aged 70 to 97 years followed for a mean of 4.3 years. Data about mobility, muscle strength, balance, medication use, cognition, medical history, lifestyle factors, renal function, and serum 25OHD were collected at baseline. Data on radiologically verified fractures were collected every 4 months. The relationship between fractures and serum 25OHD levels was analyzed using Cox's proportional hazard regression. We accounted for bone mineral density, falls, physical activity, sun exposure, and season of blood draw, in addition to anthropometric and lifestyle factors, medical history, muscle strength, balance, and medication and supplement use. There were 123 first-incident fragility fractures. The relationship between baseline 25OHD and fracture risk was U-shaped, with increased fracture risk in men with either low or high serum 25OHD levels. In multivariate analysis, the risk of fracture was greatest in men with 25OHD levels in the lowest quintile (25OHD ≤36 nmol/L; hazard ratio [HR] = 3.5; 95% confidence interval [CI] 1.7–7.0) and in men in the highest quintile (25OHD >72 nmol/L; HR = 2.7; 95% CI 1.4–5.4) compared with men in the 4th quintile (25OHD ≥60 to ≤72 nmol/L). These associations were not explained by lower BMD, increased physical activity, fall risk, or other lifestyle or anthropomorphic factors. In community-dwelling older men, there appears to be a healthy target range for serum 25OHD concentrations. Thus, serum 25OHD levels too high and too low may be harmful in regard to fracture risk. © 2014 American Society for Bone and Mineral Research.


Studies of fracture risk associated with vitamin D supplementation and 25-hydroxyvitamin D (25OHD) are conflicting.[1, 2] This has led to an international debate about the optimum levels of serum 25OHD, with key organizations recommending different minimum levels ranging from 50 nmol/L to 100 nmol/L with further disagreement about the safe upper limits.[3-5] Meta-analyses of randomized control trials (RCTs) have reported that vitamin D supplementation in older institutionalized men and women resulted in fewer fractures.[6] However, results from RCTs in community-dwelling individuals are inconsistent,[6-9] and recently, two RCTs reported greater fracture risk associated with high-dose vitamin D supplementation.[10, 11] Conclusions from RCTs of vitamin D supplementation are hindered in part because baseline 25OHD and change in 25OHD are rarely measured. In addition, there are issues around compliance,[7] difficulty accounting for ultraviolet (UV) exposure, evidence of a nonlinear association between total intake of vitamin D and serum 25OHD levels, and wide variation in individuals' serum changes with supplements and UV exposure.[12]

In light of these difficulties, prospective observational studies may provide valuable insights into the relationship between serum 25OHD and fractures, and enable exploration of factors that might confound or mediate fracture risk. Of the longitudinal population cohort studies in older community adults that have examined the association between 25OHD and fracture risk,[13-23] only a few have focused on men[18, 23] or included men.[13, 17, 19] In line with findings from clinical trials of vitamin D, observational studies of serum 25OHD and fracture have drawn different conclusions.

Few studies have investigated the potential mediating effects of baseline bone mineral density (BMD),[17, 23] physical performance,[13, 23] or falls history.[17, 18, 23] Our primary goal was to investigate the relationship between 25OHD and fracture risk in a large population-based cohort of older men. In addition, we wanted to explore the influence of BMD, balance, strength, and falls history on the relationship.

Materials and Methods


Between 2005 and 2007, the Concord Health and Ageing in Men Project (CHAMP) enrolled 1705 community-dwelling men, aged 70 years and older, living in local government areas surrounding Concord Hospital, Sydney, Australia.[24, 25] The electoral role, on which all Australians must register, was used as the sampling frame. The only exclusion criterion was living in a residential aged-care facility. The baseline participation rate was 47%.

Incident fracture ascertainment

After the baseline assessment, we contacted men every 4 months by telephone to ascertain any incident fractures. Phone calls were made up to March 2011. If a fracture was reported, radiology reports were sought from the participants, hospital medical records, and radiology practices. Only fractures confirmed by radiographic reports were included in this study. We excluded pathological fractures and fractures of hands, fingers, feet, toes, and the head. We were unable to verify 22 fractures with radiology reports. Clinical spine fractures followed an episode of acute pain that prompted the participant to seek medical attention. All first-incident fractures that met the inclusion criteria were included, regardless of trauma level.[26, 27]

Time to censorship was either date of death, date of official withdrawal from the study, or date of the last telephone contact. The date of fracture is the date on the radiology report. If this was not recorded, for example, when the X-ray noted a recent healing fracture, then the date is that reported by the participant or, if the participant cannot remember, it is the midpoint between the phone call when the fracture was reported and the previous phone call.

