The M235T Polymorphism in the AGT Gene and CHD Risk: Evidence of a Hardy-Weinberg Equilibrium Violation and Publication Bias in a Meta-Analysis

Background The M235T polymorphism in the AGT gene has been related to an increased risk of hypertension. This finding may also suggest an increased risk of coronary heart disease (CHD). Methodology/Principal Findings A case-cohort study was conducted in 1,732 unrelated middle-age women (210 CHD cases and 1,522 controls) from a prospective cohort of 15,236 initially healthy Dutch women. We applied a Cox proportional hazards model to study the association of the polymorphism with acute myocardial infarction (AMI) (n = 71) and CHD. In the case-cohort study, no increased risk for CHD was found under the additive genetic model (hazard ratio [HR] = 1.20; 95% confidence interval [CI], 0.86 to 1.68; P = 0.28). This result was not changed by adjustment (HR = 1.17; 95% CI, 0.83 to 1.64; P = 0.38) nor by using dominant, recessive and pairwise genetic models. Analyses for AMI risk under the additive genetic model also did not show any statistically significant association (crude HR = 1.14; 95% CI, 0.93 to 1.39; P = 0.20). To evaluate the association, a comprehensive systematic review and meta-analysis were undertaken of all studies published up to February 2007 (searched through PubMed/MEDLINE, Web of Science and EMBASE). The meta-analysis (38 studies with 13284 cases and 18722 controls) showed a per-allele odds ratio (OR) of 1.08 (95% CI, 1.01 to 1.15; P = 0.02). Moderate to large levels of heterogeneity were identified between studies. Hardy-Weinberg equilibrium (HWE) violation and the mean age of cases were statistically significant sources of the observed variation. In a stratum of non-HWE violation studies, there was no effect. An asymmetric funnel plot, the Egger's test (P = 0.066), and the Begg-Mazumdar test (P = 0.074) were all suggestive of the presence of publication bias. Conclusions/Significance The pooled OR of the present meta-analysis, including our own data, presented evidence that there is an increase in the risk of CHD conferred by the M235T variant of the AGT gene. However, the relevance of this weakly positive overall association remains uncertain because it may be due to various residual biases, including HWE-violation and publication biases.


Introduction
Angiotensinogen (AGT) is a liver protein that interacts with renin to produce angiotensin I, the pro-hormone of angiotensin II. Angiotensin II is the major effector molecule of the reninangiotensin-aldosterone system (RAAS) and plays a key role in the regulation of blood pressure (BP) by increasing vascular tone and promoting sodium retention. Genetic variants in the angiotensinogen gene modify the plasma concentration of angiotensinogen, which has been directly related to arterial blood pressure [1]. The molecular variant (M235T) of the AGT gene, encoding a threonine instead of a methionine at residue 235 of the mature protein, has been associated with a higher plasma AGT level and higher BP in patients homozygous for the T allele and occurs among various ethnic populations [1][2][3]. In a meta-analysis, the TT genotype was associated with a 32% increase in the risk of hypertension in white people but not in non-white people, when compared with the MM genotype [4].
Given the importance of hypertension in the occurrence of coronary heart disease [5], this finding suggests that this polymorphism may be related to increased risk of CHD. A few studies [6][7][8], including recent publications, [9,10] have found that there is an association of the M235T AGT variant with increased CHD risk; however, this relationship was not confirmed in several other studies [11][12][13] as well as in a meta-analysis [14]. Marked ethnic differences in the frequency of the T allele, small sample sizes and genotyping or phenotyping errors could partly account for discrepancies among these gene-disease association studies. Therefore, we investigated the association of the M235T polymorphism in the AGT gene (National Center for Biotechnology Information single nucleotide polymorphism cluster ID rs699) with acute myocardial infarction (AMI) and CHD in a large population-based cohort of middle-aged Dutch women and conducted an updated meta-analysis of the available studies to clarify the role of the M235T polymorphism in CHD risk.

