Skip to main content
NCBI home page
UKPMC Funders Author Manuscripts logoLink to UKPMC Funders Author Manuscripts
. Author manuscript; available in PMC: 2020 Mar 1.
Published in final edited form as: Bone. 2018 Oct 12;126:51–58. doi: 10.1016/j.bone.2018.10.011

Mendelian randomization in the bone field

Susanna C Larsson a,*, Karl Michaëlsson b, Stephen Burgess c,d
PMCID: PMC6697532  EMSID: EMS83967  PMID: 30321669

Abstract

Identification of causative risk factors amenable for modification is essential for the prevention and treatment of osteoporosis. Observational studies have identified associations between several potentially modifiable risk factors and osteoporosis. However, observational studies are susceptible to confounding, reverse causation bias, and measurement error, all of which limit their ability to provide causal estimates of the effect of exposures on outcomes, thereby reducing their ability to inform prevention and treatment strategies against bone loss and fractures. In addition, not all risk factors are suitable for an analysis in a randomized clinical trial. Mendelian randomization is a genetic epidemiological method that exploits genetic variants as unbiased proxies for modifiable exposures (e.g., biomarkers, adiposity measures, dietary factors, and behaviors) to determine the causal relationships between exposures and health outcomes. This technique has been used to provide evidence of causal associations of serum estradiol concentrations, smoking, body mass index, and type 2 diabetes with bone mineral density and the lack of associations of serum thyroid stimulating hormone, urate, C-reactive protein, and 25-hydroxyvitamin D concentrations with bone mineral density in generally healthy populations. This review will briefly explain the concept of Mendelian randomization, the advantages and potential limitations of this study design, and give examples of how Mendelian randomization has been used to investigate questions relevant to osteoporosis.

Keywords: Bone mineral density, Fracture, Genome-wide association studies, Mendelian randomization, Osteoporosis, Single nucleotide polymorphisms

1. Introduction

Osteoporosis is a multifactorial disorder influenced by genetic predisposition, age, sex, and modifiable exposures [1]. Identification of causative risk factors amenable for modification is essential for the prevention and treatment of osteoporosis. Observational epidemiological studies have identified associations of several potentially modifiable risk factors such as sex-steroid deficiency, low body mass index (BMI), physical inactivity, smoking, heavy alcohol consumption, and low calcium, vitamin D, and antioxidant intake with bone mineral density (BMD) and fracture risk [1]. However, observational studies are susceptible to various biases such as confounding (where the association between the risk factor and outcome is driven by the association with another correlated risk factor) and reverse causality (where disease affects the risk factor, and not vice versa), thereby reducing their ability to inform prevention and treatment strategies against bone loss and fractures. Furthermore, regression dilution bias due to measurement error limits their ability to detect associations. Well-conducted randomized controlled trials (RCTs) are the gold standard for inferring causality but RCTs are expensive, resource-intensive, examine mainly short-term exposures (normally < 5 years), and may not be feasible to conduct.

An alternative approach is to use genetic variants as unbiased proxies of the modifiable exposure of interest to allow estimation of the long-term causal effect of the exposure on the outcome, a technique known as Mendelian randomization [2–6]. In this review we will briefly explain the concept of Mendelian randomization, discuss the advantages and limitations of this method, and provide examples of how Mendelian randomization has been used to evaluate questions relevant to osteoporosis.

2. Underlying principles and assumptions

Mendelian randomization is a genetic epidemiologic method that utilizes one or multiple genetic variants, typically single nucleotide polymorphisms (SNPs), robustly associated with the modifiable exposure (e.g., a biomarker, adiposity measure, dietary factor, or behavior) to estimate the causal relationship between the exposure and the health outcome (e.g., a disease, BMD, or fracture risk) [2]. The basis of Mendelian randomization is that if a genetic variant influences the exposure or imitates its biological effects, then the genetic variant should be associated with the outcome to the extent predicted by its association with the exposure and the impact of the exposure on the outcome. Mendelian randomization builds on Mendel's second law or the law of independent assortment, that is, each pair of alleles segregates independently of the other pairs of alleles during gamete formation [8]. This natural randomization processes implies that a genetic variant affecting the level of a specific modifiable exposure will generally be unrelated to other traits (exposures), such as other biomarkers, dietary and lifestyle factors, and socioeconomic status. The random allocation of alleles during conception reduces potential confounding in Mendelian randomization studies in a similar way as in an RCT in which participants are randomly assigned to an exposure group (Fig. 1).

