J Rheum Dis 2021; 28(2): 94-100  
The Gut Microbiome and Osteoarthritis: A Two-Sample Mendelian Randomization Study
Young Ho Lee, M.D., Ph.D., Gwan Gyu Song, M.D., Ph.D.
Department of Rheumatology, Korea University College of Medicine, Seoul, Korea
Correspondence to: Young Ho Lee http://orcid.org/0000-0003-4213-1909
Division of Rheumatology, Department of Internal Medicine, Korea University Anam Hospital, Korea University College of Medicine, 73 Goryeodae-ro, Seongbuk-gu, Seoul 02841, Korea. E-mail:lyhcgh@korea.ac.kr
Received: October 5, 2020; Revised: February 8, 2021; Accepted: March 4, 2021; Published online: April 1, 2021.
© Korean College of Rheumatology. All rights reserved.

This is a open Access article, which permits unrestricted non-commerical use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
Objective. The aim of this study was to examine if the intestinal microbiome is causally correlated with osteoarthritis (OA) incidence. Methods. A two-sample Mendelian randomization (MR) study was conducted using inverse variance weighting (IVW), weighted median, and MR-Egger regression techniques. Publicly accessible summary statistics dataset of intestinal microbiomes of European descent from genome-wide association studies (GWASs) (a total with 3,326 individuals) was used as an exposure. As an outcome, summary data from the GWAS include 3,498 patients with OA of the knee and hip from the arcOGEN sample and 11,009 controls of European descent. Results. We identified 29 single-nucleotide polymorphisms from GWAS of intestinal microbiomes as instrumental variables. The IVW approach found no evidence to suggest a causal relationship between the intestinal microbiota and OA (beta=−0.001, standard error [SE]=0.004, p=0.748). The regression test of MR-Egger showed that the directional pleiotropy was unlikely to be a bias (intercept=0.002, SE=0.007, p=0.697) and the MR-Egger study showed no causal relation between the intestinal microbiota and the OA (beta=−0.002, SE=0.005, p=0.630). The weighted median analysis also did not have indications of a causal relationship between the intestinal microbiota and OA (beta=−0.002, SE=0.005, p=0.630). The MR results calculated using IVW, the median weighted and the MR-Egger regression approaches were consistent. Conclusion. The findings of the MR analysis did not support a causal relationship between intestinal microbiome and OA risk.
Keywords: Gut microbiome, Osteoarthritis, Mendelian randomization analysis
INTRODUCTION

Osteoarthritis (OA) is the most popular joint disorder characterized by gradual degeneration of joint cartilage, joint pain, and immobility [1]. OA is the main cause of aged people's suffering and injury. The frequency of OA continues to rise worldwide [2], thus the health burden of OA continues to increase with increasing lifespan. A clearer understanding of the causal risk factors associated with OA is therefore required. Although the etiology of OA is not well known, environmental conditions are likely to play an important role in the production of OA in genetically prone individuals.

While several factors lead to the development of OA, the intestinal microbiota has recently emerged as a pathogenic element in the initiation and progression of OA [3]. Dysbiosis of the intestinal microbiota tends to play a significant part in the etiopathogenesis of OA [4]. Intestinal microbiota DNA may be contained in cartilage and OA synovium [5]. The connection between intestinal microbiota and joint inflammation has been identified in animal arthritis models [6]. Studies have shown major alterations in the intestinal microbiota in OA patients and associations between serum concentrations of bacterial metabolites and joint degeneration, suggesting a connection between intestinal microbiome dysbiosis and OA [7]. All these evidences indicate that dysbiosis may lead to the development of OA. While these findings indicated the potential function of intestinal microbiota in OA manifestations, the mechanism for this involvement is unknown. Furthermore, the actual causal relation remains uncertain as observational studies are vulnerable to the bias, such as reversal causation and residual uncertainty, thereby preventing a clear understanding of the role of intestinal microbiota on OA [8].

Mendelian randomization (MR) is a methodology that uses genetic variants as instrumental variables (IVs) to determine whether the observational interaction between the risk factor and the disease is compatible with the causal impact [9,10]. The purpose of this research was to investigate whether intestinal microbiota is causally correlated with the occurrence of OA by means of a two-sample MR examination.

