Low serum vitamin D is associated with increased mortality in elderly men: MrOS Sweden
In elderly man, low serum 25-hydroxyvitamin D (25(OH)D) was associated with a substantial excess risk of death compared to 25(OH)D values greater than 50–70 nmol/l, but the association attenuated with time.
The aim of the present study was to determine whether poor vitamin D status was associated with an increase in the risk of death in elderly men.
We studied the relationship between serum 25(OH)D and the risk of death in 2,878 elderly men drawn from the population and recruited to the MrOS study in Sweden. Baseline data included general health and lifestyle measures and serum 25(OH)D measured by competitive RIA. Men were followed for up to 8.2 years (average 6.0 years).
Mortality adjusted for comorbidities decreased by 5% for each SD increase in 25(OH)D overall (gradient of risk 1.05; 95% confidence interval 0.96–1.14). The predictive value of 25(OH)D for death was greatest below a threshold value of 50–70 nmol/l, was greatest at approximately 3 years after baseline and thereafter decreased with time.
Low serum 25(OH)D is associated with a substantial excess risk of death compared to 25(OH)D values greater than 50–70 nmol/l, but the association attenuates with time. These findings, if causally related, have important implications for intervention in elderly men.
KeywordsComorbidity Interaction with time Mortality Population studies Serum vitamin D Spline Poisson regression model
The pleotropic activity of vitamin D has been investigated for several decades. With regard to musculoskeletal metabolism, vitamin D was associated to reduction of falls  and nonvertebral fracture . In addition, vitamin D may contribute to blood pressure regulation  and low serum levels of 25-hydroxyvitamin D (25(OH)D) have been associated with a wide variety of vascular events [4, 5, 6, 7] and with a cognitive decline . Vitamin D is also suggested to have a beneficial effect on the immune response to infectious agents . It may inhibit bacterial growth  and may be associated with a lower risk of viral and bacterial infections [11, 12]. There is continuing uncertainty about these potential non-skeletal benefits of vitamin D, however, and further research is certainly required .
Several studies have described a relationship between circulating 25(OH)D levels, widely regarded as the optimal index of vitamin D status, and the risk of death [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28]. Most studies suggest a continuous relationship between decreasing values of 25(OH)D and an increased risk of death, but in others a threshold effect or a U-shaped relationship has been described in that both low and high D vitamin levels are associated with higher mortality than values in the midrange [17, 20]. In a recent study, no significant relationship between 25(OH)D and overall mortality was found in the US stratum of MrOS . The adjusted death hazard ratio per unit deviation decrease in 25(OH)D (gradient of risk; GR) was 1.01/SD of serum 25(OH)D (95% CI = 0.89–1.14) with an average follow-up of 7.3 years. In the Third National Health and Nutrition Examination Survey (NHANES III) aged more than 60 years of age, Ginde et al.  reported that the hazard ratio was 0.95 per 10 nmol/l (95% CI = 0.92–0.98) adjusted for potential confounders and intermediate variables. They also reported that the association appeared linear (the logarithm of the hazard function) within the range of the data. Wang et al. studied the relationship between 25(OH)D and cardiovascular disease and reported a non-linear relationship .
Uncertainties over the relationship between vitamin D and clinical endpoints are reflected in recommendations for threshold values of vitamin D for the optimisation of general health. The Institute of Medicine in the USA and Standing Committee of European Doctors in Europe recommend a serum value for 25(OH)D of 50 nmol/l [13, 30] whereas the International Osteoporosis Foundation (IOF) in a recent position  reported that a majority of IOF Working Group members considered that 75 nmol/l was the appropriate target level of serum 25(OH)D.