Laboratory measurements

Fasting bloods were taken at baseline. Serum 25OHD and 1,25-dihydroxyvitamin D (1,25D) levels were measured by a commercially available radioimmunoassay detecting both 25OHD2 and 25OHD3 (Diasorin, Stillwater, MN, USA), with 100% cross reactivity of the antibody. Serum 25OHD sensitivity was 4 nmol/L; intra-assay precision was 7.6%, and interassay precision was 9.0%. Serum 1,25D had a sensitivity of <7.3 pmol/L, an intra-assay precision of 7.7%, and an interassay precision of 12.3%. Serum creatinine levels were used to estimate glomerular filtration rate (eGFR) using the Modification of Diet in Renal Disease (MDRD) formula for men:[28] eGFR = 175 × (Serum creatinine / 88.4) − 1.154 × age−0.203.

Dual-energy X-ray absorptiometry (DXA) measurements

Total hip areal BMD (g/cm2) was measured with DXA using the same fan-beam Discovery-W scanner (Hologic Inc., Bedford, MA, USA). The coefficient of variation (CV%) for scans duplicated on 30 men from the study cohort were 1.6% for the total hip. Quality-control scans conducted daily using the Hologic whole-body phantom indicated no shifts or drifts.

Vertebral deformities present at baseline were ascertained from lateral spinal scans of the area between thoracic 4/5 and lumbar 4/5, using the vertebral fracture assessment function of the DXA. The scans were assessed visually and semiquantitatively. A vertebral deformity was defined as a 20% or greater reduction in height of the anterior or mid portion of a vertebral body, relative to the posterior height of that body, or compared with the vertebral body above or below the affected vertebral body.

Other variables

Information on place of birth, smoking, alcohol-consumption lifestyle, and medical and family history was obtained from a questionnaire completed before attending the clinic. Birthplace was categorized as Australia, Mediterranean, Asia, and other. The questionnaire asked men about their use of walking aids, perception of their health, and functional limitations imposed by their health. General physical activity levels were determined using the Physical Activity Scale for the Elderly (PASE) questionnaire.[29] Participants were asked about time spent outdoors, sun avoidance behavior, and whether they had a sun tan.[30]

At baseline, weight was measured in clothing without shoes. Height was measured using a Harpenden stadiometer (Dyfed, UK). All prescription and nonprescription medications were brought to the clinic for verification. The Iowa Drug Information Service (IDIS) drug code numbers were used to code and classify the medication data.

A fracture and fall history was obtained at the clinic interview. Participants were asked about their falls during the previous 12 months. Falls were examined in a number of different categories. Grip strength (kg) of both hands was measured twice using a dynamometer (Jamar, Sammons Preston Inc., USA), and average strength was used in the analyses. Isometric quadriceps strength (kg) was measured three times on each leg using a 100-kg spring gauge with the knee at 90 degrees flexion. The best result was used for analysis. The chair-stand test (yes/no) assessed the ability and time to stand five times from sitting, without using arms. Usual gait speed was determined over a 6-m path. The narrow walk test was determined by the ability to walk 6 m within a 20-cm-wide path in two of three trials.

All participants were screened for cognitive impairment using the Mini-Mental State Examination (MMSE) and the Informant Questionnaire on Cognitive Decline.[31] Those who scored 26 or below on the MMSE or more than 3.6 on the IQCODE underwent further clinical assessment by a geriatrician. A diagnosis of depressive symptoms was made if the participant scored five or more on the self-completed short version (15 items) of the Geriatric Depression Scale.[32]

Statistical analysis

All statistical analyses were conducted using SAS version 9.1 (SAS Institute Inc., Cary, NC, USA). Student's t tests and chi-square tests were used to evaluate differences in baseline characteristics between men who had an incident fracture and those who did not. Baseline characteristics across quintiles of serum 25OHD were assessed using chi-square tests for categorical variables and ANOVA for continuous variables. Variables associated with both serum 25OHD and fractures at p < 0.2 and other factors identified by the literature, including physical activity and body mass index (BMI), were considered to be possible confounders.[12, 33-35]

Variables that showed deviations from a linear relationship were categorized. Serum 25OHD was categorized into quintiles. The cut points in nmol/L for serum 25(OH)D were: quintile 1 ≤36; quintile 2 >36 to ≤48; quintile 3 >48 to ≤59; quintile 4 >59 to ≤72; quintile 5 >72.

To account for seasonal variation[36] in serum 25(OH)D levels, all models were adjusted for season of blood draw, with seasons classified as summer (December to February), autumn (March to May), winter (June to August), spring (September to November). A new variable called “season-standardized 25OHD” was created for sensitivity analyses. The aim was to compare the risk of this derived variable with the risk of 25OHD calculated from models that accounted for season of blood draw as a categorical variable. Season-standardized 25OHD was calculated as: (individual's serum 25OHD – mean 25OHD of all blood draw in the relevant season). Similarly to 25OHD, season-standardized 25OHD had a nonlinear relationship with fracture risk and was analyzed in quintiles.