MATERIALS AND METHODS

Sources of data and choice of genetic variants

We checked the NHGRI-EBI GWASs catalogue (https:/www.ebi.ac.uk/gwas/), which is a systematic catalogue of documented associations of reported genome-wide association studies (GWAS) [11]. We used intestinal microbiota as an exposure and OA as an outcome. Publicly accessible summary statistics datasets of intestinal microbiota GWAS (total 3,326 individuals) of European ancestors were used as an exposure [12,13]. As an outcome, summary data were used from 3,498 patients with OA of the knee and hip and 11,009 controls of European descent from the United Kingdom from the arcOGEN analysis [14]. The OA patients were unrelated and of European origin, and all had primary OA of the hip or knee of radiographic Kellgren–Lawrence grade ≥2, or clinical evidence of disease to a level requiring total joint replacement. Women were 60.8% of the patients and joint replacement was performed in 78.6% [14]. We conducted a two-sample MR analysis utilizing genetic variants correlated with intestinal microbiota as IVs to enhance inference. The ‘top hits’ from a GWAS can be used to define genetic instruments for exposures in MR analyses. The selection criteria for IVs were as follows; 1) Association p-values of single-nucleotide polymorphisms (SNPs) with exposure should be less than a threshold of p<1×10−5 in GWAS dataset of exposure. 2) The SNPs from exposure dataset also should be present in GWAS dataset of outcome. Summary statistics (beta coefficients and standard errors [SEs]) were collected for 29 different SNPs correlated with intestinal microbiota as IVs (Figure 1). And we used the OA GWAS summary results for the 29 SNPs as an outcome [14].

Figure 1. Manhattan plots for IV SNPs from gut microbiome GWAS. The dark horizontal line indicates genome-wide signifi-cance. SNP: single nucleotide polymorphism, GWAS: genome- wide association study, IV: instrumental variable.

Statistical test for Mendel randomization

Study of MR demands that genetic variants is linked to, but not to, possible confounders of, exposure [15]. Therefore, first, we evaluated the independent association of SNPs with gut microbiota. Second, we looked at the relation between each SNP and the occurrence of OA. Third, we merged these results to evaluate the causal relationship between intestinal microbiota and OA occurrence using an MR analysis. Two-sample MR, a tool used to measure the causal effects of exposure (gut microbiota) on outcomes (OA) utilizing summary statistics from various GWASs [16], was used to determine the causal association between intestinal microbiome and occurrence of OA utilizing summary results from GWASs linked to intestinal microbiome and OA with 29 SNPs as IVs (Table 1). The inverse variance weighting (IVW) method utilizes a meta-analysis technique to incorporate the Wald ratio calculation of the causal impact derived from various SNPs and to provide an estimate of the causal impact of the exposure on the outcome, whenever any of the genetic variants satisfies the assumptions of the IVs [16]. While the inclusion of several variants in the MR study results in improved statistical accuracy, it has the ability to incorporate pleiotropic genetic variants that are not appropriate for IVs [17]. For the discovery and adjustment of pleiotropy, i.e., the correlation of genetic variations with more than one component, the weighted median and MR-Egger regression methods have been used. MR-Egger regression is resilient to invalid instruments, checks and accounts for the existence of imbalanced pleiotropy by introducing a parameter for this bias by integrating summary data estimates of causal effects from multiple individual variants [18,19]. MR-Egger regression was used to assess the potential for violations of MR assumptions and to improve the reliability of causal inference. The weighted median estimator offers an accurate estimate of the causal effect, even though up to 50% of the data that contributes to the study is derived from genetic variants that are invalid IVs [20]. The benefit of the weighted median estimator is that it maintains better consistency in the results relative to the MR-Egger analysis [20]. Analyzes were found statistically relevant at p< 0.05. Both MR analyzes were conducted on the MR Base Platform (App version: 1.2.2 3a435d [31 January 2019], R version: 3.5.1) [21].

Table 1 . Instrumental SNPs associated with gut microbiome and OA GWASs.