In order to justify threshold values, the relationship between vitamin D and health outcomes is optimally analysed in a way that allows for non-linearity of the relationship. In a preliminary analysis, we have previously shown that vitamin D and death risk are significantly associated with a GR of 1.20 (95% CI 1.07–1.34) in a multivariable setting, with an average follow-up of 4.5 years in the MrOS Sweden study . To further examine our hypothesis that the relationship between 25(OH)D and death (the logarithm of the hazard function) is non-linear and is therefore not appropriately described by a simple gradient of risk, we conducted a more in-depth analysis using longer follow-up and thus more deaths. The aim of the present study, therefore, was to investigate the relationship between 25(OH)D and mortality in elderly men between the ages of 70–81 years in Sweden, allowing for non-linearity as well as any interaction with time since measurement.
MrOS is a multi-centre, prospective cohort study of elderly men in Sweden, Hong Kong and the USA . The present study was from the Swedish stratum of MrOS (n = 3014) recruited at medical centres in Gothenburg (n = 1010), Malmö (n = 1005) and Uppsala (n = 999). Details have been described previously [34, 35, 36]. In brief, men aged 70–81 years were randomly identified using national population registers. To be eligible for the study, they had to be able to walk without aids, provide self-reported data and give signed informed consent. There were no other exclusion criteria. The participation rate in MrOS Sweden was 45%. In the present report, the baseline data in MrOS Sweden was used together with follow-up for death.
Areal BMD was measured at the total hip with calibrated scanners: Lunar Prodigy DXA (GE Lunar Corp., Madison, WI, USA) in Malmö and Uppsala and Hologic QDR 4500/A-Delphi (Hologic, Bedford, MA, USA) in Gothenburg. DXA measurements performed with equipment from different manufacturers were converted to a standardized BMD, as previously described [34, 35, 36, 37].
Height (in centimetres) and weight (in kilograms) were measured, and BMI was calculated as kilograms per square metre. Systolic blood pressure was measured in millimetres of mercury. The international MrOS questionnaire  collected information about current smoking, physical activity, number and type of medications, self-estimated general health, level of education, fracture history, family history of hip fracture, history of diseases (i.e. has a doctor told you that you have rheumatoid arthritis, hypertension, cancer, stroke, myocardial infarction, diabetes or angina?) and the use of alcohol. Physical activity was quantified using parts of the questions in the Physical Activity Scale for Elderly . The patients were asked to estimate the amount of time spent in the past week in activity categories: sitting, walking, light sport and recreation (bowling, fishing, boules), moderate sport (dancing, hunting, skating, golf), heavy sport and recreation (running, swimming, cycling, tennis, aerobics, skiing) and heavy training (weight training, push-ups). The relevant activity was categorised: never (0), seldom 1–2 days (1), sometimes 3–4 days (2) and often 5–7 days (3). The participants also reported if they had been doing outdoor activity (sweeping the yard, mowing the lawn, raking leaves, shovelling snow, or chopping wood) or gardening the past week.
At the first visit, participants brought their current prescription medication to the clinic where study staff recorded the name of all medication. The list of medications was searched to determine any use of D vitamin supplements. Only drugs listed in the Swedish Catalogue of registered pharmaceutical preparations (Farmaceutiska Specialiteter i Sverige) were included in the list of medications. General health was self-estimated as “compared to other people of your own age, how would you describe your own health?: very good (1), good (2), fair (3), bad (4) or very bad (5)”. Level of education was coded: 7 years education as 3, 9 years education as 4, upper secondary school as 6 and university as 8. The participants were also asked if they lived alone and if they were born in Sweden. Use of alcohol was expressed as three or more glasses of alcohol-containing drinks per day, calculated from the frequency and amount of alcohol use. Total calcium and vitamin D intake (from foods and supplements) in the past 12 months were estimated using a modified block food frequency questionnaire (Block Dietary Data Systems, Berkeley, CA), expressed in milligrams per day.
All participants were followed up in the National Cause of Death Register until the end of 2009. This register comprises records of all deaths in Sweden and is more than 99% complete. Cause of mortality was registered using International Classification of Disease codes, based on the information from death certificates, from which the underlying cause of death was determined for each subject. Emigrants were followed up to the day of emigration. In order to determine any effect of the length of follow-up, two follow-up analyses were performed. The men were followed up to the 31 December 2009.