Fig. 1.

Fig. 1

Open in a new tab

Comparison of a randomized controlled trial and a Mendelian randomization study. In a randomized controlled trial, participants are randomly assigned to receive treatment (e.g., estrogen therapy) or placebo, thereby avoiding potential confounding between treatment and other risk factors. A Mendelian randomization study generates a similar scenario. For example, the A allele of rs727479 in the CYP19A1 gene is associated with higher estradiol concentrations, which are related to higher bone mineral density (BMD). Alleles are inherited largely independently of environmental exposures, and individuals who inherit the A allele of rs727479 are assigned to an average higher estradiol concentration than those who inherit the C allele. As in a randomized controlled trial, groups defined by genotype will experience an average difference in exposure to estradiol but no difference in other exposures (confounding factors). Hence, a Mendelian randomization analysis of genotype is similar to an intention-to-treat analysis in a randomized controlled trial.

In order to obtain valid causal estimates in a Mendelian randomization study, the genetic variant (or variants) that serves as a proxy for the exposure must be: (1) reliably associated with the exposure, that is, there should be strong evidence that the genetic variant affects the exposure; (2) unrelated to confounding factors of the exposure-outcome relationship; and (3) associated with the outcome solely through the exposure and not through any other causal pathway (Fig. 2). The first assumption can be tested directly, whereas the other two can be tested but not firmly established.

Fig. 2.

Fig. 2

Open in a new tab

Assumptions underlying a Mendelian randomization study. BMD, bone mineral density; BMI, body mass index.

3. Advantages of the Mendelian randomization design

Mendelian randomization studies have several advantages over conventional observational epidemiological studies. First, confounding is mitigated due to the Mendel's second law of random assortment of alleles. Second, reverse causality is prevented because genetic variants are fixed at conception and cannot be affected by disease processes. This further mitigates against confounding, as genetic variants cannot be influenced by confounding factors that operate after conception. Third, regression dilution bias due to measurement error is avoided because genetic variants are measured with high precision. A further advantage is that genetic differences in exposure can reflect the effect of lifelong exposure on the outcome. Hence, Mendelian randomization analyses can provide unbiased estimates of exposure-outcome relationships, provided the fundamental assumptions are fulfilled (Fig. 2).

4. Limitations of the Mendelian randomization design

Although Mendelian randomization studies are profoundly less prone to confounding than conventional observational epidemiological studies, confounding of the genotype-outcome relationship can occur due to linkage disequilibrium, population stratification, and pleiotropy.

Linkage disequilibrium is the non-random association of alleles at different genetic loci. Loci that are close on the chromosome tend to be inherited together. Confounding can occur if the genetic variant that is used as proxy for the exposure is in linkage disequilibrium with a gene that directly affects the outcome under study.

Population stratification arises when allele frequencies and exposure or outcome distributions vary substantially between different subgroups of the population and produce an association between the exposure and outcome at the overall population level that is not present in any of the subgroups. This bias can be mitigated by restriction of the analysis to ethnically homogenous groups. For example, the frequency of the genetic variant underlying lactase persistence (i.e., maintenance of lactase expression after weaning) differs widely across populations. Hence, a Mendelian randomization study on milk consumption, using the genetic variant related to lactase persistence as an instrumental variable, could potentially be biased by population stratification if ancestry is not accounted for.

Pleiotropy refers to the phenomenon in which a single locus affects multiple phenotypes [9]. Depending of type, pleiotropy may be unproblematic or lead to confounded estimates. One type of pleiotropy is ‘vertical pleiotropy’ (also known as ‘mediated pleiotropy’ or ‘type II pleiotropy’ [5]), which refers to association with other phenotypes downstream the exposure of interest. MR analyses exploit vertical pleiotropy, as they assume that associations of the genetic variants with the outcome or any other variable are due to the causal pathway through the exposure. For example, genetic variants that influence BMI may also be associated with downstream phenotypes (mediators) such as estradiol and glucose concentrations and type 2 diabetes, which are associated with higher BMD [10,11]. These associations are expected and would not invalidate the results from a Mendelian randomization study of the association between BMI and BMD. However, ‘horizontal pleiotropy’ (also known as ‘biological pleiotropy’ or ‘type I pleiotropy’ [5]), which arises when a genetic variant associates with more than one phenotype on separate pathways, can invalidate the results from Mendelian randomization analyses. Whether pleiotropy may be of concern can be explored by assessing the association between the genetic variant(s) and potential confounders. Since not all genetic variants used as proxies for the exposure of interest may fulfill the ‘no pleiotropy’ assumption, several approaches have been developed to detect and adjust for pleiotropy, such as the MR-Egger [12,13], MR-PRESSO [14], and multivariable MR methods [15,16] (Box 1).

