Quantitative Relationship Between Cumulative Risk Alleles Based on Genome-Wide Association Studies and Type 2 Diabetes Mellitus: A Systematic Review and Meta-analysis

Many epidemiological studies have assessed the genetic risk of having undiagnosed or of developing type 2 diabetes mellitus (T2DM) using several single nucleotide polymorphisms (SNPs) based on findings of genome-wide association studies (GWAS). However, the quantitative association of cumulative risk alleles (RAs) of such SNPs with T2DM risk has been unclear. The aim of this meta-analysis is to review the strength of the association between cumulative RAs and T2DM risk. Systematic literature searches were conducted for cross-sectional or longitudinal studies that examined odds ratios (ORs) for T2DM in relation to genetic profiles. Logarithm of the estimated OR (log OR) of T2DM for 1 increment in RAs carried (1-ΔRA) in each study was pooled using a random-effects model. There were 46 eligible studies that included 74,880 cases among 249,365 participants. In 32 studies with a cross-sectional design, the pooled OR for T2DM morbidity for 1-ΔRA was 1.16 (95% confidence interval [CI], 1.13–1.19). In 15 studies that had a longitudinal design, the OR for incident T2DM was 1.10 (95% CI, 1.08–1.13). There was large heterogeneity in the magnitude of log OR (P < 0.001 for both cross-sectional studies and longitudinal studies). The top 10 commonly used genes significantly explained the variance in the log OR (P = 0.04 for cross-sectional studies; P = 0.006 for longitudinal studies). The current meta-analysis indicated that carrying 1-ΔRA in T2DM-associated SNPs was associated with a modest risk of prevalent or incident T2DM, although the heterogeneity in the used genes among studies requires us to interpret the results with caution.


INTRODUCTION
Intensive lifestyle interventions (eg, promoting increased physical activity and weight loss) can be effective in decreasing the incidence of type 2 diabetes mellitus (T2DM). 1 However, healthcare resources are limited, and participants in interventions to prevent diabetes should be prioritized. Identification of individuals at high risk of T2DM could facilitate the targeting of prevention efforts to those who could benefit from them and reduce the cost of preventing T2DM.
Predicting T2DM in healthy individuals has been attempted using a diabetes risk score that is derived from common clinical information, such as adiposity, blood pressure, and family history of T2DM. However, using the risk score is inevitably limited in predicting T2DM because T2DM has a strong genetic basis; concordance of T2DM is about 70% for monozygotic twins, compared to about 20-30% for dizygotic twins. 2 Limitations in predicting T2DM have driven researchers to employ genetic risk assessments. Moreover, unlike clinical markers, genetic markers do not change with time, so they possess the advantage of identifying high-risk individuals long before disease onset, which could enable early interventions for preventing T2DM. Conventionally, family-based linkage studies have played an important role in identifying genes having a large effect in monogenic disorders, such as maturity-onset diabetes of the young. 3 However, linkage studies have low power for polygenic diseases that are influenced by multiple genes, as is the case with the majority of those with T2DM. Therefore, using monogenic mutations would have very limited value for predicting risk of disease in the general population because of their low frequency.
Genome-wide association studies (GWAS) capture the great majority of common genetic differences among individuals and relate them to health and diseases. 4 The GWAS potentially represent a powerful new tool for identifying genes that influence common diseases. Recently, GWAS identified an increasing number of loci associated with susceptibility to T2DM. However, each risk allele (RA) of single nucleotide polymorphisms (SNPs) at these loci has a modest effect size. 5 Therefore, a combination of several SNPs is required to substantially influence T2DM risk. A genetic risk score potentially has the ability to predict disease risk as a function of the combined effects of SNPs. The most typical approach to produce the genetic risk score is to count the total number of RAs at each T2DM-associated SNP identified by GWAS. Because the number of RAs carried is consequently a quantifiable variable, it can be potentially used for clinical risk assessment, similar to measurement of body mass index (BMI). Recently, an increasing number of epidemiological studies have investigated the effect of cumulative RAs on T2DM risk. However, the quantitative association between cumulative RAs and T2DM risk has not been established, while the association of BMI with T2DM risk was quantified using meta-analysis. 6 The aim of this meta-analysis is to comprehensively estimate the strength of the association between cumulative RAs and T2DM risk, including exploration of differences in the magnitude of T2DM risk according to several study characteristics.