Correlation tests between 25(OH)D and other variables used nonparametric Pitman's permutation test . Linear regression was used to adjust for other variables.
A special Poisson regression model was used to study the relationship between age, the time since baseline, the z-score of 25(OH)D, other covariates on the one hand and on the other hand, the risk of death. In contrast to logistic regression, the Poisson regression utilises the length of each individual's follow-up period and the hazard function is assumed to be exp(β0 + β1∙current time from baseline + β2∙current age + β3∙current variable of interest). The observation period of each patient was divided in intervals of 1 month. A similar approach was used to examine other predictors of mortality, and a final multivariable model was constructed to determine predictors, which had an independent contribution to risk. A forward stepwise manner was used, choosing the variable with the smallest p value. All variables that had a univariate p value less than 0.05, for the association with death, were candidates for the multivariable analysis.
A spline Poisson regression model was fitted using knots at the 10th, 50th and 90th percentiles, as recommended by Harrell , of the z-score of 25(OH)D to study the association between serum 25(OH)D and mortality in more detail. The splines were second-order functions between the breakpoints and linear functions at the tails resulting in a smooth curve.
According to the findings in the analysis, a model with a quadratic function in the interaction between the z-score of 25(OH)D and time was used (a∙current time + b∙current time2). An interaction between the spline coefficients and time since baseline was also used. In order to minimize the number of β-coefficients when using this complex model, the variables that did not significantly correlate with the z-score of 25(OH)D were deleted from the model, since they would not influence the association between serum 25(OH)D and mortality. The hazard function was exp(β1 + β2∙current age + β3∙current time since baseline + β4∙general health + β5∙TH BMD + β6∙diabetes + β7∙outdoor activity + β8∙physical activity walking + β9–β11∙spline coefficient∙exp(a∙current time since baseline + b∙current time2)). The parameters β0–11, a and b were determined by the maximum likelihood method. The parameters a and b determined the time to the optimum influence of vitamin D.
The difference between the likelihood values of two models, of which one is included in the other one, has a χ2 distribution with the degrees of freedom equal to the difference of the number of parameters. This methodology was used to compare the model above and a simpler one that assumed a linear GR and no time interaction. Two-sided p values were used for all analyses and p < 0.05 considered to be significant
Characteristics of the MrOS study population by intervals of 25(OH)D
75.7 ± 3.4
75.4 ± 3.2
75.4 ± 3.2
75.5 ± 3.1
75.7 ± 3.1
82.7 ± 17.8
82.4 ± 13.4
81.4 ± 12.3
79.6 ± 10.7
77.2 ± 10.7
26.9 ± 5.4
27.0 ± 4.0
26.6 ± 3.5
26.1 ± 3.3
25.5 ± 3.4
Born outside Sweden
General health (1–5)
2.5 ± 0.7
2.3 ± 0.8
2.0 ± 0.8
2.0 ± 0.7
2.0 ± 0.8
Physical activity (walking), 0–3
2.0 ± 1.1
2.3 ± 0.9
2.4 ± 0.8
2.4 ± 0.8
2.4 ± 0.8
Physical activity moderate (0–3)
0.2 ± 0.7
0.3 ± 0.6
0.4 ± 0.7
0.4 ± 0.7
0.5 ± 0.8
Physical activity heavy (0–3)
0.2 ± 0.5
0.4 ± 0.8
0.6 ± 0.9
0.6 ± 0.9
0.6 ± 0.9
Outdoors activity during last week
Gardening during last week
Number of medications
2.0 ± 3.0
1.6 ± 2.6
1.2 ± 2.2
1.1 ± 2.2
1.1 ± 2.1
Fatty fish (standard portions per week)
1.3 ± 1.9
1.4 ± 3.2
1.3 ± 3.1
1.8 ± 11.3
2.2 ± 15.3
Calcium intake per day (mg)
781 ± 391
855 ± 473
910 ± 439
905 ± 563
963 ± 688
Vitamin D supplements
Total hip BMD (g/cm2)
0.88 ± 0.13
0.92 ± 0.15
0.94 ± 0.15
0.94 ± 0.14
0.93 ± 0.15
Systolic blood pressure (mmHg)
145 ± 23
146 ± 22
145 ± 21
145 ± 20
145 ± 21
Winter (Nov to April) when measuring 25(OH)D
−2.1 ± 0.3
−1.1 ± 0.4
−0.3 ± 0.4
0.6 ± 0.4
1.9 ± 0.7
Number of deaths during follow-up
Death risk per year
A low 25(OH)D (original value) was correlated with higher weight (p < 0.001), higher BMI (p < 0.001), more medications (p < 0.001), poorer general health (p < 0.001), lower physical activity (p < 0.001), diabetes (p < 0.001), winter measurement (p < 0.001), living alone (p < 0.001), born outside Sweden (p < 0.001), less outdoor activity (p < 0.001), less garden work (p < 0.001) and lower calcium intake (p = 0.0066). None of the variables above accounted for the differences in 25(OH)D between centres.