Instrumental SNPGeneEffect alleleExposure (gut microbiome)Outcome (OA)


BetaSEp-valueBetaSEp-value
rs1009634AKAP3G−1.3100.2267.00E-090.0300.0550.561
rs10928827HS6ST1G−0.2300.0394.00E-09−0.0100.0210.600
rs11626933C14orf102G−0.2400.0432.00E-080.0100.0250.720
rs11724031SHROOM3G−0.9700.1522.00E-10−0.0770.0350.036
rs11877825NAPGG−0.2700.0413.00E-11−0.0100.0250.756
rs11915634CNTN6T−1.3000.2063.00E-100.0620.0480.237
rs1230666MAGI3A−4.5831.0045.00E-060.0100.0300.841
rs12442649TMCO5AG−1.4900.2714.00E-080.0730.0380.062
rs13096731FLNBA−0.4300.0783.00E-080.0200.0220.352
rs17085775C9orf71C−1.0300.1842.00E-080.0410.0340.288
rs17661843ABCA13T−1.4000.1854.00E-14−0.0200.0330.606
rs2071199HNF4A-AS1T−0.3200.0561.00E-080.0100.0310.857
rs2294239ZNRF3G−4.6761.0013.00E-060.0100.0230.802
rs2318350COL22A1T−1.1500.1954.00E-090.0510.0400.210
rs249733SPRY4T−0.6500.1055.00E-10−0.0300.0210.174
rs35275482BNIP2C−0.5400.0824.00E-110.0000.0280.883
rs4621152AC007557.1C−0.2900.0511.00E-080.0410.0240.099
rs4669413RP11-521D12.1T−0.1800.0311.00E-080.0300.0220.128
rs479105PRMT8T−0.2200.0381.00E-08−0.0200.0210.351
rs56006724C2orf83A−0.8800.1426.00E-100.0000.0260.866
rs59042687LINC00879T−0.2300.0406.00E-090.0000.0200.962
rs597205C1orf183T−0.6200.1078.00E-09−0.0390.0390.344
rs62295801LINC01192G−0.2700.0435.00E-100.0100.0230.648
rs75036654LINC01137C−1.3900.2245.00E-100.0000.0430.910
rs9291879CD180C−0.5800.0994.00E-09−0.0620.0300.062
rs9300430RAP2AC−0.6100.1001.00E-090.0000.0410.937
rs9323326SLC35F4A−0.2100.0349.00E-10−0.0200.0240.272
rs9831278LINC00973C−1.1600.2093.00E-08−0.0680.0550.215
rs986417SIX6C−1.4000.2363.00E-09−0.0580.0390.132

SNP: single nucleotide polymorphism, GWAS: genome-wide association study, Beta: beta coefficient, SE: standard error, OA: osteoarthritis..



Tests of heterogeneity and sensitivity

Heterogeneity between SNPs was tested using Cochran Q-statistics and funnel plots [22]. We have carried out a “leave one out” study to explore the likelihood of a causal relationship being powered by a single SNP. We conducted a subgroup study on the anatomical location (hip/knee) for the sensitivity examination.

RESULTS

Instrumental variables for Mendelian randomization

A total of 29 SNPs from GWAS on intestinal microbiota were chosen as IVs (Table 1). All of them were correlated with intestinal microbiota of genome-wide significance, except for rs1230666 and rs2294239 (Table 1). Rs11724031 was strongly correlated with OA, while some were not substantially linked to OA (Figure 2).

Figure 2. Forest plot of the causal effects of gut microbiome-associated single-nucleotide polymorphisms on osteoarthritis. IVW: inverse variance weighting, MR: Mendelian randomization.

Results of Mendelian randomization

The IVW approach finds no evidence to prove a causal relationship between intestinal microbiota and OA (beta=−0.001, SE=0.004, p=0.748) (Table 2, Figures 2 and 3). The regression of MR-Egger showed that directional pleiotropy was unlikely to result in a bias (intercept=0.002, SE=0.007, p=0.697). The MR-Egger study showed no causal relation between intestinal microbiota and OA (beta=−0.002, SE=0.005, p=0.630) (Table 2, Figure 3). The weighted median analysis also did not have evidence of a causal relationship between intestinal microbiota and OA (beta=−0.002, SE=0.005, p=0.630) (Table 2, Figure 3). The MR estimates calculated using IVW, the median weighted and the MR-Egger regression approaches were consistent. Therefore, the findings of the MR analysis did not support a causal association between intestinal microbiome and OA risk.