Cox's proportional hazards regression models[37, 38] were used to estimate relative risks (hazard ratios) and 95% confidence intervals (CI). The significance of parameter estimates derived from Cox's proportional hazards model was tested with the likelihood ratio statistic.[38] The final model was created using backward elimination to remove non-confounders.

Insight into possible mechanisms through which serum 25OHD might influence fracture risk were examined by observing the change in the hazard ratio (HR) when a variable associated with both fractures and serum 25OHD was included in a model. The percentage change in HR was calculated as: ([HR model 1–HR model 2) / HR model 1] × 100. A reduction of more than 10% of the HR has been used in previous studies to suggest a mediating role of a variable that lies on the causal pathway.[23]

Sensitivity analyses were conducted, excluding men on vitamin D supplementation (n = 130), men on calcium supplements (n = 184), and men on bisphosphonate medication (n = 86). Characteristics of men with missing serum 25OHD data (n = 43, 2.5% of cohort) were compared with men in this study using Student's t tests and chi-square tests.

Ethics approval

The CHAMP study was approved by the Concord Hospital Human Research Ethics Committee. Written and informed consent was given by all participants before their inclusion in the study.


Of the 1705 men in CHAMP, serum was collected at baseline in 1662 men. There were no differences in age, BMI, vitamin D supplement use, or fracture incidence, in the 43 men who did not have a serum 25OHD test.

There were 123 fractures during a mean follow-up of 4.3 ± 1.4 years. Most fractures resulted from a fall from standing height or below (76.6%). The mean age of the 1662 men who had baseline measures of 25OHD was 76.9 (70 to 97) years. Men who sustained a fracture were older, had lower BMD, were less physically active, had poorer neuromuscular function, and reported having more falls in the 12 months before baseline measures (Table 1).