Box 1. MR methods to explore and adjust for pleiotropy.

MR-Egger regression. This method can (1) test for directional pleiotropy, (2) test for a causal effect, and (3) provide an estimate of the causal effect of the exposure on the outcome. While the standard inverse-variance weighted method for Mendelian randomization assumes that all genetic variants fulfill the instrumental variable assumptions, the MR-Egger method can test whether the genetic variants have pleiotropic effects on the outcome that differ on average from zero (directional pleiotropy). Provided that pleiotropic effects of genetic variants are uncorrelated with the associations of genetic variants with the exposure, MR-Egger can consistently estimate the causal effect even when genetic variants are pleiotropic. However, MR-Egger estimates typically have low precision, and can be strongly influenced by outlying genetic variants.

MR-PRESSO (Mendelian randomization pleiotropy residual sum and outlier). This method can (1) identify horizontal pleiotropy, (2) adjust for horizontal pleiotropy via outlier removal, and (3) test the significant differences in the causal estimates before and after removal of outliers. The method is best used to identify inconsistencies between genetic associations for different genetic variants, and removing outlying genetic variants. However, if the method removes several variants, then the user should be suspicious about whether consistency in the remaining variants is really meaningful.

Multivariable MR analysis uses multiple genetic variants associated with two or more measured risk factors to simultaneously estimate the causal effect of each risk factor on the outcome. This approach is similar to the concurrent evaluation of several treatments in a factorial randomized controlled trial. The method is most useful when there are related risk factors with shared genetic predictors, but it is difficult to find specific genetic predictors of each individual trait.

Another shortcoming is that genetic variants generally have small effects on the exposure (i.e., explain only a small proportion of the variance) and therefore large sample sizes are necessary to obtain statistically significant results. The power can be increased by the use of multiple genetic variants (explaining more of the variance in the exposure), which can be combined into a weighted genetic risk score. A further potential limitation is canalization, which refers to compensatory processes during development that can atone for disrupting environmental or genetic forces and mitigate the results. Furthermore, since genetic variants reflect lifelong exposure, the MR approach cannot be used to assess whether an exposure at a certain induction period in the life course is related to the outcome. Finally, Mendelian randomization can only be used if one or more suitable genetic variants associated with the exposure of interest are available.

5. Estimating the causal effect of exposure on outcome

Mendelian randomization studies can make use of summary-level data from genome-wide association studies (GWAS). Publicly available summary-level data for BMD and fracture are provided by the GEnetic Factors for OSteoporosis Consortium (GEFOS) (http://www.gefos.org/). The conventional and easiest way to estimate the causal effect of an exposure on an outcome in a Mendelian randomization study based on summary-level data is the ratio method where the coefficient for the effect of the genetic variant on the outcome is divided by the coefficient for the effect of the genetic variant on the exposure (see example in Fig. 3). This method can be used for a single genetic variant or multiple genetic variants in combination. Other approaches for estimating causal associations in Mendelian randomization studies have been described in detail elsewhere [13,17–21].

Fig. 3.

Fig. 3

Open in a new tab

Example of a Mendelian randomization study assessing the causal association between body mass index (BMI) and bone mineral density (BMD). In this example, the genetic variant (single-nucleotide polymorphism, SNP) rs1558902 in the FTO gene is used as proxy for BMI. The A-allele of rs1558902 is associated with 0.02 g/cm2 higher BMD (p = 1.8 × 10−9) [36] and 0.08 standard deviation (SD) higher BMI (p = 7.5 × 10−153) in individuals of European ancestry [37]. The causal estimate is derived by dividing the beta coefficient for the SNP–outcome association with the coefficient for the SNP–exposure association: 0.02/0.08 = 0.25, i.e., BMD is increased by 0.25 g/cm2 per SD increase in genetically predicted BMI.

6. Mendelian randomization studies in the bone field

Over the past decade, GWAS have made an important contribution to the identification of genetic variants associated with numerous potential risk factors for health-related outcomes. The GWAS results have facilitated the use of Mendelian randomization to evaluate causal relationships between modifiable exposures and outcomes. This has resulted in a growing number of Mendelian randomization studies.