Literature searches
Electronic literature search using EMBASE and MEDLINE (up to 2015) was conducted for studies that quantified the genetic risk of T2DM. This search was limited to articles published after the International HapMap Project had identified a majority of the common SNPs examined by GWAS in 2003. 7 Manual searches were added using the reference list of each included study.

Inclusion criteria
Study keywords were thesaurus terms related to genetic profiles and T2DM (Table 1). Inclusion criteria were: 1) cross-sectional design (for assessing the possibility of having undiagnosed T2DM) or longitudinal design (for assessing the risk of developing T2DM); 2) total number of RAs carried calculated by summing the number of RAs in each SNP according to an additive model (ie, 0, 1, and 2 were assigned for homozygosity for a non-RA, heterozygosity for an RA, and homozygosity for an RA, respectively) were analyzed as an exposure; 3) two or more SNPs were combined for assessing genetic risk; and 4) the odds ratio (OR) for each increment in the number of RAs carried (1-ΔRA) and its corresponding standard error (SE) (SE is usually provided for the natural logarithm of OR [log OR] rather than for the OR) were presented or could be estimated. Exceptionally, studies that used the dominant model (ie, counting 1 for homozygosity or heterozygosity for having an RA and 0 for having no RA) were also considered.
Haplotypes into which alignments of multiple SNPs were classified were regarded as one unit because these SNPs could not be separated into individual SNPs. Therefore, studies that examined the association between several haplotypes and T2DM risk were excluded, even if two or more SNPs were used. Due to the necessity of maintaining homogeneity across studies as closely as possible, the OR needed to be adjusted for at least two of the following three covariates: age, gender, and BMI. For the same reason, studies were excluded if they examined T2DM risk in relation to a weighted risk score that was produced by summing the number of RAs multiplied by the SNP-specific effect size. In longitudinal studies, if the risk measure was expressed as a relative risk (RR) rather than OR, we included such studies on the condition that the incidence rate in the study population was reported. In that case, the RR was converted into an OR using the following formula: OR ¼ RRÂð1ÀPoÞ 1ÀPoÂRR , in which the incidence rate was imputed to P o . 8 Also, as the SE corresponding to the natural logarithm of RR (log RR) was converted into the SE corresponding to log OR, the Miettinen test-based approach was applied: SEðlog ORÞ ¼ SEðlog RRÞ Â ffiffiffiffiffiffiffiffiffiffi ffi log OR log RR q . 9

Data extraction
For each included study, two authors (S.K. and H.So.) extracted the following information relevant to study characteristics: country, ethnic group (if information was presented), design (cross-sectional or longitudinal), observational periods (in case of a longitudinal study), number of participants and cases, mean age, proportion of men and women, mean BMI, study-specific covariates, criteria of T2DM cases and non-cases, and used SNPs. Inconsistencies were resolved via discussion. Study quality was assessed by 10 basic questions that should be answered in the affirmative to indicate a reliable report (Table 2). 4 One point was awarded to a study for each "Yes" answer, with a maximum score of 10.
