2.1. Literature search
To identify studies eligible for this meta-analysis, the computerized search was conducted on three online electronic English databases (Medline, Embase, and Web of Science) and one online Chinese CNKI database using the following key words: “DRD3”, “dopamine receptor 3”, “dopamine D3 receptor”, “dopamine receptor D3”, “schizophrenia”, and “Ser9Gly”. We also screened the reference lists of the accessed articles and of potentially relevant review articles to identify additional studies.
2.2. Inclusion criteria
Only the studies examining Ser9Gly SNP were included in the present meta-analysis. Moreover, the studies needed to meet the following inclusion criteria: (1) family-based design (original transmission disequilibrium test (TDT)  or haplotype-based haplotype relative risk (HRR)); (2) original data, or available data to calculate an effect; (3) independent from other studies (i.e., studies reported by the same authors contained the same or overlapping data, the latest literature was selected). Using this approach, a total of 13 articles were identified and included in our meta-analysis. The flow diagram of the literature search process was showed in Fig. 1.
2.3 Data extraction
According to the inclusion criteria listed above, two authors extracted information from all eligible publications independently. Any disagreement was resolved through discussion until the two authors reached a consensus. The following details of each article were recorded: the first author’s last name, publication year, location, ethnicity, diagnostic criteria, and numbers of transmissions.
2.4. Meta-analytic methods
The meta-analysis of the family-based association studies was divided into two parts: TDT and HRR. For the TDT study, each study provided the two-by-two transmission disequilibrium table, which classifies heterozygous parental alleles (informative meioses) by transmission status (Ser9 allele transmitted to the schizophrenic offspring) and data type (the number of observed transmission vs. the number of theoretic transmission). For one informative meiosis, the expected transmitted number that the allele is transmitted from heterozygous parents to the proband is 0.5 and the expected untransmitted number that the allele is not transmitted from heterozygous parents to proband is also 0.5. For the HRR studies, each study provided the two-by-two HRR table, which classifies parental alleles by type of allele (Ser9 or Gly9) and transmission status (transmitted to the schizophrenic offspring or not).
The degree of heterogeneity between studies was determined by means of the Q statistic.[37, 38] Specifically, P > 0.05 by the Q test indicated the absence of heterogeneity, and P < 0.05 indicated heterogeneity. I2 was defined as the proportion of observed variance in effect sizes attributable to true differences among studies. Conventional interpretations of I2 include limits for low (<25%), moderate (approximately 50%), and high (>75%) heterogeneity. A random effect model was used when heterogeneity was present (p<0.05 and/or I2>50%); otherwise, a fixed effect model was applied and the fixed effect model used the method of Mantel and Haenszel .
For the pooled analysis, odds ratios (ORs) with accompanying 95% confidence intervals (CIs) were used to assess the strength of the association in the two-by-two tables. P > 0.05 indicated the absence of statistical significance, and P < 0.05 indicated statistical significance. When P < 0.05, OR < 1 meant the variation as a protective factor, and OR > 1 meant the variation as a risk factor. Pooled calculations of ORs were obtained and compared with the controls (observed transmission vs. expected transmission for TDT study or transmitted vs. untransmitted for HRR study) using test statistic z and 95% CIs. Subgroup analysis were carried out by ethnicity (i.e., East Asian, Caucasian, and other populations) and diagnostic criteria (i.e., DSM-III-R, DSM-IV, and CCMD-III).
Publication bias was assessed by the funnel plot (the standard normal deviate of the OR is regressed on the precision of the OR). When there is no publication bias, the regression line should pass through the origin, and the expected value of intercept will be zero.
All the calculations of the meta-analysis were conducted by Comprehensive Meta Analysis V2 software (Biostat, Englewood, NJ, USA).