Table 2 . The MR estimates from each method of assessing the causal effect of gut microbiome on the incidence of OA.

MR methodNumber of SNPsBetaSEAssociation
p-value
Cochran Q statisticHeterogeneity
p-value
Inverse variance weighted29−0.0010.0040.74833.140.231
MR Egger29−0.0020.0050.63032.950.199
Weighted median29−0.0020.0050.636NANA

MR: Mendelian randomization, OA: osteoarthritis, SNP: single nucleotide polymorphism, Beta: beta coefficient, SE: standard error, NA: not available..



Figure 3. Scatter plots of genetic associations with gut microbiome against the genetic associations with osteoarthritis. The slopes of each line represent the causal association for each method. The dash line represents the inverse-variance weighting estimate; the dot line, the weighted median estimate; and the linear line, the Mendelian randomization‐Egger estimate. SNP: single nucleotide polymorphism.

Tests of heterogeneity and sensitivity

The Cochran Q test and the funnel test displayed no signs of heterogeneity between the IV estimates dependent on the individual variants, and the MR-Egger regression indicated no signs of directional pleiotropy (Table 2, Figure 3). Results from the “leave one out” study revealed that the IVW point estimate was not guided by a single SNP, and the MR results stayed non-significant after excluding rs1230666 and rs2294239 (Figure 4). When the MR study was stratified by anatomical location (hip/knee), the findings were not relevant in both the knee and hip OA.

Figure 4. “Leave-one-out” analysis to investigate the possibility of the causal association being driven by a unique single-nucleotide polymorphism. MR: Mendelian randomization.
DISCUSSION

Several studies have reported that intestinal microbiota could be linked with an OA risk [3,4]. These findings suggest the possibility that the microbiome can mediate the effects and outcomes of OA. However, the exact function of intestinal microbiota involvement in OA pathophysiology remains under investigation. Hypotheses have been proposed to clarify the following findings [3]: (1) dysbiosis is correlated with metabolic syndrome, which may facilitate OA [23]; (2) in mice, obesity is correlated with depletion of beneficial Bifidobacteria in the intestine, contributing to macrophage migration to synovium and increased knee OA [24]; (3) OA is associated with lower-grade inflammation, which is mediated mainly by the innate immune system. Metabolic endotoxemia induced by damaged intestinal mucosa, and low-grade chronic inflammation can lead to the onset and progression of OA in obese patients [25].

The causal link between intestinal microbiota and OA remains uncertain, as previously recorded correlations might have been distorted by biases or conflicting factors inherent in observational studies, such as reverse causation interpretation, few small-scale trials, and selection bias. Thus, in order to understand this causality relationship, we tested the causality using three separate estimation approaches (IVW method, weighted median method, and MR-Egger regression) for MR research. Our analysis did not suggest a causative link between intestinal microbiome and OA risk. The MR estimates using IVW, MR Egger, and the weighted median analysis were consistent, and did not endorse a causal relationship between intestinal microbiota and OA.

MR minimizes the probability of bias present in observational trials owing to residual confounding or reverse causality. However, the MR experiments are vulnerable to pleiotropic distortion (a mixture of genetic variations of more than one variable) [26]. While the inclusion of multiple variants in the MR study contributes to improved statistical strength, it also results in the possible inclusion of pleiotropic genetic variants that are not true for IVs [27]. In order to adjust pleiotropy, we used a weighted median estimator, which offers accurate estimates even though 50 % of the SNPs are not accurate instruments [20], and we used MR-Egger regression to include an unbalanced pleiotropy test and evaluate a causal estimation of the impact of the treatment on the result in its presence [18]. The MR-Egger analysis and weighted median estimator results were comparable to the IVW estimator, giving increased confidence in these correlations. Our data indicate that the relationship between intestinal microbiome and OA previously documented could be the result of bias or conflicting factors inherent in observational studies.