Table 1. Baseline Characteristics of Men Who Experienced a Fracture and Men Who Did Not Fracture
 All men (N = 1662)No fracture (n = 1538)Fracture (n = 123) 
CharacteristicPrevalence (%) or mean ± SDpa
  1. ap value: ANOVA (continuous variables); chi-square (categorical variables).
  2. bp for nonlinear association.
  3. cSerum1,25D had neither a linear nor nonlinear association with fracture.
  4. dPASE: Physical Activity Scale for the Elderly (range 0 to 376, median 115).
  5. eCorticosteroid use >3 months was self-reported; vitamin D supplementation included sighted supplements and self-reported cod liver oil; all other meds were sighted at baseline clinic.
  6. fBaseline vertebral deformity: ≥20% decreased height using DXA vertebral fracture assessment (VFA).
  7. gNarrow walk test (6 m length, 20 cm wide × 2 laps).
  8. hSun tan self-reported as having had “obvious tan during past 6 months.”
Age (years)76.9 ± 5.576.8 ± 5.4578.7 ± 5.610.0001
Height (m)1.7 ± 0.071.7 ± 0.071.7 ± 0.070.4
BMI (kg/m2)27.8 ± 4.027.8 ± 4.0527.7 ± 3.990.7
Australian born825 (49.7%)758 (49.3%)67 (54.5%)0.02
Asian born67 (4.0%)63 (4.1%)4 (3.3%)0.7
BMD: total hip (g/cm2)0.939 ± 0.140.944 ± 0.140.866 ± 0.17<0.0001
Serum tests    
25OHD (nM) continuous55.7 ± 21.955.8 ± 21.955.6 ± 26.20.9
25(OH)D quintiles(Q)    
Q1 ≤36 nM323 (19.4%)287 (18.7%)36 (29.3%) 
Q2 >37 to ≤48 nM355 (21.4%)333 (21.6%)22 (17.9%) 
Q3 >48 to ≤59 nM324 (19.5%)307 (20.0%)17 (13.8%) 
Q4 >59 to ≤72 nM319 (19.2%)306 (20.0%)13 (10.6%) 
Q5 73+ nM340 (20.5%)306 (20.0%)35 (28.5%)0.0009b
1,25D (pM/L) continuousc109.5 ± 66.1109.4 ± 65.6109.6 ± 71.20.9
Creatinine (micromol/L)96.8 ± 40.196.1 ± 37.2105 ± 66.30.02
eGFR (mL/min/1.73m2)71.2 ± 1971.4 ± 18.768.4 ± 21.70.09
Current smokers101 (6.1%)89 (5.8%)12 (9.8%)0.08
Alcohol consumption (drinks/wk)8.3 ± 10.58.4 ± 10.58.0 ± 11.40.7
PASEd118 ± 58118 ± 57110 ± 600.1
Medical history    
Epilepsy20 (1.2%)16 (1%)4 (3.3%)0.03
Kidney disease55 (3.3%)48 (3.1%)7 (5.7%)0.1
Liver disease35 (2.1%)29 (1.9%)6 (4.9%)0.02
Osteoarthritis300 (18.1%)271 (17.6%)29 (23.6%)0.09
Paget's disease33 (2.0%)27 (1.8%)6 (4.9%)0.01
Rheumatoid arthritis163 (9.8%)150 (9.8%)13 (10.6%)0.8
Stroke138 (8.3%)121 (7.9%)17 (13.8%)0.02
Calcium supplement186 (11.2)161 (10.5%)25 (20.3%)0.0009
Vitamin D supplement130 (7.8%)120 (7.8%)10 (8.1%)0.8
Bisphosphonates86 (5.2%)74 (4.8%)12 (9.8%)0.01
Psychotropic226 (13.6%)202 (13.1%)24 (19.5%)0.04
Corticosteroids276 (16.6%)254 (16.5%)22 (17.9%)0.6
Fracture history    
Low-trauma fracture age ≥50 years166 (10%)146 (9.5%)20 (16.3%)0.02
Vertebral deformity at baselinef284 (17.1%)258 (16.8%)26 (21.1%)0.2
No. of falls in preceding 12 months    
01342 (80.8%)1257 (81.7%)85 (69.1%) 
1171 (10.3%)156 (10.1%)15 (12.2%) 
≥2138 (8.3%)118 (7.7%)20 (16.2%)0.0002
Good/excellent health (self-reported)1152 (69.4%)1097 (70.5%)78 (63.4%)0.1
Chair-stand time (s)16.7 ± 5.916.5 ± 5.419.6 ± 10.40.003
Grip strength (kg)32.6 ± 7.432.8 ± 7.430.8 ± 8.10.008
Leg strength (kg)30.9 ± 831.0 ± 828.7 ± 7.70.01
Unable to do chair stands137(8.2%)118(7.7%)19(15.4%)0.002
Unable to do narrow walk test × 2g723(43.5%)645(41.9%)78(63.4%)0.0005
6-m gait speed (m/s)0.93 ± 0.20.93 ± 0.20.88 ± 0.20.04
Depressive symptoms241 (14.5%)214 (13.9%)27 (22%)0.01
Dementia89 (5.4%)74 (4.8%)15 (12.2%)0.005
Season of blood draw    
Summer345 (20.8%)321 (20.9%)24 (19.5%) 
Autumn472 (28.4%)436 (28.3%)36 (29.3%) 
Winter409 (24.6%)381 (24.8%)28 (22.8%) 
Spring435 (26.2%)400 (26%)35 (28.5%)0.7
Sun tanh    
No sun tan1007 (60.6%)926 (60.2%)81 (65.9%) 
Slight sun tan485 (29.2%)453 (29.5%)32 (26.0%) 
Obvious tan147 (8.9%)140 (9.1%)7 (5.7%)0.1
Avoids direct sun    
Always avoid direct sun160 (9.6%)147 (9.6%)13 (10.6%) 
Usually avoid direct sun887 (53.4%)817 (53.1%)70 (56.9%) 
Never avoid direct sun590 (35.5%)553 (36%)37 (30.1%)0.3
Time spent outdoors    
Never goes outside13 (0.8%)12 (0.8%)1 (0.8%) 
Goes outside a few times a month62 (3.7%)57 (3.7%)5 (4.1%) 
Goes outside weekly82 (4.9%)77 (5%)5 (4.1%) 
Goes outside most days1482 (89.2%)1374 (89.3%)108 (87.8%)1.0

The mean (± SD) serum 25OHD level at baseline was 55.8 ± 22 nmol/L. Serum 25OHD levels were higher in autumn (mean 65.2 nmol/L) than in summer (57.8 nmol/L), winter (49.7 nmol/L), or spring (49.8 nmol/L). Table 2 shows characteristics of the study cohort by quintile of 25OHD. Men with the lowest serum 25OHD levels had lower hip BMD and poorer neuromuscular performance. They were more likely to report poorer health, diabetes, liver disease, and lower levels of physical activity. Men with both high and low serum 25OHD levels tended to be older, walk less for exercise, and were more likely to have self-reported kidney disease.