We searched PubMed until August 2018 for Mendelian randomization studies assessing the causal role of potentially modifiable risk factors for osteoporosis. We used the search terms “Mendelian randomization” combined with “bone mineral density”, “fracture” or “osteoporosis” and identified several studies of which all but one were performed during the last 2–3 years [10,11,22–31] (Table 1). The Mendelian randomization approach has been used to show that increased serum estradiol concentrations have a causal effect on increasing BMD [10] and that decreased BMD, earlier menopause, and late puberty are associated with increased fracture risk [31]. The use of multiple SNPs in Mendelian randomization analyses has also established that increased BMI are associated with increased BMD in children [25]; that type 2 diabetes and higher fasting glucose concentrations are associated with increased BMD [11] but not with reduced fracture risk in adults [31]; that decreased grip strength is associated with an increased fracture risk [31]; and that smoking is associated with decreased BMD [28]. Furthermore, the Mendelian randomization method has been used to show that genetically higher serum 25‑hydroxyvitamin D concentrations are not associated with increased BMD [10,24] or fracture risk [31] in generally healthy populations, and that serum thyroid stimulating hormone [29,31], homocysteine [31], urate [23], and C-reactive protein (an inflammatory marker) [22,26] concentrations are not associated with BMD or fracture risk. A recent Mendelian randomization study showed no association between genetically higher alcohol consumption and BMD [28]. Other Mendelian randomization studies found no association of genetically predicted milk consumption, using an SNP (rs4988235) located upstream from the lactase gene as an instrumental variable, with BMD [27,30] or fracture risk [30,31]. However, the lactase persistence genotype has been found to be associated with higher BMI [27,32] and height [33] and with lower fruit and vegetable consumption [33]. No association has been observed between genetic predisposition to type 1 diabetes, coronary artery disease, rheumatoid disease, and inflammatory bowel disease and risk of fracture [31]. While those diseases are unlikely causally associated with fracture risk, the MR study design cannot be used to assess whether complications or treatment of the diseases influence fracture risk.

Table 1. Examples of Mendelian randomization studies in the bone field.