Mortality during follow-up
Two thousand eight hundred seventy-eight men aged 70–81 years had been followed up for up to 8.2 years, with an average follow-up of 6.0 years. Of these, 577 men (20.0%) had died (all cause mortality). Serum 25(OH)D was significantly lower in men who subsequently died compared with men who remained alive during the follow-up.
In a multivariable analysis for predictors of mortality, current age, current time since baseline, total hip BMD, past history of cancer, angina and diabetes, outdoor activity, physical activity walking, number of medication and general health were significantly associated with overall mortality. None of the other variables listed above had any significant independent effect, including the z-score for serum 25(OH)D (GR = 1.05; 95% CI = 0.96–1.14). Total hip BMD was a more marked independent determinant of death than serum 25(OH)D. The GR for 1 standard deviation decrease in BMD was 1.22 (95% CI = 1.12–1.33).
Time since baseline measurement
In order to minimize the number of β-coefficients since the interaction with time is needed, the variables cancer, angina and number of medications were deleted from the model, since they were not significantly correlated with 25(OH)D. The association between 25(OH)D and mortality remained the same with the more simple model (data not shown). When comparing the fit of the models with and without allowing for non-linearity and interaction with time, the former (shown in Fig. 3) had a significantly better fit using the maximum likelihood test (p = 0.0043). When just spline functions with no interaction with time since baseline were added to the simple model, there was no significant improvement in fit (p > 0.30), because the relationship between 25(OH)D was still obscured by the long duration of follow-up.
In this prospective study of older men, we describe a significant inverse relation between serum 25(OH)D and all cause mortality using a univariate analysis. When the relationship was investigated by multivariable analysis, a significant relationship no longer was evident. The association between vitamin D and mortality was, however, confounded by a longer duration of follow-up and the assumption of a linear relationship between death and vitamin status. Thus, when we took these variables into account, a non-linear relationship became evident such that low serum values were associated with increased short-term mortality up to a value of about 60 nmol/l, but serum values above this were not associated with improved survival. Models assuming this non-linearity and interaction with time since baseline provided a significantly better fit than a model that assumed a simple gradient of risk.
Our findings may partly explain the disparate conclusions reported in previous studies that have examined the relationship between vitamin D status and death. Whereas the finding of a simple gradient of risk is consistent with our results [18, 19, 26], so too is the absence of a significant effect, particularly after multivariable analysis [14, 15, 21, 27, 28] or after a follow-up that extends beyond about 5 years [14, 21, 27]. Similarly, the reporting of a non-linear relationship in some studies [17, 20] is not inconsistent with the finding of a linear relationship in others , particularly where time to follow-up has not been taken into account. Only one previous analysis has formally taken time since follow-up into account ; they showed that the predictive power appeared after 3 years. A recent report in abstract form  noted that a significant association between low 25(OH)D and death risk in elderly women was stronger after 5 years than after 10 years of follow-up. These considerations indicate that time dependency is an important aspect of the future research agenda.