Table 2. Baseline Characteristics of Men Across Quintiles of Serum 25OHD (Mean ± SD or n [%])
 Quintiles of serum 25OHD (nmol/L)a 
 ≥3 to ≤36>36 to ≤48>48 to ≤59>59 to ≤72>72 to ≤148 
Characteristicn = 323 (19.4%)n = 355 (21.4%)n = 324 (19.5%)n = 319 (19.2%)n = 340 (20.5%)pb
  1. a25-Hydroxyvitamin D units 1 ng/mL = 1/2.496 nmol/L.
  2. bp value: ANOVA (continuous variables); chi-square (categorical variables).
  3. cPASE: Physical Activity Scale for the Elderly (range 0 to 376, median 115).
  4. dCorticosteroids >3 months at any time.
  5. eUnable to do narrow walk test (6 m length, 20 cm wide × 2 laps).
Mean serum 25OHD28.1 ± 6.642.7 ± 3.653.4 ± 3.165.5 ± 3.788.5 ± 14.2 
Age (years)77.6 ± 5.876.7 ± 5.576.3 ± 5.176.3 ± 5.577.6 ± 5.60.002
Height (m)1.69 ± 0.071.68 ± 0.0741.69 ± 0.071.69 ± 0.071.69 ± 0.070.3
Weight (kg)80.0 (13.0)80.1 (13.0)80.4 (13.5)79.1 (12.9)77.1 (12.3)0.002
BMI (kg/m2)28.2 ± 4.128.3 ± 4.328.1 ± 4.227.6 ± 3.726.9 ± 3.6<0.0001
Percent total body fat30.4 (3.9)30.5 (6.0)29.9 (5.8)29.2 (6.1)28.7 (5.9)<0.0001
Australian born142 (44)163 (45.9)146 (45.1)164 (51.4)210 (61.8)0.0005
Asian born13 (4)15 (4.2)17 (5.2)9 (2.8)13 (3.8)0.0005
BMD: total hip (g/cm2)0.907 ± 0.140.932 ± 0.140.953 ± 0.130.957 ± 0.150.945 ± 0.15<0.0001
Creatinine (micromol/L)96.1 ± 32.398.5 ± 6893.9 ± 2495.2 ± 28.399.9 ± 27.30.3
eGFR (mL/min/1.73m2)71.8 ± 19.772.4 ± 19.672.4 ± 18.571.8 ± 18.767.7 ± 17.90.005
Serum 1,25D (pM/L)86.2 ± 57.195 ± 53.8114.6 ± 59.7121 ± 72.9131.8 ± 74.7<0.0001
Current smokers30 (9.3)16 (4.5)16 (4.9)20 (6.3)19 (5.6)0.2
Alcohol (drinks/week)8 ± 9.57.8 ± 9.58.3 ± 12.98 ± 9.59.4 ± 10.90.3
PASEc99.1 ± 58.1112.2 ± 59129.2 ± 54.8123.8 ± 53.6125.8 ± 58.7<0.0001
Medical history      
Diabetes75 (23.2)80 (22.5)52 (16)53 (16.6)43 (12.6)<0.0001
Epilepsy8 (2.5)2 (0.6)6 (1.9)3 (0.9)1 (0.3)0.04
Kidney disease16 (5)12 (3.4)3 (0.9)8 (2.5)16 (4.7)0.7
Liver disease12 (3.7)7 (2)7 (2.2)6 (1.9)3 (0.9)0.02
Osteoarthritis48 (14.9)66 (18.6)57 (17.6)54 (16.9)75 (22.1)0.07
Paget's disease4 (1.2)9 (2.5)5 (1.5)5 (1.6)10 (2.9)0.3
Rheumatoid arthritis32 (9.9)45 (12.7)30 (9.3)28 (8.8)28 (8.2)0.1
Stroke37 (11.5)28 (7.9)17 (5.2)24 (7.5)32 (9.4)0.3
Calcium supplement24 (7.4)37 (10.4)45 (13.9)37 (11.6)43 (12.6)0.04
Vitamin D supplement13 (4)37 (10.4)29 (9)27 (8.5)24 (7.1)0.04
Bisphosphonates13 (4)25 (7)16 (4.9)18 (5.6)14 (4.1)0.4
Corticosteroidsd52 (16.1)74 (20.8)53 (16.4)44 (13.8)53 (15.6)0.2
Fracture history      
Low-trauma fracture age ≥50 years19 (5.9)22 (6.2)17 (5.2)15 (4.7)20 (5.9)0.7
Prevalent vertebral deformity55 (17)62 (17.5)60 (18.5)43 (13.5)64 (18.8)0.7
No. of falls in preceding 12 months      
0254 (78.6)283 (79.7)272 (84)264 (82.8)269 (79.1) 
130 (9.3)39 (11)32 (9.9)25 (7.8)45 (13.2) 
≥235 (10.9)28 (7.9)20 (6.2)30 (9.4)25 (7.3)0.3
Self-reported health      
Good/excellent health198 (61.3)231 (65.1)227 (70.1)233 (73)263 (77.4)<0.0001
Chair-stand time (s)16.7 ± 5.216.7 ± 5.516 ± 5.117 ± 7.517.1 ± 6.10.2
Grip strength (kg)31.1 ± 7.632.1 ± 7.633.5 ± 7.233.8 ± 7.232.5 ± 7.3<0.0001
Leg strength (kg)30.6 ± 8.130.3 ± 7.931.9 ± 7.931.2 ± 8.330.5 ± 7.80.2
6-m gait speed (m/s)0.90 ± 0.20.92 ± 0.20.96 ± 0.20.94 ± 0.20.91 ± 0.20.005
Unable to do chair stands41 (12.7)37 (10.4)17 (5.2)19 (6)23 (6.8)0.0007
Unable to do narrow walke163 (50.5)172 (48.5)130 (40.1)115 (36.1)143 (42.1)0.0008
Depressive symptoms65 (20.1)57 (16.1)42 (13)38 (11.9)39 (11.5)0.0002
Dementia25 (7.7)16 (4.5)11 (3.4)20 (6.3)17 (5)0.9
Season of blood draw      
Summer (n = 349)46 (14.2)74 (20.8)74 (22.8)75 (23.5)76 (22.4) 
Autumn (n = 493)47 (14.6)73 (20.6)83 (25.6)113 (35.4)156 (45.9) 
Winter (n = 421)120 (37.2)92 (25.9)81 (25)64 (20.1)52 (15.3) 
Spring (n = 442)110 (34.1)116 (32.7)86 (26.5)67 (21)56 (16.5)<0.0001
Sun tan      
No sun tan239 (74)229 (64.5)201 (62)161 (50.5)177 (52.1) 
Slight sun tan64 (19.8)99 (27.9)97 (29.9)113 (35.4)112 (32.9) 
Obvious tan11 (3.4)23 (6.5)23 (7.1)41 (12.9)49 (14.4)<0.0001
Avoidance of direct sun      
Always avoid direct sun33 (10.2)50 (14.1)28 (8.6)26 (8.2)23 (6.8) 
Usually avoid direct sun166 (51.4)176 (49.6)180 (55.6)178 (55.8)187 (55) 
Never avoid direct sun115 (35.6)125 (35.2)112 (34.6)111 (34.8)127 (37.4)0.2
Time spent outdoors      
Never goes outside4 (1.2)4 (1.1)3 (0.9)1 (0.3)1 (0.3) 
Goes outside a few times a month24 (7.4)11 (3.1)12 (3.7)8 (2.5)7 (2.1) 
Goes outside weekly18 (5.6)22 (6.2)17 (5.2)16 (5)9 (2.6) 
Goes outside most days268 (83)313 (88.2)289 (89.2)291 (91.2)321 (94.4)<0.0001