Exposure Genetic variantsa Outcome Sample size and data sources for the outcome data Unit Estimate (95% CI) p-Value Interpretation Ref.
FN-BMD 43 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in FN-BMD Fracture: 1.55
(1.48 to 1.63)		 < 0.001 Supports a causal association between decreased FN-BMD and increased fracture risk [31]
LS-BMD 40 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in LS-BMD Fracture: 1.43
(1.37 to 1.50)		 < 0.001 Supports a causal association between decreased LS-BMD and increased fracture risk [31]
Earlier menopause 54 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD change, i.e.,3.9 years earlier menopause Fracture: 1.10
(1.00 to 1.21)		 0.05 Supports a causal association between earlier menopause and increased fracture risk [31]
Late puberty 106 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD change, i.e.,1.4 years late puberty Fracture: 1.06
(1.00 to 1.13)		 0.04 Supports a causal association between late puberty and increased fracture risk [31]
Estradiol 1 BMD 32,965 individuals of European descent, GEFOS Consortium SD change in BMD and g/cm2 in eBMD per 10% increase in estradiol FN-BMD: 0.038
(NA)
LS-BMD: 0.031
(NA)
eBMD: 0.030 (NA)		 4.6 × 10−6
0.001
6.0 × 10×−18 		Supports a causal association between serum estradiol levels and increased BMD [10]
Vitamin D 4 BMD 1824 Chinese postmenopausal women 1 g/cm2 change in BMD per 1 ln-ng/mL increase in 25OHD FN-BMD: −0.04
(−0.13 to 0.03)
LS-BMD: 0.05
(−0.16 to 0.06)
TH-BMD: −0.04
(−0.13 to 0.04)		 0.26
0.38
0.33		 No support for a causal association of increased serum 25OHD levels with increased BMD [24]
5 BMD 32,965 individuals of European descent, GEFOS Consortium and 142,487 individuals from the UK Biobank SD change in BMD and g/cm2 change in eBMD per 1 SD increase in 25OHD FN-BMD: 0.02
(−0.03 to 0.07)
LS-BMD: 0.02
(−0.04 to 0.08)
eBMD: −0.03
(−0.05 to −0.01)		 0.37
0.49
0.02		 No support for a causal association of increased serum 25OHD levels with increased BMD [10]
4 Fracture 264,973 individuals of predominantly European ancestry, GEFOS and GENOMOS consortia and UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in 25OHD Fracture: 0.84
(0.70 to 1.02)		 0.07 No support for a causal association of decreased serum 25OHD levels with increased fracture risk [31]
TSH 20 BMD 32,735 individuals of European descent, GEFOS Consortium SD change in BMD per 1 SD decrease in serum TSH FN-BMD: 0.003
(−0.053 to 0.048)
LS-BMD: 0.010
(−0.069 to 0.049)		 0.92
0.73		 No support for a causal association of serum TSH levels with BMD [29]
20 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in serum TSH levels Fracture: 0.99
(0.94 to 1.04)		 0.78 No support for a causal association of serum TSH levels with fracture risk [31]
Homocysteine 13 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD increase in homocysteine levels Fracture: 0.98
(0.92 to 1.05)		 0.78 No support for a causal association of homocysteine levels with fracture risk [31]
Urate 5 BMD 2501 individuals, Framingham Heart Study 1 g/cm2 change in BMD per 1 mmol/L increase in urate levels TF-BMD: −0.36
(−0.73 to 0.07)
FN-BMD: −0.33
(−0.68 to 0.03)
LS-BMD: 0.04
(−0.39 to 0.46)		 0.06
0.07
0.87		 No support for a causal association of urate levels with BMD [23]
CRP 20 BMD 32,965 individuals of European descent, GEFOS Consortium 1 g/cm2 change in BMD per 1 ln-mg/L increase in CRP levels FA-BMD: −0.02 (NA)
FN-BMD: −0.04 (NA)
LS-BMD: −0.04 (NA)		 0.69
0.22
0.30		 No support for a causal association between CRP levels and BMD [26]
29 Fracture 6386 individuals, The Rotterdam Study OR of fracture per 1 SD increase in CRP levels Fracture: 1.00 (0.99 to 1.00) 0.23 No support for a causal association between CRP levels and fracture risk [22]
Fasting glucose 30 BMD 32,961 individuals of European descent, GEFOS Consortium SD change in BMD per 1 mmol/L increase in fasting glucose levels FN-BMD: 0.13 (0.01 to 0.25)
LS-BMD: 0.08 (−0.04 to −0.21)		 0.03
0.21		 Supports a causal association of increased fasting glucose levels with increased BMD [11]
35 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD increase in fasting glucose levels Fracture: 1.04 (0.97 to 1.12) 0.24 No support for a causal association of fasting glucose levels with fracture risk [31]
T1D 19 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per doubling in odds of T1D susceptibility Fracture: 1.00 (1.00 to 1.01) 0.57 No support for a causal association between T1D and fracture risk [31]
T2D 32 BMD 32,961 individuals of European descent, GEFOS Consortium SD change in BMD per increase in log-odds of T2D FN-BMD: 0.034 (0.001 to 0.067)
LS-BMD: 0.022 (−0.01 to −0.051)		 0.04
0.13		 Supports a causal association of T2D with increased BMD [11]
T2D 38 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per doubling in odds of T2D susceptibility Fracture: 0.99 (0.99 to 1.01) 0.37 No support for a causal association between T2D and fracture risk [31]
BMI 32 BMD 5221 children, Avon Longitudinal Study of Parents and Children SD change in BMD per 1 SD increase in BMI UL-BMD: 0.46 (0.31 to 0.61)
LL-BMD: 0.55 (0.41 to 0.68)
SP-BMD: 0.48 (0.33 to 0.63)
SK-BMD: −0.02 (−0.20 to 0.15)		   < 0.001
  < 0.001
  < 0.001
0.78		 Supports a causal association between increased BMI and increased BMD in children [25]
Grip strength 15 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in grip strength Fracture: 2.14 (1.13 to 4.04) 0.01 Supports a causal association between decreased grip strength and fracture risk [31]
Milk consumption 1 BMD 32,965 individuals of European descent, GEFOS Consortium SD change in BMD per 1 SD increase in milk consumption FA-BMD: 0.05 (−0.13 to 0.23)
FN-BMD: 0.02 (−0.09 to 0.06)
LS-BMD: 0.02 (−0.07 to 0.10)		 NA
NA
NA		 No support for a causal association between milk consumption and BMD [27]
1 BMD and fracture 105,256 European-descent individuals 1 g/cm2 change in BMD and OR of fracture for TT and TC genotypes versus CC genotype FN-BMD: 0.10 (0.02 to 0.18) TT
FN-BMD: 0.06 (−0.04 to 0.17) TC
Fracture: 0.86 (0.61 to 1.21) TT
Fracture: 0.90 (0.68 to 1.21) TC		 NA
NA
NA
NA		 No support for a causal association between milk consumption and BMD and fracture risk [30]
1 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per 1 SD decrease in milk consumption Fracture: 1.01 (0.80 to 1.23) 0.94 No support for a causal association between milk consumption and fracture risk [31]
Alcohol consumption 5–6 BMD 32,735 individuals from the GEFOS Consortium and 445,921 individuals of European descent from the UK Biobank NA FN-BMD: −0.008 (NA)
LS-BMD: 0.067 (NA)
FA-BMD: 0.194 (NA)
eBMD: 0.010 (NA)		 0.96
0.20
0.36
0.04		 No support for a causal association between alcohol consumption and BMD [28]
Smoking status 139–142 BMD 32,735 individuals from the GEFOS Consortium and 445,921 individuals of European descent from the UK Biobank NA FN-BMD: −0.139 (NA)
LS-BMD: 0.003 (NA)
FA-BMD: 0.264 (NA)
eBMD: 0.053 (NA)		 0.05
0.98
0.08
0.003		 Supports a causal association between smoking and decreased BMD [28]
Coronary artery disease 38 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per doubling in odds of coronary artery disease susceptibility Fracture: 1.00 (0.99 to 1.02) 0.76 No support for a causal association between coronary artery disease and fracture risk [31]
Rheumatoid disease 30 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per doubling in odds of rheumatoid disease susceptibility Fracture: 1.01 (1.10 to 1.02) 0.14 No support for a causal association between rheumatoid disease and fracture risk [31]
Inflammatory bowel disease 151 Fracture 264,973 individuals of predominantly European ancestry, GEFOS Consortium, UK Biobank and EPIC-Norfolk study OR of fracture per doubling in odds of inflammatory bowel disease susceptibility Fracture: 1.00 (1.10 to 1.01) 0.92 No support for a causal association between inflammatory bowel disease and fracture risk [31]
Open in a new tab