The waning of effect of a risk variable with time might at first sight seem counterintuitive, but on closer consideration is not unexpected. Given the inevitability of death, all risk factors for death will have no predictive value with an infinite follow-up time. With shorter observation times, a waning of effect may be due to shorter term heterogeneity in the natural history of disease. A good example in the field of bone disease is that the predictive value of BMD tests for fracture risk wanes with time . The loss of predictive value is much greater than the precision errors of the BMD measurement and arises because individuals lose bone mineral with time at variable rates. In the case of BMD, the loss of predictive value with time is modest but is more marked in the case of some of the biochemical markers of bone turnover [43, 44]. In one of these studies with a 5-year follow-up period, the ratio of carboxylated to total serum osteocalcin was a very strong risk factor for fracture (HR = 5.32; 95% CI = 3.26–8.68) over the period of observation, but the predictive value of the maker lasted 3 years . For normally distributed variables, the long-term predictive value can be derived from the correlation coefficient between repeated samples over a defined interval of time . Another example of predictive power changing with time is the association between HbA1c and progression of retinopathy in type 1 diabetes. HbA1c up to 5 years earlier made a greater contribution than current values . In the case of 25(OH)D, there are few studies that have examined correlation coefficients over long time intervals, though one study found a correlation coefficient of 0.7 between values measured 3 years apart.
The work presented in this paper has a number of strengths and weaknesses. A strength is the detail of the baseline assessment so that we were able to examine a large number of potential confounders and adjust for these where appropriate. Measurement errors were also modulated by adjusting baseline serum 25(OH)D for season. There are also several limitations. The participation rate (45%) will likely cause a healthy selection bias. A biased sample is expected to converge towards the general population over time and was shown in the present study in that an increasing time of follow-up was associated with an increasing risk of death even after adjustment for age (p < 0.001). By the inclusion of time since baseline assessment, we have in part adjusted for this effect. The measurement of 25(OH)D by radioimmunoassay has lower accuracy than measurements made by mass spectrometry [47, 48], but errors of measurement are likely to affect both cases and controls and attenuate rather than strengthen any association. A significant limitation is the narrow age range of men that were recruited to MrOS (70–81 years), so that our findings may not be extrapolated to younger ages. Finally, mean serum 25(OH)D differed between centres, despite the use of the same methodology and the use of a single laboratory for the analysis. We were unable to find any explanatory variables in multivariable analysis and was the reason for z-score transformation of serum 25(OH)D.
A further strength of this paper is the analytical treatment of potential confounding factors. When studying the relationship between two variables as, e.g. vitamin D and death, there are several ways to take other variables into account. The reason for including other variables is either to adjust for confounding or to decrease measurement error, which means that the relationship will be strengthened and the statistical power larger. The variables, which we want to take into account, could be included in a regression analysis or could be the base for a division of the material into more homogenous groups. By the Mantel-Haenszel procedure, the results from the subgroups could then be pooled to a test. A third method is to determine a z-score depending on variables, which we want to take into account. The first mentioned method is useless if the variable, e.g. season, is related only to one of the two main variables, and the purpose is to reduce the measurement error. In fact we have applied all the three types of methods in this study (though the Mantel-Haenszel procedure was not applied).
A principal aim of the present study was to determine whether there was a threshold effect of vitamin D status on the risk of death. By allowing for a potential threshold effect in the analytical framework, we were able to demonstrate a threshold for above which mortality no longer decreased with increasing values. The threshold varies on season for measuring D vitamin. In the present paper, we back transformed our data using the values in May (150 days from 1 January). Under these assumptions, a threshold of about 60 nmol/l was apparent but other assumptions would suggest a tolerance interval of 50–70 nmol/l. This is comfortably consistent with the several previews and position statements on target values for serum 25-OH D in the elderly population [30, 31, 49, 50].