MR rests on three assumptions of the instrument strength independent of direct effect (InSIDE). The first assumption was easily satisfied in MR by restricting the instruments to genetic variants that are discovered using genome-wide levels of statistical significance. To improve reliability of causal inference, MR results should be presented alongside sensitivity analyses that make allowance for various potential patterns of horizontal pleiotropy. Different extensions of the MR design such as MR-Egger or a weighted median test can be used to detect the presence of pleiotropy and/or estimate the causal effect of exposure, even in the presence of assumption 2 and/or 3 violations [6]. We proved the second and third assumptions by MR-egger regression and weighted median tests, which indicated that directional pleiotropy was unlikely to bias the result. Our MR analysis may hold the InSIDE assumptions and the causal estimate may be reliable.

There are some drawbacks to this study. First, to draw conclusions about negative association between gut microbiota and OA may have limitation, because of OA’s complex mechanism with a limited number of SNPs. Genetic differences have only a minimal impact on a given exposure, as they may explain just a small proportion of the total variation of a given exposure [28]. Our research could have had insufficient power to detect a link. Second, our study involved a limited number of SNPs as IVs, which might not have been statistically significant enough to identify association; statistical precision may be improved and a more reliable causal approximation can be achieved by adding many genetic variants together. Third, research on intestinal microbiome in OA was focused on participants with European ancestry. Since causality may rely on race, MR studies including other population are required. This MR also has its strength. While intestinal microbiome has been examined as a possible risk factor for OA, to the best of our knowledge, this is the first MR research to evaluate the causal association between intestinal microbiome and OA.

CONCLUSION

In conclusion, the findings of this MR analysis did not support a causal association between intestinal microbiome and OA risk. Epidemiological evidence of an association between intestinal microbiota and OA may not seem to be causal. Well-designed epidemiological and MR experiments utilizing a lot of genetic variants that describe a higher proportion of intestinal microbiome are required to determine the causality.

CONFLICT OF INTEREST

No potential conflict of interest relevant to this article was reported.

AUTHOR CONTRIBUTIONS

Y.H.L. was involved in conception and design of study, acquisition of data, analysis and/or interpretation of data, drafting the manuscript, revising the manuscript critically for important intellectual content. G.G.S. was involved in conception and design of study, analysis and interpretation of data, drafting the manuscript.