Case-cohort study
Study design, general questionnaire, anthropometric and laboratory measurements have been described in detail elsewhere [15][16]. Briefly, the study population consisted of participants of the Prospect-EPIC cohort. Participants were recruited between 1993 and 1997 among women living in Utrecht and the vicinity who attended the regional population-based breast cancerscreening program. A total of 17,357 women, aged 49-70, were included. At baseline, a general and a dietary questionnaire were administered, a limited physical examination was performed and a non-fasting blood sample was taken. Follow-up event information was obtained from the Dutch Centre for Health Care Information, which holds a standardized computerized register of hospital discharge diagnoses. Using the International Classification of Diseases, ninth Revision (ICD-9) codes for the main discharge reason, we categorized cardiovascular disease (codes 390-459) as CHD (codes 410-414), including AMI (code 410), and other cardiovascular diseases. Whenever multiple events (AMI and CHD) occurred, the first occurrence of that endpoint was taken as the endpoint of interest in endpoint-specific analyses. All women signed an informed consent form prior to study inclusion. The study was approved by the Institutional Review Board of the University Medical Center Utrecht.
We applied the case-cohort design introduced by Prentice [17]. From the 17,357 women in the total cohort, we randomly selected a sample of 10% as the sub-cohort (n = l736). Women who did not consent to linkage with vital status registries or who were not traceable (cases n = 3/sub-cohort n = 38) were not included. Women who reported a diagnosis of cardiovascular disease (ICD-9; 390-459) at baseline or who had missing questionnaires, blood, or DNA samples were excluded. This resulted in 15,236 women in the total cohort and 1522 women in the sub-cohort (as the control group) at baseline. All individuals with first fatal and non-fatal CHD and ischemic stroke events that arose during follow-up until January 1 st 2000 were selected as cases. These were 211 CHD cases, including 71 AMIs. For all case subjects, followup ended at the date of diagnosis or at the date of death due to cardiovascular disease.
Genetic analysis. Genetic analysis was performed at the Cardiovascular Genotyping (CAGT) laboratory of the Department of Internal Medicine of the University Hospital Maastricht. Genomic DNA was extracted from buffy coats using the QIAampH Blood Kit (Qiagen Inc., Valencia, California, USA). Genotyping of the polymorphisms was performed using a multilocus genotyping assay for candidate markers of cardiovascular disease risk (Roche Molecular Systems Inc., Pleasanton, CA, USA) [18]. Briefly, each DNA sample was amplified using two multiplex polymerase chain reactions, and the alleles were genotyped simultaneously using an array of immobilized sequence-specific oligonucleotide probes. This array of probes was blotted on plastic strips, and, after staining, genotypes were scored based on blue (positive) and white (negative) bands. Each blue band, representing a specific genotype, was scored by specific software (counting the pixel intensity of each band) and checked manually. Genotyping was performed blinded to the case-control status. A random doublecheck was performed to detect potential genotyping errors in a subset of 100 samples. The check confirmed the previous genotyping results by 100%.
Data analysis. Hardy-Weinberg equilibrium (HWE) was tested with the x 2 test among the controls. Allele frequencies were estimated by gene counting. We used the ANOVA F test to estimate relationships among the M235T genotypes and continuous variables, while we tested the significance of any difference in proportions by applying the x 2 statistic. A p-value ,0.05 (2-sided) was considered statistically significant.
To assess the relationship of the M235T polymorphism in the AGT gene with the outcome, we used a Cox proportional hazards model with an estimation procedure adapted for case-cohort designs. We used the unweighted method by Prentice [17,19], which is incorporated in a SAS macro at http://lib.stat.cmu.edu/ general/robphreg. A previous meta-analysis [14] showed that the effect of the AGT M235T variant on its intermediate phenotype (plasma angiotensinogen level) follows an additive model according to the number of T alleles [5% (95% CI: 2 to 8%) increase for the MT and 11% (95% CI: 7 to 15%) increase for the TT genotype versus the MM genotype]. Therefore, our priori hypothesis was that the association between the M235T polymorphism in the AGT gene and CHD follows an additive model according to the number of T alleles. However, other genetic models were evaluated as well. We considered different modes of inheritance as follows: the additive ''per-allele'' model, the T allele was compared between cases and controls by assigning scores of 0, 1, and 2 to homozygotes for the M allele, heterozygotes, and homozygotes for the T allele, respectively; the recessive model, the TT genotype versus the MT and MM combined genotypes; and the dominant model, the MT and TT genotypes combined versus the MM genotype. We also performed separate pairwise comparisons of the MT and TT genotypes versus the MM genotype.