The clinical significance of our findings should be cautiously interpreted. The finding of an association of this kind is not evidence for causality, irrespective of the strength of the association. Rather, the finding is hypothesis generating. Notwithstanding, if the association were causal, this would have marked clinical consequences.
We conclude that there is a significant association between low serum 25(OH)D levels and mortality in elderly men. The association is non-linear in that the effect is not evident above a threshold value for 25(OH)D. Threshold values lie in the range of 50–70 nmol/l, consistent with consensus views on vitamin D nutrition. The attainment of target values of serum 25(OH)D levels could have marked clinical benefits but need to be tested in studies of efficacy. In addition the association wanes with time. The time effect is very important to consider when analysing the relationship, since it could both hide and create an association. If a non-linear relationship is confirmed in other studies, where both sexes and other ages are represented, we could obtain a threshold of the 25(OH)D to be recommended as a minimum level.
Conflicts of interest
JAK and EVM consult for a large number of companies involved in skeletal metabolism but have no competing interest to declare in relation to the context of this paper.
- 2.Bischoff-Ferrari HA, Willett WC, Wong JB, Stuck AE, Staehelin HB, Orav EJ, Thoma A, Kiel DP, Henschkowski J (2009) Prevention of nonvertebral fractures with oral vitamin D and dose dependency: a meta-analysis of randomized controlled trials. Arch Intern Med 169(6):551–561. doi:10.1001/archinternmed.2008.600 PubMedCrossRefGoogle Scholar
- 9.Liu PT, Stenger S, Li H, Wenzel L, Tan BH, Krutzik SR, Ochoa MT, Schauber J, Wu K, Meinken C, Kamen DL, Wagner M, Bals R, Steinmeyer A, Zugel U, Gallo RL, Eisenberg D, Hewison M, Hollis BW, Adams JS, Bloom BR, Modlin RL (2006) Toll-like receptor triggering of a vitamin D-mediated human antimicrobial response. Science 311(5768):1770–1773. doi:10.1126/science.1123933 PubMedCrossRefGoogle Scholar
- 13.Ross AC, Manson JE, Abrams SA, Aloia JF, Brannon PM, Clinton SK, Durazo-Arvizu RA, Gallagher JC, Gallo RL, Jones G, Kovacs CS, Mayne ST, Rosen CJ, Shapses SA (2011) The 2011 report on dietary reference intakes for calcium and vitamin D from the Institute of Medicine: what clinicians need to know. J Clin Endocrinol Metab 96(1):53–58. doi:10.1210/jc.2010-2704 PubMedCrossRefGoogle Scholar
- 16.Dobnig H, Pilz S, Scharnagl H, Renner W, Seelhorst U, Wellnitz B, Kinkeldei J, Boehm BO, Weihrauch G, Maerz W (2008) Independent association of low serum 25-hydroxyvitamin d and 1,25-dihydroxyvitamin d levels with all-cause and cardiovascular mortality. Arch Intern Med 168(12):1340–1349. doi:10.1001/archinte.168.12.1340 PubMedCrossRefGoogle Scholar
- 20.Michaelsson K, Baron JA, Snellman G, Gedeborg R, Byberg L, Sundstrom J, Berglund L, Arnlov J, Hellman P, Blomhoff R, Wolk A, Garmo H, Holmberg L, Melhus H (2010) Plasma vitamin D and mortality in older men: a community-based prospective cohort study. Am J Clin Nutr 92(4):841–848. doi:10.3945/ajcn.2010.29749 PubMedCrossRefGoogle Scholar