References
  1. Dieppe PA, Lohmander LS. Pathogenesis and management of pain in osteoarthritis. Lancet 2005;365:965-73.
    Pubmed CrossRef
  2. Litwic A, Edwards MH, Dennison EM, Cooper C. Epidemiology and burden of osteoarthritis. Br Med Bull 2013;105:185-99.
    Pubmed KoreaMed CrossRef
  3. Berthelot JM, Sellam J, Maugars Y, Berenbaum F. Cartilage- gut-microbiome axis: a new paradigm for novel therapeutic opportunities in osteoarthritis. RMD Open 2019;5:e001037.
    Pubmed KoreaMed CrossRef
  4. Favazzo LJ, Hendesi H, Villani DA, Soniwala S, Dar QA, Schott EM, et al. The gut microbiome-joint connection: implications in osteoarthritis. Curr Opin Rheumatol 2020;32:92-101.
    Pubmed KoreaMed CrossRef
  5. Zhao Y, Chen B, Li S, Yang L, Zhu D, Wang Y, et al. Detection and characterization of bacterial nucleic acids in culture-negative synovial tissue and fluid samples from rheumatoid arthritis or osteoarthritis patients. Sci Rep 2018;8:14305.
    Pubmed KoreaMed CrossRef
  6. Griffin TM, Huebner JL, Kraus VB, Yan Z, Guilak F. Induction of osteoarthritis and metabolic inflammation by a very high-fat diet in mice: effects of short-term exercise. Arthritis Rheum 2012;64:443-53.
    Pubmed KoreaMed CrossRef
  7. Lorenzo D, GianVincenzo Z, Carlo Luca R, Karan G, Jorge V, Roberto M, et al. Oral-gut microbiota and arthritis: is there an evidence-based axis? J Clin Med 2019;8:1753.
    Pubmed KoreaMed CrossRef
  8. Ranstam J. Bias in observational studies. Acta Radiol 2008;49:644-5.
    Pubmed CrossRef
  9. Burgess S, Daniel RM, Butterworth AS, Thompson SG. Network Mendelian randomization: using genetic variants as instrumental variables to investigate mediation in causal pathways. Int J Epidemiol 2015;44:484-95.
    Pubmed KoreaMed CrossRef
  10. Song GG, Lee YH. Causal association between bone mineral density and osteoarthritis: a Mendelian randomization study. J Rheum Dis 2019;26:104-10.
    CrossRef
  11. Welter D, MacArthur J, Morales J, Burdett T, Hall P, Junkins H, et al. The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic Acids Res 2014;42(Database issue):D1001-6.
    Pubmed KoreaMed CrossRef
  12. Wang J, Thingholm LB, Skiecevičienė J, Rausch P, Kummen M, Hov JR, et al. Genome-wide association analysis identifies variation in vitamin D receptor and other host factors influencing the gut microbiota. Nat Genet 2016;48:1396-406.
    Pubmed KoreaMed CrossRef
  13. Bonder MJ, Kurilshikov A, Tigchelaar EF, Mujagic Z, Imhann F, Vila AV, et al. The effect of host genetics on the gut microbiome. Nat Genet 2016;48:1407-12.
    Pubmed CrossRef
  14. Zeggini E, Panoutsopoulou K, Southam L, Rayner NW, Day-Williams AG, Lopes MC, et al; arcOGEN Consortium and arcOGEN Collaborators. Identification of new susceptibility loci for osteoarthritis (arcOGEN): a genome-wide association study. Lancet 2012;380:815-23.
    Pubmed KoreaMed CrossRef
  15. Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol 2013;37:658-65.
    Pubmed KoreaMed CrossRef
  16. Pierce BL, Burgess S. Efficient design for Mendelian randomization studies: subsample and 2-sample instrumental variable estimators. Am J Epidemiol 2013;178:1177-84.
    Pubmed KoreaMed CrossRef
  17. Hartwig FP, Davies NM, Hemani G, Davey Smith G. Two-sample Mendelian randomization: avoiding the downsides of a powerful, widely applicable but potentially fallible technique. Int J Epidemiol 2016;45:1717-26.
    Pubmed KoreaMed CrossRef
  18. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol 2015;44:512-25.
    Pubmed KoreaMed CrossRef
  19. Burgess S, Thompson SG. Interpreting findings from Mendelian randomization using the MR-Egger method. Eur J Epidemiol 2017;32:377-89.
    Pubmed KoreaMed CrossRef
  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;40:304-14.
    Pubmed KoreaMed CrossRef
  21. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife 2018;7:e34408.
    Pubmed KoreaMed CrossRef
  22. Egger M, Smith GD, Phillips AN. Meta-analysis: principles and procedures. BMJ 1997;315:1533-7.
    Pubmed KoreaMed CrossRef
  23. Courties A, Sellam J, Berenbaum F. Metabolic syndrome- associated osteoarthritis. Curr Opin Rheumatol 2017;29:214-22.
    Pubmed CrossRef
  24. Schott EM, Farnsworth CW, Grier A, Lillis JA, Soniwala S, Dadourian GH, et al. Targeting the gut microbiome to treat the osteoarthritis of obesity. JCI Insight 2018;3:e95997.
    Pubmed KoreaMed CrossRef
  25. Metcalfe D, Harte AL, Aletrari MO, Al Daghri NM, Al Disi D, Tripathi G, et al. Does endotoxaemia contribute to osteoarthritis in obese patients? Clin Sci (Lond) 2012;123:627-34.
    Pubmed CrossRef
  26. Thompson JR, Minelli C, Bowden J, Del Greco FM, Gill D, Jones EM, et al. Mendelian randomization incorporating uncertainty about pleiotropy. Stat Med 2017;36:4627-45.
    Pubmed CrossRef
  27. Smith GD, Ebrahim S. Mendelian randomization: prospects, potentials, and limitations. Int J Epidemiol 2004;33:30-42.
    Pubmed CrossRef
  28. Swerdlow DI, Kuchenbaecker KB, Shah S, Sofat R, Holmes MV, White J, et al. Selecting instruments for Mendelian randomization in the wake of genome-wide association studies. Int J Epidemiol 2016;45:1600-16.
    Pubmed KoreaMed CrossRef


This Article


Author ORCID Information

Services
Social Network Service

e-submission

Archives