If a study presented several ORs for T2DM with different levels of adjustment, the most completely adjusted OR was chosen. If a study presented several ORs with different combinations of SNPs used for T2DM screening, the OR that had the greatest effect size was chosen. In addition, if a study [related to genetic backgrounds] "genetic variability" OR "genetic polymorphism" OR "single nucleotide polymorphism [Exp]" OR "genetic association" OR "genotyping" OR "genetic susceptibility" OR "genetic resistance" OR "genetic predisposition" #2 [related to type 2 diabetes mellitus] "non insulin dependent diabetes mellitusepidemiology" OR "diabetes mellitusepidemiology" #3 #1 AND #2 Using MEDLINE #4 [related to genetic backgrounds] "Genome-Wide Association Study" OR "Genetic Association Studies" OR "Polymorphism, Single Nucleotide" OR "Genetic Variation [Exp]" OR "Polymorphism, Genetic [Exp]" OR "Genotype" OR "Genetic Predisposition to Disease" #5 [related to type 2 diabetes mellitus] "Diabetes Mellitus, Type 2epidemiology" OR "Diabetes Mellitus, Type 2genetics" OR "Diabetes Mellitusgenetics" OR "Diabetes Mellitusepidemiology" #6 #4 AND #5 Combination of EMBASE and MEDLINE #7 #3 OR #6 [Exp] indicates automatic inclusion of all the narrower terms under the specified descriptor in the thesaurus hierarchy. Using the connector "-" limits a descriptor term to a Subheading appearing after the connector.
Cumulative Risk Alleles and Type 2 Diabetes Mellitus Each question used for assessment of study quality is given below. 4 If the answer is "Yes" to a question, a study could be awarded 1 point. Full score is 10. Q.1. Are the cases defined clearly and reliably so that they can be compared with patients typically seen in clinical practice? Q.2. Are case and control participants demonstrated to be comparable to each other for important characteristics that might also be related to genetic variation and to the disease? In longitudinal studies, the answer was considered to be "Yes" if the diabetes risk was adjusted for age, gender, and obesity index. Q.3. Was the study of sufficient size to detect modest odds ratios or relative risks (1.3-1.5)? (At least 1,500 cases are needed to obtain 90% statistical power for detecting a 30% allelle with a 1.5 odds ratio at P < 10 −8 of a significant level. + ) Q.4. Was the genotyping platform of sufficient density to capture a large proportion of the variation in the population studied? Q.5. Were appropriate quality control measures applied to genotyping assays, including visual inspection of cluster plots and replication on an independent genotyping platform? (The concordance rate needed to be presented by using a duplicate sample if the study met this criterion.) Q.6. Did the study reliably detect associations with previously reported and replicated variants (known positives)? Q.7. Were stringent corrections applied for the many thousands of statistical tests performed in defining the P value for significant associations? Q.8. Were the results replicated in independent population samples? Q.9. Were the replication samples comparable in geographic origin and phenotype definition, and if not, did the differences extend the applicability of the findings? Q.10. Was evidence provided for a functional role for the gene polymorphism identified?
performed both overall and subgroup analyses, we chose the data based on subgroup analyses if characteristics, such as mean age, proportion of men, and mean BMI, in each subgroup were described; otherwise, data based on an overall analysis were chosen.

Data synthesis
Most of the included studies did not directly present data on the OR for the increment in the number of RAs carried. To estimate the OR, the log ORs in several genetic risk groups in the individual study were regressed to their corresponding mean number of RAs carried. 10  Cross-sectional studies and longitudinal studies were analyzed separately. In each study, the OR of T2DM for 1-ΔRA was transformed into log OR because its corresponding SE is provided for log OR rather than for the OR. The log OR was pooled with a random-effects model 12 followed by exponentiation of the pooled log OR to obtain the OR of interest. Study heterogeneity was assessed by Q-statistics or I-squared overall and within each strata after stratification. 13 Publication bias was assessed by two formal statistical tests: Begg's rank correlation test 14 and Egger's regression asymmetry test. 15 For statistically suspected publication bias, the trim and fill method was adopted to adjust the pooled T2DM risk, assuming that the asymmetry of the funnel plot is entirely due to publication bias. This method includes detection of unpublished studies that distorted the funnel plot, filling the results of these hypothetical studies to recover the symmetry of the funnel plot, and recalculation of the pooled effect size as if these studies had actually existed. 16

Sensitivity analysis
The pooled OR was estimated for subgroups after stratification on the basis of the following pre-specified study characteristics: number of SNPs (<10 or ≥10), mean age (<55 years or ≥55 years in cross-sectional studies; <50 years or ≥50 years in longitudinal studies), proportion of men (<50% or ≥50%), country where the study was conducted (Western or non-Western), ethnic group (Asian or non-Asian), mean BMI (<27 kg=m 2 or ≥27 kg=m 2 in cross-sectional studies; <25 kg=m 2 or ≥25 kg=m 2 in longitudinal studies), whether the OR was or was not adjusted for BMI, whether the oral glucose tolerance test (OGTT) was thoroughly performed for detecting any cases of T2DM, and whether subjects with impaired glucose tolerance (IGT) or impaired fasting glucose (IFG) were excluded from non-T2DM cases. The cut-off values for mean age and BMI used in the stratified analyses corresponded to these median values in cross-sectional and longitudinal studies (median age: 55.7 years for cross-sectional studies and 50.6 years for longitudinal studies; median BMI: 26.7 kg=m 2 for cross-sectional studies and 25.9 kg=m 2 for longitudinal studies). Univariate log-linear metaregression analysis was used to test the differences in the magnitude of the OR between strata.