- 21.Cawthon PM, Parimi N, Barrett-Connor E, Laughlin GA, Ensrud KE, Hoffman AR, Shikany JM, Cauley JA, Lane NE, Bauer DC, Orwoll ES, Cummings SR (2010) Serum 25-hydroxyvitamin D, parathyroid hormone, and mortality in older men. J Clin Endocrinol Metab 95(10):4625–4634. doi:10.1210/jc.2010-0638 PubMedCrossRefGoogle Scholar
- 24.Semba RD, Houston DK, Bandinelli S, Sun K, Cherubini A, Cappola AR, Guralnik JM, Ferrucci L (2010) Relationship of 25-hydroxyvitamin D with all-cause and cardiovascular disease mortality in older community-dwelling adults. Eur J Clin Nutr 64(2):203–209. doi:10.1038/ejcn.2009.140 PubMedCrossRefGoogle Scholar
- 28.Ensrud KE, Blackwell TL, Cauley JA, Cummings SR, Barrett-Connor E, Dam TT, Hoffman AR, Shikany JM, Lane NE, Stefanick ML, Orwoll ES, Cawthon PM (2011) Circulating 25-hydroxyvitamin D levels and frailty in older men: the osteoporotic fractures in men study. J Am Geriatr Soc 59(1):101–106. doi:10.1111/j.1532-5415.2010.03201.x PubMedCrossRefGoogle Scholar
- 30.Vitamin D nutritional policy in Europe (2009) Comité Permanente des Médecins Européen, http://cpme.dyndns.org:591/adopted/2009/CPME_AD_Brd_241009_179_final_EN.pdf. Accessed 24 Oct 2009
- 32.Johansson H, Oden A, Kanis J, McCloskey E, Lorentzon M, Ljunggren O, Karlsson MK, Tivesten A, Barrett-Connor E, Ohlsson C, Mellstrom D (2010) Low vitamin D is related to increased risk of death in elderly men. MrOS Sweden. J Bone Miner Res 25(Suppl 1):S6(1019)Google Scholar
- 33.Orwoll E, Blank JB, Barrett-Connor E, Cauley J, Cummings S, Ensrud K, Lewis C, Cawthon PM, Marcus R, Marshall LM, McGowan J, Phipps K, Sherman S, Stefanick ML, Stone K (2005) Design and baseline characteristics of the osteoporotic fractures in men (MrOS) study—a large observational study of the determinants of fracture in older men. Contemp Clin Trials 26(5):569–585. doi:10.1016/j.cct.2005.05.006 PubMedCrossRefGoogle Scholar
- 34.Mellstrom D, Vandenput L, Mallmin H, Holmberg AH, Lorentzon M, Oden A, Johansson H, Orwoll ES, Labrie F, Karlsson MK, Ljunggren O, Ohlsson C (2008) Older men with low serum estradiol and high serum SHBG have an increased risk of fractures. J Bone Miner Res 23(10):1552–1560PubMedCrossRefGoogle Scholar
- 36.Johansson H, Oden A, Kanis J, McCloskey E, Lorentzon M, Ljunggren O, Karlsson MK, Orwoll E, Tivesten A, Ohlsson C, Mellstrom D (2010) Low bone mineral density is associated with increased mortality in elderly men: MrOS Sweden. Osteoporos Int. doi:10.1007/s00198-010-1331-1
- 39.Good P (ed) (2000) Permutation tests. Springer Verlag, New YorkGoogle Scholar
- 40.Harrell FJ (2001) General aspects of fitting regression models: regression modeling strategies. Springer Science + Business Media, Inc, New YorkGoogle Scholar
- 41.Hochberg MC, Lui L, Black DM, Cauley J, Cawthon PM, Ensrud K, Fredman L, Hillier TA, Cummings S (2010) Serum 25-hydroxyvitamin D levels. Mortality and risk of non-spine and hip fractures in older white women: data from the study of osteoporotic fractures. J Bone Miner Res 25(Suppl 1):s201(SA0361)Google Scholar
- 43.Luukinen H, Kakonen SM, Pettersson K, Koski K, Laippala P, Lovgren T, Kivela SL, Vaananen HK (2000) Strong prediction of fractures among older adults by the ratio of carboxylated to total serum osteocalcin. J Bone Miner Res 15(12):2473–2478. doi:10.1359/jbmr.2000.15.12.2473 PubMedCrossRefGoogle Scholar