Abbreviations: 25OHD, 25-hydroxyvitamin D; BMD, bone mineral density; BMI, body mass index; CRP, C-reactive protein; eBMD, estimated bone mineral density from ultrasound; EPIC, European Prospective Investigation of Cancer; FA, forearm; FN, femoral neck; GEFOS, Genetic Factors for Osteoporosis; LL, lower limbs; LS, lumbar spine; NA, not available; OR, odds ratio; SD, standard deviation; SK, skull; T1D, type 1 diabetes; T2D, type 2 diabetes; TF, total femur; TH, total hip; TSH, thyroid stimulating hormone; UL, upper limbs bone mineral density.

a

Number of genetic variants used as instrumental variables in the Mendelian randomization analysis.

Most of the exposures that have been assessed for association with BMD or fracture risk using the MR study design would have been unfeasible to test in RCTs. These include for example the effect of alcohol consumption, smoking, estrogen levels, and higher BMI as these exposures have adverse effects on other health outcomes. Other modifiable exposures that would unlikely to be properly evaluated for associations in future RCTs include long-term intake of calcium (may increase the risk of cardiovascular disease [34,35]), other nutrients and dietary compounds, heavy metals, and coffee (difficult to prevent individuals in the placebo group from consuming coffee) as well as comorbidities, body height and other anthropometric measures.

7. Conclusions

The use of the Mendelian randomization study design to infer causality has been evolving over the last decade. This technique has been used to provide evidence of possible causal associations of serum estradiol concentrations, earlier menopause, late puberty, BMI, grip strength, and smoking with BMD and/or fracture risk and the lack of associations of 25‑hydroxyvitamin D, serum thyroid stimulating hormone, homocysteine, urate, and C-reactive protein concentrations with BMD and/or fracture risk. If used prudently, Mendelian randomization can provide an important contribution to our understanding of the etiology of osteoporosis and can be valuable to prioritize exposures for evaluation in RCTs.

Funding

This work did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors. Stephen Burgess is supported by a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (Grant Number 204623/Z/16/Z).

Footnotes

Disclosure

The authors have no conflicts of interest to declare.