Multivariate log-linear meta-regression analyses were added using several study characteristics simultaneously as explanatory variables. In these regression analyses, log OR of T2DM for 1-ΔRA was used as an objective variable. The T2DM-susceptible genes that were used in each study were inconsistent. Therefore, we added a meta-regression analysis of whether or not each of these genes was examined. This analysis was limited to the top 10 of the commonly used genes in this meta-analysis in order to maintain statistical power. All analyses were based on statistical software STATA version 14 (STATA Corp., College Station, TX, USA). The meta-regression analyses were performed by installing the "metareg" command into STATA. We also installed the "glst" command to estimate the OR for carrying 1-ΔRA using GLST. Figure 1 is a flow chart showing the procedure for identifying studies that met the initial inclusion criteria. Of 7,381 studies retrieved through the electronic literature search, 96 investigated the genetic risk for T2DM combining two or more SNPs. After 10 studies in which the study population overlapped with other studies were eliminated, there remained 86 studies for further review. Forty studies did not allow estimation of the adjusted OR for T2DM for one increment in risk alleles carried. Finally, 46 original studies  in which the OR for T2DM for 1-ΔRA was presented or could be estimated, including 75,651 cases among 249,365 participants, were eligible. Twenty-eight studies 17,19,20,23,26,27,29,30,[33][34][35]38,40,42,43,[45][46][47]49,51,53,[55][56][57][58][59][60]62 analyzed T2DM risk in Western countries. Table 3 shows characteristics of the 46 eligible studies. Most studies were not considered to have targeted exclusively a single ethnicity, and only 5 studies 19,26,39,47,57 performed analyses by ethnicity. Only 2 studies 49,62 analyzed men and women separately, and all but 3 studies 37,46,53 included both genders. One study 48 analyzed T2DM risk from both cross-sectional and longitudinal perspectives. Table 4 summarizes the criteria for T2DM cases and non-cases in each study. OGTT was used for any participant in whom T2DM was undiagnosed using other methods in 20 studies. 22 Table 5 shows details of covariates that each study arbitrarily considered. Study-specific covariates were heterogeneous among studies. However, T2DM risk was adjusted for BMI in most studies (37 studies). [17][18][19][20][22][23][24][25][26][27][28][31][32][33][34][35][36][37][38][40][41][42][44][45][46][47][48][49][50][51][52][53][54][55][56]59,62 Table 2 describes the assessment of study quality. No study received a full score of 10. Also, there were some "no" responses to each of the items. Mean (standard deviation) score was 4.3 (1.6). Table 6 summarizes the T2DM-associated loci used in each study. The number of SNPs ranged from 2 to 65 (median, 14.0 SNPs). Table 7 indicates the rank of the genes that were commonly used in association with prevalent or incident Cumulative Risk Alleles and Type 2 Diabetes Mellitus T2DM. The top 10 genes were CDKAL1, TCF7L2, CDKN, HHEX, IGFBP2, SLC30A8, KCNJ11, PPARG, FTO, and KCNQ1. Total number of SNPs that were covered by at least one of the included studies was 116.
Kodama S, et al.

DISCUSSION
The current meta-analysis indicated that 1-ΔRA in T2DMassociated SNPs was associated with a modest risk of prevalent or incident T2DM. A previous meta-analysis 6 indicated that the pooled RR of incident T2DM was 1.87 for an increment of 4.3 kg=m 2 in the BMI. If the magnitude of T2DM risk is compared between 1-ΔRA and 1-ΔBMI, 1-ΔRA corresponds to only 0.58-ΔBMI (Δ1.7 kg, if body height is 1.7 meters), assuming that the cumulative incidence rate of T2DM in the referent is p 0 = 10% and the OR is transformed into an RR using the following formula: RR ¼ In other words, having 1 RA was equivalent to losing less than 2 kg body weight.
Assuming the monotonicity between the number of RAs carried and T2DM risk, the magnitude of disease risk will endlessly expand with increases in the RAs carried, even if the effect size of each SNP is modest. Therefore, GWAS may be able to detect individuals at extremely high risk of T2DM if the number of the identified T2DM-associated SNPs is progressively increased. However, the magnitude of T2DM risk might reach a plateau, which was suggested from the result of the stratified analysis that indicated that studies using a larger number of SNPs (10 or more) revealed a smaller OR of T2DM for 1-ΔRA compared with those using a smaller number of SNPs. Although the observed higher risk when using a smaller number of genes could be spurious due to the winner's curse effect (eg, T2DM risk would be higher when using only the top two genes that were strongly associated with susceptibility to T2DM than when using the top 10 genes), this result reflects the fact that the T2DMsusceptible loci were detected in order of descending effect size related to T2DM risk among all proposed loci. 65 Of note, although the magnitude of genetic T2DM risk was modest, this did not mean that the risk was ignorable from the result of the stratified analysis that indicated that the association of cumulative RAs with T2DM risk was not weakened after adjusting the T2DM risk for BMI. This result suggested that the genetic T2DM risk existed independently of obesity, which is well known as one of the greatest risk factors for T2DM.
The current meta-analysis indicated a large heterogeneity in the magnitude of the T2DM risk for cumulative RAs, which was in a large part explained by the choice of the T2DM-susceptible genes. This suggested that the magnitude of T2DM risk depended on what genes were used. In particular, the cross-sectional studies that used the FTO gene revealed a lower OR compared with those that did not use it, which suggested that it is unnecessary to examine the FTO gene if clinical risk factors for T2DM were simultaneously assessed. FTO is one of the most well-known of the obesity-associated genes. 66 Considering that the OR for T2DM was adjusted for obesity in most of the studies included in this meta-analysis, the T2DM risk associated with the FTO gene could have been masked by that associated with obesity. Although the magnitude of T2DM risk was not influenced by whether or not FTO was used in the longitudinal studies, the inconsistency in the results between the meta-analysis of crosssectional studies and that of longitudinal studies could be explained by differences in the exposure to risk factors (ie, participants in longitudinal studies were recruited before they became obese, although those in cross-sectional studies had already become obese at recruitment). Despite the heterogeneity in the choice of the T2DM-susceptible genes, the forest plots shown in Figure 2 and Figure 3 indicate that the OR for carrying 1-ΔRA was within 1.0-1.3. This value corresponds to a modest effect size. 4 Therefore, it is unlikely that heterogeneity in the genes among studies changed the main conclusion of this study.   Talmud 17  2015  65  ADAMTS9, ADCY5, ANK1, ANKRD55, AP3S2, BCAR1, BCL11A, C2CD4A, CCND2, CDC123=CAMK1D, CDKAL1, CDKN2A=B, CENTD2, CILP2,  DGKB, DUSP8, FTO, GCC1, GCK, GCKR, GIPR, GLIS3, GRB14, HHEX=IDE, HMG20A, HMGA2, HNF1A, HNF1B, HNF4A, IGF2BP2, IRS1, JAZF1,  KCNJ11, KCNK16, KCNQ1, KLF14, KLHDC5, MAEA, MC4R, MTNR1B, NOTCH2, PEPD, PPARG, PRC1, PROX1, PSMD6, PTPRD, RBMS1, SLC30A8,  SPRY2, SRR, ST64GAL1, TCF7L2, THADA, TLE1, TLE4, TP53INP1, TSPAN8=LGR5 Qi 44  2010  17  ADAMTS9, CDC123-CAMK1D, CDKAL1, CDKN2A=2B, GCKR, HHEX, HNF1B, IGF2BP2, JAZF1, KCNJ11, KCNQ1, MTNR1B, PPARG, SLC30A8,  TCF7L2, TSPAN-LGR5, WFS1  Rotger 45  2010  4  FTO, KCNJ11, TCF7L2, TSPAN-LGR5  Wang 46  2010  20  ADAMTS9, CDC123, CDKAL1, CDKN2A=2B, FTO, HHEX, HNF1B, IGF2BP2, JAZF1, KCNJ11, KCNQ1, LOC387761, MTNR1B, NOTCH2, PPARG,  SLC30A8, TCF7L2, THADA, TSPAN8, WFS1  Waters 47  2010  19  ADAMTS9, CDC123, CDKAL1, CDKN2A=2B, FTO, HHEX, HNF1B, IGF2BP2,  Nevertheless, the heterogeneity in the magnitude of T2DM risk due to the heterogeneity among the used genes urges us interpret the results with caution. The current stratified meta-analysis of cross-sectional studies indicated that a larger magnitude of T2DM risk in relation to cumulative RAs was observed in studies with a proportion of men 50% or greater compared with women-dominant studies, suggesting that the magnitude of the genetic association with T2DM was larger in men than in women. Although evidence for gender-related differences in genetic associations has been insufficient, 67 some major longitudinal studies suggested that the environmental contribution to T2DM risk differed by gender. For example, regarding moderate physical activity, which has been established to reduce T2DM risk, 68 the MONICA=KORA Augsburg Cohort study suggested that leisure time physical activity was effective in preventing T2DM, especially in women. 69 In the intensive lifestyle modification group in the Diabetes Prevention Program, the effect in preventing incident T2DM did not differ by sex, although men lost significantly more body weight and increased physical activity more than did women. 70 The relative contribution of genetic risk to incident T2DM could be smaller in women than in men. Further studies would need to analyze the T2DM risk in relation to genetic risk by gender to clarify the gender difference in the relative contribution of genetic profiles. Of note, this suggestion was  Figure 2. Forest plot of odds ratio (OR) with 95% confidence interval (CI) for risk of type 2 diabetes mellitus with 95% CI for 1 increment in risk alleles carried in cross-sectional studies. The OR in each study and the overall OR are indicated by squares and a diamond, respectively. Horizontal lines indicate the range of the 95% CI. The area of each square is proportional to the study weight expressed as the inverse of the square of standard error based on a random-effects model.  contradictory to the finding in one of the included studies in this meta-analysis, which indicated that carrying RAs was associated with T2DM in women but not in men. 62 However, this study focused on ADIPOQ among the established T2DM-susceptible genes and limited the subjects to those who already had IGT at cohort entry. ADIPOQ might play a role in the determination of the severity of glucose tolerance, especially in women, although the majority of SNPs used for T2DM screening was associated with discrimination of glucose intolerance from normal glucose tolerance. Current sensitivity analyses indicated that there was a stronger association between cumulative RAs and T2DM risk in study populations with lower mean BMI. This finding was supported by previous studies that stratified the analyses by BMI, although these data could not be included in this meta-analysis because characteristics other than BMI in each subgroup were not presented. 20,25 However, ethnicity or geographic region, rather than obesity, might affect the strength of the association between cumulative RAs and T2DM risk, considering that the prevalence of overweight and obesity in Asia is relatively low compared with Western populations. 71 To our knowledge, no studies have investigated the difference in the magnitude of T2DM risk by ethnicity using two or more SNPs, although a difference in the attributable risk of a certain T2DM-susceptible SNP has been suggested. 72 Whether an included study was targeted on an Asian-dominant ethnicity or whether it was conducted in Western countries affected the magnitude of T2DM risk for an increment in the number of RAs carried in both univariate and multivariate meta-regression analyses. However, modification of the magnitude of T2DM risk by the mean BMI that was shown in univariate analyses disappeared in the multivariate analyses. It could be interpreted that the contribution of genes to T2DM risk could  Figure 3. Forest plot of odds ratio (OR) with 95% confidence interval (CI) for risk of type 2 diabetes mellitus with 95% CI for 1 increment in risk alleles carried in longitudinal studies. The OR in each study and the overall OR are indicated in squares and a diamond, respectively. Horizontal lines indicate the range of the 95% CI. The area of each square is proportional to the study weight expressed as the inverse of the square of standard error based on a random-effects model.