Meta-analysis
Searching. We searched PubMed/MEDLINE, Web of Science, and EMBASE up to February 2007 for observational studies evaluating an association between the M235T polymorphism in the AGT gene and CHD. Terms used for the search contained both medical subject heading terms and text words: (Met235Thr OR M235T OR T704C) AND (angiotensinogen OR AGT) AND (polymorphism OR mutation OR genetic OR genotype) AND (''coronary disease'' OR ''coronary heart disease'' OR CHD OR ''myocardial infarction'' OR MI OR ''myocardial infarct'' OR ''coronary artery disease'' OR CAD OR ''ischemic heart disease'' OR IHD OR ''cardiovascular disease'' OR ''heart disease'' OR angina). We also retrieved additional studies by hand searching the bibliographies of original research reports and review articles and through the MEDLINE option ''related articles''. Search results were limited to articles published in English and studies on human subjects.
Selection. All studies were considered potentially eligible if they aimed to investigate the relationship between the M235T genotypes and risk of CHD or MI. Any observational study, regardless of sample size, which fulfilled the following criteria, was included: (i) AGT M235T genotype frequencies were provided by case-control status (studies without controls were excluded); (ii) risk of CHD or MI was evaluated (studies on recurrent coronary events were excluded); (iii) relevant data were presented to calculate the effect size and its 95% CI; (iv) non-overlapping data were contained. For duplicate publications, the study with the smaller data set was excluded.
Data abstraction. The following information was extracted from each study that we included: the first author's name; country; year of publication; the population evaluated; study design; mean age or age range for case-patients and controls; definition and number of cases and controls; allele frequencies and genotype distribution in case-patients and controls (where data were not given, they were calculated from the corresponding genotyping frequencies of the case and control groups); consistency of genotype frequencies with HWE (calculated); gender in the evaluated population and male percentage, matching variables, use of blinding of genotyping staff, performing regenotyping of a random sample, and crude ORs and 95% CIs for development of CHD or MI related to the AGT gene genotypes based on different genetic models (from the original paper or calculated from crude data if not provided). We again considered a dominant, a recessive, an additive ''per-allele'' model and pairwise comparisons. Data were extracted independently and entered into separate databases by two authors (performed by MHZ and MLB). Results were compared, and disagreements were resolved by a consensus.
Quantitative data synthesis. The method of Mantel-Haenszel was used to calculate the odds ratio for the pooled data in a fixed-effects model, and, if there was evidence for heterogeneity, the DerSimonian-Laird method was used for the pooled odds ratio in a random-effects model, under pairwise comparisons of the different genotypes and dominant, recessive, and additive inheritance models. For all the models used, the T allele was considered the risk allele. The genetic model to be considered as the priori hypothesis was the additive model. In each study, we tested for HWE by using the x 2 test or an exact test among the controls by using the genhwi command in Stata 9.2 [20].
In addition, we used Cochran's x 2 -based Q statistic for between-study heterogeneity, which is considered to be significant for P,0.10, as well as the I 2 statistic for estimation of inconsistency in meta-analyses [20]. I 2 represents the percentage of the observed between-study variability due to heterogeneity rather than to chance. It ranges between 0% and 100%, where a value of 0% indicates no observed heterogeneity, and larger values indicate an increasing degree of heterogeneity (roughly suggested cut-off points include: I 2 = 0-25%, no heterogeneity; I 2 = 25-50%, moderate heterogeneity; I 2 = 50-75%, large heterogeneity; I 2 = 75-100%, extreme heterogeneity) [21].
We used funnel plots to examine the publication bias of reported associations. We also used Egger's test and the Begg-Mazumdar test with 95% CI for evaluation of publication bias, which are considered to be significant for P,0.10. Meta-analysis was carried out using STATA 9.2. We used random effect metaregression models with restricted maximum likelihood estimation to evaluate the extent to which different variables explained heterogeneity among the individual ORs. The pre-specified characteristics for assessment of sources of inter-study heterogeneity were: study size (for detailed definition see [22]); ethnicity of population evaluated (of Caucasian descent, East Asian, and others); male percentage in each study, matching (matched or unmatched); blinding of genotyping staff (blinded, or not reported); performing regenotyping of a random sample (performed or not reported); violating HWE (violated or confirmed; the term ''violated'' used for statistically significant deviation of HWE) in sub-group analysis as well as in meta-regression analysis.
HWE Correction. For evaluating the impact of HWEviolated studies on effect estimates (at the 0.05 significance level) under different genetic models, odds ratios, and variances were corrected by using the HWE-predicted genotype counts in the control instead of the observed counts as previously suggested [20]. Thereafter, they were included in the sensitivity analysis.