References

  • [1].Hendrickx G, Boudin E, Van Hul W. A look behind the scenes: the risk and pathogenesis of primary osteoporosis. Nat Rev Rheumatol. 2015;11:462–474. doi: 10.1038/nrrheum.2015.48. [DOI] [PubMed] [Google Scholar]
  • [2].Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, Davey Smith G. Best (but oft-forgotten) practices: the design, analysis, and interpretation of Mendelian randomization studies. Am J Clin Nutr. 2016;103:965–978. doi: 10.3945/ajcn.115.118216. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [3].Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey Smith G. Mendelian randomization: using genes as instruments for making causal inferences in epidemiology. Stat Med. 2008;27:1133–1163. doi: 10.1002/sim.3034. [DOI] [PubMed] [Google Scholar]
  • [4].Sheehan NA, Didelez V, Burton PR, Tobin MD. Mendelian randomisation and causal inference in observational epidemiology. PLoS Med. 2008;5:e177. doi: 10.1371/journal.pmed.0050177. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [5].Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet. 2014;23:R89–R98. doi: 10.1093/hmg/ddu328. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [6].Burgess S, Thompson SG. Mendelian Randomization: Methods for Using Genetic Variants in Causal Estimation. Chapman and Hall/CRC Press; 2015. [Google Scholar]
  • [8].Smith GD, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol. 2003;32:1–22. doi: 10.1093/ije/dyg070. [DOI] [PubMed] [Google Scholar]
  • [9].Stearns FW. One hundred years of pleiotropy: a retrospective. Genetics. 2010;186:767–773. doi: 10.1534/genetics.110.122549. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [10].Larsson SC, Melhus H, Michaelsson K. Circulating serum 25-hydroxyvitamin D levels and bone mineral density: Mendelian randomization study. J Bone Miner Res. 2018;33:840–844. doi: 10.1002/jbmr.3389. [DOI] [PubMed] [Google Scholar]
  • [11].Ahmad OS, Leong A, Miller JA, Morris JA, Forgetta V, Mujammami M, Richards JB. A Mendelian randomization study of the effect of type-2 diabetes and glycemic traits on bone mineral density. J Bone Miner Res. 2017;32:1072–1081. doi: 10.1002/jbmr.3063. [DOI] [PubMed] [Google Scholar]
  • [12].Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol. 2015;44:512–525. doi: 10.1093/ije/dyv080. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [13].Burgess S, Thompson SG. Interpreting findings from Mendelian randomization using the MR-Egger method. Eur J Epidemiol. 2017;32:377–389. doi: 10.1007/s10654-017-0255-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [14].Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50:693–698. doi: 10.1038/s41588-018-0099-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [15].Burgess S, Thompson SG. Multivariable Mendelian randomization: the use of pleiotropic genetic variants to estimate causal effects. Am J Epidemiol. 2015;181:251–260. doi: 10.1093/aje/kwu283. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [16].Rees JMB, Wood AM, Burgess S. Extending the MR-Egger method for multivariable Mendelian randomization to correct for both measured and unmeasured pleiotropy. Stat Med. 2017;36:4705–4718. doi: 10.1002/sim.7492. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [17].Burgess S, Small DS, Thompson SG. A review of instrumental variable estimators for Mendelian randomization. Stat Methods Med Res. 2017;26:2333–2355. doi: 10.1177/0962280215597579. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [18].Burgess S, Bowden J, Fall T, Ingelsson E, Thompson SG. Sensitivity analyses for robust causal inference from Mendelian randomization analyses with multiple genetic variants. Epidemiology. 2017;28:30–42. doi: 10.1097/EDE.0000000000000559. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [19].Burgess S, Zuber V, Gkatzionis A, Foley CN. Modal-based estimation via heterogeneity-penalized weighting: model averaging for consistent and efficient estimation in Mendelian randomization when a plurality of candidate instruments are valid. Int J Epidemiol. 2018;47:1242–1254. doi: 10.1093/ije/dyy080. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [20].Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol. 2016;40:304–314. doi: 10.1002/gepi.21965. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [21].Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol. 2017;46:1985–1998. doi: 10.1093/ije/dyx102. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [22].Oei L, Campos-Obando N, Dehghan A, Oei EH, Stolk L, van Meurs JB, Hofman A, Uitterlinden AG, Franco OH, Zillikens MC, et al. Dissecting the relationship between high-sensitivity serum C-reactive protein and increased fracture risk: the Rotterdam study. Osteoporos Int. 2014;25:1247–1254. doi: 10.1007/s00198-013-2578-0. [DOI] [PubMed] [Google Scholar]