Cumulative Risk Alleles and Type 2 Diabetes Mellitus
Cumulative Risk Alleles and Type 2 Diabetes Mellitus be affected by ethnicity rather than physical background. Unfortunately, when applied to the included studies in the current meta-analysis, this hypothesis was directly elucidated by only one study, which compared the magnitude of T2DM risk between Asian and non-Asian ethnic groups. 47 Further studies are needed to confirm the characteristics of individuals or populations that are susceptible to genetic T2DM risk, including determinations of whether the characteristic was the extent of obesity or ethnicity.
The result of the current stratified meta-analyses might reflect characteristics regarding metabolic traits of persons at high genetic risk of T2DM: a higher OR was observed in studies that excluded subjects without T2DM who had IGT compared with studies that did not exclude subjects with IGT, but no difference   Kodama S, et al. in ORs was observed according to whether or not subjects with IFG were excluded, which meant that inclusion of subjects with IGT in the non-case group weakened the association between cumulative RA and T2DM risk. A previous study reported that the genetic risk score was reported to be cross-sectionally associated with the risk of having IGT but not IFG using nine genes (FTO, HHEX, KCNJ11, KCNQ1, MTNR1B, PPARG, SLC30A8, TCF7L2, WFS1), which were established to be major loci associated with the risk of T2DM. 73 When applying this finding to this meta-analysis, seven of these nine genes were ranked in the top 10 genes that frequently appeared in the included studies in this meta-analysis. Major T2DM-associated genes that were discovered in the current GWAS might have a stronger association with glucose intolerance than T2DM itself. Considering that T2DM is a partly a result of progression of glucose intolerance, incident T2DM could be prevented by nongenetic factors, although mild glucose tolerance was inevitably regulated by genes. This suggestion was supported by evidence for the benefit of lifestyle interventions in reducing the risk of progression from IGT to T2DM. 74 Several limitations should be addressed. First, this metaanalysis had to assume the log-linearity between the number of RAs carried and the observed T2DM risk. Second, this metaanalysis assumed that each SNP within one study would equally contribute to the risk of T2DM, although the OR for each SNP varied. However, it was reported that the difference in the discriminative power between using an unweighted and weighted genetic risk score was not significant 43 or was modest. 51 Third, as previously mentioned, the genes that were used in the examination of the association with prevalence and incidence of T2DM were too heterogeneous to analyze by the combination of the used genes. Nevertheless, the genes used for genetic risk should be the same across the studies. Fourth, this meta-analysis also did not consider heterogeneity in covariates for which each study adjusted T2DM risk. To minimize this heterogeneity, analyses were limited to the ORs that were adjusted for at least two of three covariates: age, gender, and BMI. In addition, sensitivity analyses confirmed that the magnitude of T2DM risk was not influenced by whether these ORs were adjusted for BMI. Therefore, the current meta-analysis could confirm that the genetic T2DM risk was independent of age, gender, and obesity. Although these are well known to be major classic risk factors for T2DM, it needs to be emphasized that other residual confounders could not be ruled out.

Conclusions
The current meta-analysis indicated that carrying one RA in T2DM-associated SNPs was associated with a modest risk of prevalent or incident T2DM as far as the SNPs discovered in GWAS were concerned, although the heterogeneity in the genes examined among studies indicates that we should interpret the results with caution.