Prospect-EPIC study results
The general characteristics of the randomly sampled participants of the cohort (N = 1522) are given in Table 1. The genotype distribution was in Hardy-Weinberg equilibrium (x 2 = 0.020; P = 0.89). General and clinical characteristics of CHD cases and controls are shown in Table 1. The median follow up time for the random sample was 4.3 years, with a total of 6,523 person years. The actual follow-up in the baseline cohort of 15,236 women was 64,768 person years. Due to the case-cohort design, 23 women in the sub-cohort eventually were CHD cases (among which there were nine AMI cases).
Due to the association of the M235T genotypes with some risk factors of CHD, we presented crude models and models adjusted for hypertension, total cholesterol and waist to hip ratio as potential confounding factors. Table 2 presents hazard ratios of AMI and CHD under different genetic models. Under the additive model of inheritance, no increased risk for CHD was found (HR = 1.20; 95% CI, 0.86 to 1.68; P = 0.28), which did not alter after adjustment (HR = 1.17; 95% CI, 0.83 to 1.64; P = 0.38). The same was true for other comparisons ( Table 2). Analyses for AMI risk did not show any statistically significant associations ( Table 2).

Meta-Analysis results
Flow of included studies. A total of 44 gene-disease association studies, including the present study, evaluating the AGT M235T gene variant and CHD risk were identified. Seven articles were excluded, three of which were duplicate publications [12,23,24], three of which did not provide relevant data [25][26][27], and one of which studied the risk of recurrent coronary events [28]. Finally, 37 studies met the selection criteria. In one paper, the provided results were based on two different studies [6], so both were included in the meta-analysis. Therefore, 38 studies with 13,284 cases and 18,722 controls were included in the final meta-analysis ( Figure 1).

Study characteristics
Characteristics of the studies are shown in Table 3 [6][7][8]10,11,13,. There were 25 studies in Caucasians, eight studies in East Asians, and five studies in other populations (West Asian, South Asian, African, African-American, and South American). The last was collapsed into a miscellaneous group. The design of the studies was case-control, except for three studies that were prospective cohort [56], case-cohort (present study), and cross-sectional [40]. The T allele frequency varied from 26 to 54 percent in Caucasians, 65 to 91 percent in East Asians, and 34 to 83 percent in the miscellaneous group.
All studies used polymerase chain reaction methods for genotyping, and most used a restriction fragment length method for polymorphism analysis. Blinding of investigators involved in genotyping with respect to the case/control status of the participants was reported in six studies [8,32,50,51,56]. A random double-check to detect potential genotyping errors was mentioned in five studies [37,50,53,56]. In most of the studies, the genotype frequencies were consistent with HWE. However, statistically significant deviations from HWE were found in five studies