  • [23].Dalbeth N, Topless R, Flynn T, Cadzow M, Bolland MJ, Merriman TR. Mendelian randomization analysis to examine for a causal effect of urate on bone mineral density. J Bone Miner Res. 2015;30:985–991. doi: 10.1002/jbmr.2434. [DOI] [PubMed] [Google Scholar]
  • [24].Li SS, Gao LH, Zhang XY, He JW, Fu WZ, Liu YJ, Hu YQ, Zhang ZL. Genetically low vitamin D levels, bone mineral density, and bone metabolism markers: a Mendelian randomisation study. Sci Rep. 2016;6 doi: 10.1038/srep33202. 33202. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [25].Kemp JP, Sayers A, Smith GD, Tobias JH, Evans DM. Using Mendelian randomization to investigate a possible causal relationship between adiposity and increased bone mineral density at different skeletal sites in children. Int J Epidemiol. 2016;45:1560–1572. doi: 10.1093/ije/dyw079. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [26].Huang JV, Schooling CM. Inflammation and bone mineral density: a Mendelian randomization study. Sci Rep. 2017;7 doi: 10.1038/s41598-017-09080-w. 8666. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [27].Yang Q, Lin SL, Au Yeung SL, Kwok MK, Xu L, Leung GM, Schooling CM. Genetically predicted milk consumption and bone health, ischemic heart disease and type 2 diabetes: a Mendelian randomization study. Eur J Clin Nutr. 2017;71:1008–1012. doi: 10.1038/ejcn.2017.8. [DOI] [PubMed] [Google Scholar]
  • [28].Guo R, Wu L, Fu Q. Is there causal relationship of smoking and alcohol consumption with bone mineral density? A Mendelian randomization study. Calcif Tissue Int. 2018 doi: 10.1007/s002230180452y. [Epub ahead of print] [DOI] [PubMed] [Google Scholar]
  • [29].Van Vliet NA, Noordam R, Van Klinken JB, Westendorp RGJ, Bassett JHD, Williams GR, Van Heemst D. Thyroid stimulating hormone and bone mineral density: evidence from a two-sample Mendelian randomization study and a candidate gene association study. J Bone Miner Res. 2018;33:1318–1325. doi: 10.1002/jbmr.3426. [DOI] [PubMed] [Google Scholar]
  • [30].Bergholdt HKM, Larsen MK, Varbo A, Nordestgaard BG, Ellervik C. Lactase persistence, milk intake, hip fracture and bone mineral density: a study of 97 811 Danish individuals and a meta-analysis. J Intern Med. 2018 doi: 10.1111/joim.12753. [Epub ahead of print] [DOI] [PubMed] [Google Scholar]
  • [31].Trajanoska K, Morris JA, Oei L, Zheng HF, Evans DM, Kiel DP, Ohlsson C, Richards JB, Rivadeneira F. Assessment of the genetic and clinical determinants of fracture risk: genome wide association and Mendelian randomisation study. BMJ. 2018;362:k3225. doi: 10.1136/bmj.k3225. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [32].Dairy Consumption, Body Mass Index Among Adults, Mendelian randomization analysis of 184802 individuals from 25 studies. Clin Chem. 2018;64:183–191. doi: 10.1373/clinchem.2017.280701. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [33].Bergholdt HK, Nordestgaard BG, Ellervik C. Milk intake is not associated with low risk of diabetes or overweight-obesity: a Mendelian randomization study in 97,811 Danish individuals. Am J Clin Nutr. 2015;102:487–496. doi: 10.3945/ajcn.114.105049. [DOI] [PubMed] [Google Scholar]
  • [34].Bolland MJ, Grey A, Avenell A, Gamble GD, Reid IR. Calcium supplements with or without vitamin D and risk of cardiovascular events: reanalysis of the Women's Health Initiative limited access dataset and meta-analysis. BMJ. 2011;342:d2040. doi: 10.1136/bmj.d2040. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [35].Larsson SC, Burgess S, Michaelsson K. Association of genetic variants related to serum calcium levels with coronary artery disease and myocardial infarction. JAMA. 2017;318:371–380. doi: 10.1001/jama.2017.8981. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [36].Kemp JP, Morris JA, Medina-Gomez C, Forgetta V, Warrington NM, Youlten SE, Zheng J, Gregson CL, Grundberg E, Trajanoska K, et al. Identification of 153 new loci associated with heel bone mineral density and functional involvement of GPC6 in osteoporosis. Nat Genet. 2017;49:1468–1475. doi: 10.1038/ng.3949. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • [37].Locke AE, Kahali B, Berndt SI, Justice AE, Pers TH, Day FR, Powell C, Vedantam S, Buchkovich ML, Yang J, et al. Genetic studies of body mass index yield new insights for obesity biology. Nature. 2015;518:197–206. doi: 10.1038/nature14177. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES