|
nif:isString
|
-
Papers included in the systematic review were retrieved from Medline (http://www.ncbi.nlm.nih.gov/PubMed) and ISI web of Knowledge (http://wok.mimas.ac.uk/) using the search terms: “Aldehyde dehydrogenase 2” and “ALDH2” in combination with “Hypertension”, “Blood pressure”, “Cardiovascular diseases”, “Coronary disease”, “Heart disease”, “Coronary artery disease”. The bibliographies of retrieved articles were also scanned for relevant publications. We searched for all papers published before October 2006. Screening of studies for inclusion was performed by two researchers (LC, SJL) independently of each other to increase objectivity. We also searched for all studies of the ALDH2 genotype regardless of outcome for use in separate analyses, and we scanned the abstracts of these studies to determine whether they may include any of the outcomes we were interested in. If we were uncertain we scanned the paper and then contacted the authors. There was no exclusion on the basis of language. For multiple publications based on the same study, the latest publication was included. For inclusion we required, as a minimum, data on systolic or diastolic blood pressure by genotype or prevalence of hypertension by genotype. Where insufficient data for inclusion were available in the paper we contacted authors directly, and if sufficient information could still not be obtained the study was excluded. A flow diagram showing the number of studies found in our search strategy, and reasons for exclusion is given in Figure 1.
Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pmed.0050052.g001 Flow Diagram Showing Reasons for Exclusion and Number of Papers Excluded We followed a standard protocol for data extraction. For each study, the following data were recorded (where available): first author's name, year of publication, country and city in which the study was performed, name of the study, study design, number and source of participants, sex, method of assessment of drinking status, categories of alcohol drinking, mean alcohol drinking and standard deviation by ALDH2 genotype, distribution of potential confounding factors by genotype with p-values, distribution of genotypes among hypertensive and nonhypertensive participants, mean blood pressure and standard deviation by ALDH2 genotype, and reported effect estimates (i.e., relative risks, odds ratios, or mean difference of blood pressure level) and 95% confidence intervals (CIs) for ALDH2 genotype and hypertension risk. Data extraction was carried out by two researchers (LC, SJL) independently of each other.
Data on alcohol intake were converted to a uniform measure of grams per day for comparison across all studies with 1 unit = 10 ml = 7.9 g [10]. We calculated the unadjusted mean difference in blood pressure or the odds ratio of hypertension by comparing *1*2 and *1*1 to *2*2 genotypes. In the analyses of hypertension, fixed-effects meta-analysis was used to calculate summary odds ratio estimates, and in the analyses of diastolic and systolic blood pressure, mean differences between genotype groups were obtained. We carried out fixed-effects meta-analyses separately for men and women, excluding two very small studies (Nishimura [11], n = 36, and Mackenzie [12], n = 28) that did not provide data separately by sex, and two further small studies [13,14] for which we were unable to obtain estimates of blood pressure separately for *1*2 and *2*2 genotypes. All statistical analyses were carried out with the use of Stata statistical software (version 9.2; Stata Corporation). All statistical tests were two-sided. We assessed evidence of small-study effects (which may include publication bias) for studies of ALDH2 polymorphism and hypertension risk, and for ALDH2 and blood pressure by computing both the Egger [15] and Begg [16] tests. For the studies that reported both blood pressure and mean alcohol intake by genotype, we calculated estimates of the causal effect of alcohol on blood pressure, using the method of instrumental variables with genotype as the instrument assuming the effect is linear. In order to use standard instrumental variable estimation methods (‘two-stage least squares') [17] the corr2data command in Stata was used to construct an artificial dataset matching the reported number of participants with each genotype, and the means and standard deviations of alcohol intake and blood pressure within each genotype. One study reported the mean but not the standard deviation of alcohol intake by genotype [18], so it was estimated from a histogram. As the correlation between alcohol intake and blood pressure within each genotype was not available, this correlation was assumed to be zero, which is slightly conservative, as the standard error of the estimated effect of alcohol on blood pressure reduces slightly as this correlation increases. As a sensitivity analysis we repeated the analysis assuming a correlation of 0.2 in all genotypes and all studies. Instrument strength was assessed using the first-stage F-statistic; values considerably above 10 are generally taken to show sufficient instrument strength to ensure the validity of two-stage least-squares [17]. The resulting estimates were then meta-analysed using the same methods as above.
|
![[This page as RDF]](http://data.gesis.org/somesci/static/rdf-icon.gif){kind=icon}