The relationship between baseline serum 25OHD and fracture risk was U-shaped, with an increased fracture risk in men with either low or high serum 25OHD levels. Of men who suffered a fracture, a third had 25OHD levels below 37 nmol/L and a third had 25OHD levels above 72 nmol/L. The number of spine and hip fractures in the highest and lowest quintiles was the same (Fig. 1). There was no difference in the fracture site (p = 0.3) (Fig. 1) or the mechanism of injury (p = 0.5) (data not shown) by 25OHD quintile. There was a trend for decreasing fracture risk in each 25OHD quintile up to quintile 4 (p = 0.002), after which fracture risk increased. Men with serum 25OHD between 59 and 73 nmol/L (quintile 4) had the lowest fracture risk. This group was used as the referent. In unadjusted models, men in the lowest quintile had a 3.2-fold higher risk of fracture compared with the referent group. Men in the highest quintile had a 2.6-fold higher risk of fracture (Table 3 and Fig. 2).

Figure 1.

Number of fractures by site in each 25OHD quintile.

Table 3. Effect of Adjustment for Possible Mediators on the Hazard Ratio (95% CI) of Fracture According to 25OHD Levels
 Quintiles of serum 25OHD (nmol/L)
Models≥3 to ≤36>36 to ≤48>48 to ≤59>59 to ≤72>72 to 148
  1. aMultivariate model adjusts for age, country of birth, BMI, physical activity, season of blood draw, previous low-trauma fracture after age 50 years, calcium supplement, and vitamin D supplement.
  2. bNeuromuscular measures: time taken to do five chair stands (in quartiles) and ability to complete two lengths of narrow walk test.
  3. cSensitivity analysis: multivariate model excluding any men taking any form of vitamin D supplementation (cod liver oil or tablets).
Unadjusted3.2 (1.7, 5.9)1.7 (0.8, 3.3)1.3 (0.6, 2.7)ref2.6 (1.4, 4.9)
Age adjusted2.9 (1.6, 5.5)1.7 (0.8, 3.3)1.3 (0.6, 2.7)ref2.4 (1.3, 4.6)
Multivariate modela3.5 (1.7, 7.0)1.9 (0.9, 4.0)1.4 (0.6, 3.0)ref2.7 (1.3, 5.4)
Multivariate modela + falls3.5 (1.7, 7.0)1.9 (0.9, 4.1)1.4 (0.6, 3.1)ref2.8 (1.4, 5.6)
Multivariate modela + BMD2.8 (1.4, 5.7)1.6 (0.8, 3.5)1.5 (0.7, 3.2)ref2.7 (1.3, 5.4)
Multivariate modela + neuromuscular measuresb3.3 (1.7, 6.8)1.8 (0.9, 3.9)1.5 (0.7, 3.2)ref2.7 (1.4, 5.5)
Multivariate modela + serum 1,25D3.6 (1.7, 7.5)2.1 (1.0, 4.5)1.4 (0.6, 3.2)ref2.6 (1.2, 5.2)
Sensitivity analysisc multivariate modela3.1 (1.6, 6.0)1.6 (0.8, 3.3)1.5 (0.7, 3.1)ref2.1 (1.0, 4.1)
Figure 2.

Survival curve depicting time to first fracture according to baseline serum 25OHD quintile.

The final model accounted for age, country of birth, physical activity, BMI, season of blood draw, history of low-trauma fracture, and calcium and vitamin D supplements (Table 3). There was no effect on the relationship between 25OHD and fracture risk when renal function, dementia, health status, and medications were accounted for. These factors were not significant and were not included in the final model. In the multivariate analyses, fracture risk remained greatest in men with serum 25OHD in the lowest quintile (HR = 3.5, 95% CI 1.7–7.0) followed by the risk of fracture for men in the highest quintile (HR = 2.7, 95% CI 1.3–5.4) compared with men in the 4th quintile. The relationships were similar when men on vitamin D supplements were excluded (Table 3). Analyses of season-standardized 25OHD and sensitivity analyses, which excluded men taking bisphosphonates or calcium supplements or men whose fracture had resulted from a motor vehicle accident or a fall above standing height, showed the same strong and statistically significant relationships between fractures and 25OHD (data not shown). In addition, we examined further the top quintile of 25OHD. Men with serum 25OHD ≥73 to ≤100 nmol/L (n = 297) had a 2.5-fold (1.3- to 5.1-fold) fracture risk, whereas men with serum 25OHD ≥100 nmol/L (n = 65) had a 3.1-fold (1.2- to 7.9-fold) fracture risk compared with quintile 4 (data not shown).

To further explore the relationship between 25OHD and fractures, we examined the effect of potential factors through which 25OHD may influence risk. The inclusion of BMD in the multivariate model decreased the HR for low serum 25OHD by 20%. Neuromuscular performance (chair-stand time and the ability to complete a narrow walk) decreased the HR of men in the first quintile by 6%. Fall history and serum 1,25D had no effect on the HRs (Table 3). Adding BMD and neuromuscular function to the model did not change the HR of men in the fifth quintile.

The risk of fracture increased with time for men in 25OHD quintile 1 (time interaction p = 0.04). To examine this interaction, the adjusted model was stratified using a median cut point of 57 weeks. Up to 57 weeks, the HR for the lowest quintile was 1.8. After 57 weeks, the HR rose to 3.5. There was no change in fracture risk over time in higher 25OHD quintiles.


In our prospective population-based study, we found older men with either high (≥73 nmol/L) or low levels (≤36 nmol/L) of 25OHD had a marked increased risk of all fractures. The U-shaped relationship remained after accounting for possible confounding factors.

A number of other studies in men have reported nonlinear relationships between 25OHD and fracture risk.[13, 17, 18] Similar to our study, men in the third quartile of NHANES III (61 to 83 nmol/L) had the lowest fracture risk,[17] and in the EPIC Oxford study, the risk of fractures in men with lower and higher levels of 25OHD was greater than for intermediate levels.[19] However, in the EPIC study, there were wide confidence intervals and the risk was not statistically different between groups. This lack of significance could reflect the lower risk of fractures in the relatively young EPIC cohort (mean age 50 years), the high mean serum 25OHD levels (81 nmol/L), or that fractures were self-reported 6 years after baseline. The Uppsala longitudinal study also reported a nonlinear relationship with higher fracture risk in men who had 25OHD below 40 nmol/L (HR = 1.65).[18] In this study, only 5% of men had 25OHD <40 nmol/L. In contrast to our findings, they found no increased risk at high levels of 25OHD. MrOS reported a strong relationship between 25OHD and hip fractures, with the highest risk in their lowest quartile (<47 nM) and no difference in risk between men in quartile 3 (63 to 69 nmol/L) and quartile 4 (70+ nmol/L) (Fig. 2).[15] However, they did not find any association between 25OHD and “all non-spine” fractures.

An important finding in our study was the high fracture risk in men with higher levels of serum 25OHD compared with men with moderate levels. Few men took vitamin D supplements, and potential confounders, including fall history and time spent outside exercising, did not explain this relationship. Some but not other studies[7] have reported increased fractures,[10, 14] falls,[10] frailty,[39] and decreased longevity[40] in people with higher levels of serum 25OHD.

Discordance in findings from different studies may in part reflect different populations and lifestyles. Australia has very limited vitamin D food fortification, and only 8% of men in CHAMP took any form of vitamin D supplementation. In addition, health marketing campaigns encourage Australians to avoid the sun to prevent skin cancers. Our study had the power to investigate both low and higher levels of 25OHD. The median 25OHD level in CHAMP men was considerably lower (53.0 nmol/L) than in most European and US studies of fractures in men, which reported medians between 61 and 69 nmol/L.[17, 18, 23] Furthermore, more than 40% of men in the CHAMP study had serum 25OHD below the minimum recommended rate of 50 nmol/L. This proportion of 25OHD-deficient men is significantly greater than in MrOS,[23] EPIC,[19] and NHANES III,[17] in which about 20% of men had levels less than 50 nmol/L.

We investigated several potential factors that might explain the influence of 25OHD on fracture risk. Lower baseline BMD appeared to explain 20% of the fracture risk seen in men in the lowest quintile of 25OHD, whereas measures of neuromuscular function (chair-stand time and ability to complete a narrow walk) reduced the hazard ratio by only 6%. Even after including all potential baseline confounding variables, the strong and significant risk of fractures in men in the lowest quintile of 25OHD remained. Our previous study found that only 25OHD <36 nmol was associated with more rapid BMD loss. This accelerated loss of BMD in the lowest 25OHD quintile group may explain why fracture risk increased over time in these men but did not increase in men in the highest 25OHD quintile.[41] The inclusion of potential confounders, including season of blood draw, fall history, physical activity, poor health status, and vitamin D supplementation, did not change the strong relationship between higher levels of 25OHD and greater fracture risk. Nor was the increased risk in the high 25OHD quintile mediated by either BMD or neuromuscular function.

One of the major strengths of our study was that we assessed a large number of potential confounders of the relationship between serum 25OHD and fractures, including many validated quantitative measures of neuromuscular function, cognition, BMI, physical activity, and health status. We adjusted for the season at the time of the blood draw and also created a variable to season-standardize 25OHD because of the importance of seasonal variation in serum 25OHD levels.[42, 43]

Fracture rates in CHAMP were similar to previously published Australian studies of fracture incidence.[44] Identification of incident fractures relied on participants reporting fractures, although it is unlikely that symptomatic fractures will have been missed because we conducted four monthly phone calls with participants.

Our study has several limitations. CHAMP is a prospective, observational study, not a randomized clinical trial. RCTs would be needed to test that bringing 25OHD to within 59 to 72 nM reduces the risk of fracture in order to show causation. Serum 25OHD assays are reported to vary greatly according to laboratory and assay method.[46] To counter this, assays in CHAMP were measured at the same time, in the same laboratory, and we used quintiles rather than specific cut points to examine the data. Our assay detected total serum 25OHD, which includes both vitamin D2 and vitamin D3. Our findings are based on baseline measures, and serum 25OHD, fall risk, and health status may change over time. There were few men in our cohort on vitamin D supplements at baseline; however, this may have changed as community awareness of vitamin D has increased. We do not have measures of parathyroid hormone, markers of bone modeling, or calcium intake, which could contribute to the relationship between 25OHD and fracture. We did not measure dietary intake, and vitamin D requirements may vary based on customary calcium intake.[45] Self-reported variables, such as fall history, may be inaccurate, and this may impede the identification of mediators in the 25OHD fracture relationship. Men in the CHAMP study are community dwelling and are likely to be healthier than men who did not participate in the study, which makes it likely that the low serum 25OHD may be even more prevalent in older Australian men than reported in this study.

Despite these limitations, after adjustment for many potential confounding factors, both high and low levels of serum vitamin D were strongly associated with increased fracture risk in a cohort of community-dwelling men older men who were not taking vitamin D supplements at baseline. Men with serum 25OHD levels between 59 and 73 nmol/L had the lowest risk. Fracture risk increased over time in men with the lowest 25OHD <37 nmol/L. The U-shaped relationship between 25OHD and fractures implies that vitamin D supplementation might need to be titrated to an optimum level with avoidance of higher levels.


All authors state that they have no conflicts of interest.


The CHAMP Study is funded by the Australian National Health and Medical Research Council (NHMRC Project Grant No. 301916) and the Ageing and Alzheimer's Research Foundation (AARF). The funding organizations were independent of the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Many thanks to Melisa Litchfield for coordinating CHAMP, to scientists Lynley Robinson and Beverly White for assessing the DXA scans, and to the participants who have graciously given their time for this study.

Authors' roles: Study concept and design: KB, RC, VN, and MS. Acquisition of data: KB. Analysis and interpretation of data: KB, RC, VN, and MS. Drafting of the manuscript: KB. Critical revision of the manuscript for important intellectual content and approving final version: KB, RC, VN, MS, FB, D le C, DH, and LW. Statistical analysis: KB. Obtained funding: RC, D le C, DH, and LW. KB had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.