Quantitative data synthesis
The overall OR under a random-effects model using an additive model for CHD risk was 1.08 (95% CI, 1.01 to 1.15; P = 0.025; Figure 2). However, there was evidence of substantial betweenstudy heterogeneity (I 2 = 55.5%, P,0.001). Table 4 shows the association of the AGT T235M polymorphism with CHD risk under different genetic contrasts. When a recessive model was evaluated, a significant association was found between individuals homozygous for the T allele (T235T genotype) and CHD risk, when compared to carriers of the M allele (OR = 1.11; 95% CI, 1.02 to 1.22; P = 0.016). Under the dominant model, the association was not significant. Under pairwise comparisons, there was a significant modest association between the T235T genotype and CHD risk, as compared with the M235M genotype (OR = 1.15; 95% CI, 1.00 to 1.32; P = 0.045). There was evidence for moderate to large between-study heterogeneity under all models (Table 4). Sub-group analysis, by study characteristics under the additive model, showed that matching, blinding of genotyping staff, and regenotyping of a random sub-sample explained little of the heterogeneity. However, stratification showed an attenuated effect estimates in the large studies, in studies that CHD was defined based on angiography or WHO criteria, and in particular in studies that were in HWE (Table 5). Further evaluation of potential sources of the heterogeneity was performed using a meta-regression analysis.

Meta-regression
First, an empty regression was run with only the log of the effect estimate of pooled studies under the additive model to determine the baseline value for t 2 , an estimate of between-study variation (baseline t 2 = 0.025). Next, single covariates were added in a series of univariate models. We performed the regression analysis for ten pre-defined potential sources of heterogeneity, including ethnicity, sex, mean age of cases, study size, case definition, source of controls, HWE-violation, blinding in genotyping, performing a sub-sample regenotyping, and matching (we hypothesized that studies that used matching might produce more conservative estimates of association). Univariate regression analyses showed that violation of HWE (b coefficient = 0.27 (0.06 to 0.48); P Het = 0.015, t 2 = 0.019), the mean age of cases (b = 20.01 (20.02 to 0.0008); P Het = 0.066, t 2 = 0.024), and the method of case definition, clinically diagnosed CHD versus WHO criteria adjusted for other definitions (b = 0.26 (0.02 to 0.50); P Het = 0.038, t 2 = 0.020), were significant sources of heterogeneity among studies. The study size (P Het = 0.241, t 2 = 0.024), the ethnicity (P Het = 0.591, t 2 = 0.025), the male percentage in the study (P Het = 0.701, t 2 = 0.029), blinded genotyping (P Het = 0.890,  Table 3. Characteristics of published studies of the association between the M235T polymorphism in AGT gene and CHD included in the meta-analysis.     Table 3. Cont. t 2 = 0.026), sub-sample regenotyping (P Het = 0.131, t 2 = 0.023), the source of controls (P Het = 0.640, t 2 = 0.025), and matching (P Het = 0.942, t 2 = 0.026) were not significant sources of heterogeneity among studies. Violation of HWE in multivariable regression analysis remained a statistically significant source of heterogeneity after adjustment for the effect of study size (P Het = 0.031, t 2 = 0.020). Adding the mean age of cases and method of case definition to the model with violation of HWE decreased the t 2 value to 0.017 (P Het = 0.073 for violation of HWE, P Het = 0.057 for the mean age of cases, and P Het = 0.162 for clinically diagnosed CHD). It also showed that the effect of method of case definition on the variation among the studies was through the effect of the mean age on the heterogeneity and not as an independent factor. A model that included only violation of HWE and the mean age of cases reduced the t 2 value to 0.018 (P Het = 0.019, and 0.052, respectively).

Sensitivity Analysis
First, the influence of deviation from the HWE on effect estimates was examined by using HWE-deviated adjusted ORs. Table 6 presents the genotype-based contrasts with corrected ORs, as well as the allele-based contrast. After adjustment, a smaller overall effect was seen under the additive, dominant, and pairwise comparisons. Moreover, after adjustment, the previously significant association under the additive model, as well as the TT vs. MM comparison, was no longer statistically significant. The association under the recessive model still remained significant. Figure 3 shows a funnel plot in which the log of the OR of CHD risk under the additive genetic model was plotted against the standard error of the log of the OR in each study. The funnel plot for the overall results was substantially asymmetric for small negative studies. Moreover, tests for potential publication bias (The Egger's test and the Begg-Mazumdar test; P-values equal to 0.066 and 0.074, respectively) suggested the presence of a publication bias. By using the trim and fill method, we showed that, if the publication bias was the only source of the funnel plot asymmetry, it needed seven more studies to be symmetrical ( Figure 4).

Prospect-EPIC study
In this prospective study of healthy women aged 49 to 70 years, we investigated the relationship between the M235T polymorphism in the AGT gene and risk of AMI and CHD later in life. Under the additive genetic model, increased risks, albeit not statistically significant, were found for the incidence of AMI and CHD, which did not alter after adjustment. Likewise, we did not find a clear association between the variant and risk of CHD or AMI using different genetic models. This may be explained by: (i) the absence of a biological effect, (ii) the presence of real genetic heterogeneity according to ethnic background, or (iii) failure to detect a small effect because the epidemiologic risk for an individual genetic variant is likely to be small and a large sample size is needed for adequate statistical power. It has been commonly proposed that, as well as a need for much larger and more rigorous studies those that are currently used, there is a greater need for international collaborations, particularly for a complex disease like CHD [59].
Strengths and limitations. In our study, the data collection was prospective, before the diagnosis of AMI or CHD and equal for all participants. This ensures that the cases and the randomly selected controls are comparable [17]. For a multifactorial trait, like CHD, this provides a valid approach to evaluate the Author End point Case definition

Source of controls
Tsai et al. [54] CHD .50% stenosis of at least one coronary vessel CHD-free group documented by angiography Frossard et al. [39] CHD, MI CHD: Exertional angina, unstable angina or MI MI: ECG changes; presence of regional wall motion abnormalities on trans-thoracic echocardiography; and serial enzyme elevations  Table 3. Cont.
relationship between genetic factors and the risk of AMI and CHD, while taking into account co-existing and risk-modifying factors. In this study, prevalent cases of CHD were excluded from the analyses to prevent introducing bias due to potentially selective survival. The Prospect study was a population-based cohort, which makes it less susceptible to selection bias. Additional strengths were the comprehensiveness of our data and sample collection, as well as the morbidity and mortality follow-up for the entire cohort through linkage with nation-wide registries. The case-cohort design of the study combined the advantages of cohort studies (multiple outcomes and time-dependent covariates) with those of case-control analyses (fewer subjects); thus, it was more efficient than cohort studies. Classical case-control studies might be affected by selection bias since only non-fatal cases can be included, which was not the case in this study because of our endpoint definition. Moreover, we did not have misclassification of exposure (genotypes), which, when present, generally lead to a bias toward the null because we used standard laboratory protocols,  performed a random double-check to detect potential genotyping errors, and had our AGT genotypes in Hardy-Weinberg equilibrium. The limitations of this study were the relatively short period of follow-up and the small number of cases. Moreover, because this cohort was exclusively composed of Dutch women, these results cannot be generalized to men or other ethnic groups, for whom the rates of the events or the allele frequency are known to differ.

Meta-Analysis
The current meta-analysis, which includes new data from a prospective study in a large population-based cohort of Dutch women, represents a comprehensive evaluation of the M235T variant of the AGT gene in CHD risk. Although a pooled per-allele OR was suggestive of a modest increase in the risk of CHD of 1.08 (95% CI, 1.01 to 1.15), the robustness of this summary estimate is uncertain. First, in the pre-specified sub-groups analyses in the meta-analysis, larger studies, those with validated genotyping quality controls, and studies that used standardized criteria for case definition did not provide strong evidence for a positive statistically significant association between the M235T variant of the AGT gene and CHD risk. Second, the meta-regression analysis revealed that the HWE violation was a significant source of the moderate to large heterogeneity in the meta-analysis. Taking violation of HWE into account in the meta-analysis decreased the overall effect (Table 5). Third, the previous result was confirmed by using HWE-deviation adjusted ORs in the meta-analysis   (Table 6). Moreover, there was evidence for publication bias in the meta-analysis. Taken together, these findings point to a violation of HWE and publication biases as the potential explanations for the results observed in the meta-analysis. Some aspects of the current meta-analysis need to be considered to appreciate the findings. First, it might not be very practical to adjust for violation of HWE in the studies that mentioned that the violation is not due to genotyping errors. However, in the current meta-analysis, the HWE-violated studies that were included in the pooled estimate did not provide any reason for the violation. Therefore, we performed sensitivity analyses by using HWEadjusted ORs and corresponding variances. Thereafter, a smaller overall effect was seen under most of the genetic models. Second, the power of tests for HWE and the power to detect genotyping errors are low. Therefore, the inability to detect a deviation from the HWE does not mean that there is no deviation, nor does it rule out the presence of genotyping errors, especially for small sample sizes. Third, our meta-analysis was based on published studies and we did not have access to the original data. However, it could be possible that an association between the genotype and disease exists in certain contexts rather than in all people studied. For example, a case-control study showed that the TT genotype was associated with an increased risk of CHD and MI only in smokers [33]. Finally, in all meta-analyses of gene-disease association studies, the inclusion criteria of cases and controls can be a potentially confounding factor. In this meta-analysis, cases were well defined and the source of controls was not a significant source of variation. However, the advantages of this study were the large sample size of the meta-analysis of 38 studies with 13284 cases and 18722 controls, which was twice the number of studies and sample sizes that had been reported in the previous meta-analysis [14], the exploration of potential sources of heterogeneity in the metaanalysis, and the evaluation of the association under different modes of inheritance.
Approximately 10% of gene-disease association studies are affected by statistically significant deviation from HWE, which could result from genotyping error, chance, inbreeding, nonrandom mating, differential survival of marker carriers, genetic drift, population stratification, or a combination of these reasons [20,60]. Of these, genotyping error could be avoided by using standard genotyping methods and performing quality assessment. It has been recommended that authors specify the quality measures for the genotyping analysis, such as the blinding of laboratory staff to the donor subjects and hypotheses being investigated, procedures for establishing duplicates, degree of reproducibility between quality control replicates, and the inspection for conformity to HWE [61]. In the current meta-analysis, in studies where the blinding of genotyping staff was not reported, a statistically significant increased risk of CHD was found, while those that used blinding methods did not find a significant association. Moreover, for studies without regenotyping of a random sub-sample, a significant increase in CHD risk was found, but not for studies that performed regenotyping. Although overlapping confidence intervals for before-mentioned risks indicate caution in any interpretations, no report on blinding and regenotyping can point towards an uncertainty in quality control of genotyping in these studies. However, violation of HWE, which tends to inflate the chance of a false positive association, may be the strongest indicator of genotyping error [62].
Violation of HWE cannot solely explain the observed betweenstudy variation in gene-disease association studies. The large between-study heterogeneity presented in most meta-analyses could be due to true heterogeneity (i.e., racial differences or differences in gene-environment interactions among various populations) or bias [63]. Bias, which could invalidate the results of the studies, should, therefore, be explored in detail. Biological plausibility, publication bias, selection bias, biased definition of cases, biased selection of controls, and population stratification should be assessed [63]. In this meta-analysis, we found strong evidence for publication bias. This is said to occur when the chance of the publication of a smaller study increases when it shows a stronger effect. Further exploration for sources of biases among studies showed that the selection of controls was not biased. However, using different case definitions resulted in a significant difference in the risk of CHD between those studies using WHO criteria and those using clinically diagnoses of CHD. Studies using definition of cases based on coronary angiography or based on WHO criteria had the same results. Considering a multivariate model in the meta-regression results, case definition was not a significant source of bias in the meta-analysis, while the different mean age of cases and violation of HWE were significant sources of heterogeneity. Since increasing age is a risk factor for CHD and the mean age of cases in the included studies ranged from 42 to 67 years, it is more likely that the studies with older individuals would show a stronger effect and produce heterogeneity. As case-parental controls, or other family-based designs, and genomic controls, using unlinked genetic markers which have no effect on the risk of CHD, were not available to evaluate the potential problem of population stratification among the studies, we presented effect estimates by different ethnic groups. However, there is controversy about the potential importance of population stratification for genetic-association studies using unrelated subjects [64].
In conclusion, the present meta-analysis, including our own data, indicated that, although a weak association between the M235T variant in the AGT gene and CHD was found, the relevance of this weakly positive overall association remains uncertain because it may be due to various residual biases. Moderate to large heterogeneity was identified between studies, and violation of HWE and the mean age of cases were statistically significant sources of the observed variation.