Introduction

Lung cancer is the most common cancer worldwide and the most common causes of cancer death are cancers of the lung and bronchus in both man and woman1, 2. About 80% of lung cancer was non-small cell lung cancer (NSCLC), which is diagnosed at an advanced stage with approximate 15% of the 5-year survival rate3. Current studies showed that the prognosis of NSCLC was contributed to patients’ clinical status and genetic factors such as TNM staging, surgery, chemotherapy drugs, genetic heterogeneity including EGFR, KRAS, PIK3CA, ALK et al.4, 5. Chemotherapy is the main conventional and useful therapeutic method for advanced and metastatic tumors6, 7. NSCLC accounts for approximately 70% of patients harboring advanced stages at the time of diagnosis and chemotherapy is the important treatment strategy for NSCLC patients8, 9. Platinum-based chemotherapy is one of effective treatments in advanced lung cancer patients, which could improve the survival of patients10, 11. The efficacy of platinum-based chemotherapy was individual differences among patients12, 13. Scientists have spared no efforts to search for relevant therapeutic and prognostic biomarkers to improve the accuracy and sensitivity of prognostic and predictive assessment in NSCLC patients. However, there is still a lack of perfect biomarkers and clinical practice.

It is well known that platinum acts through the formation of bulky intrastrand and interstrand DNA adducts that inhibit DNA synthesis and transcription14. Moreover, studies have suggested that the resistant mechanisms of platinum may be via the inactivation of platinum compounds through the glutathione metabolic pathway and via the increase of the DNA repair capacity and of the tolerance to DNA damage14,15,16. Glutathione S-transferases (GSTs) are a series of phase II metabolic enzymes, which are involved in the platinum detoxification17. GSTM1, GSTP1 and GSTT1 are the most important GSTs enzymes18. Evidences showed that the variants of GSTP1 (rs1695, Ile105Val), GSTM1 (null/present) and GSTT1 (null/present) may be involved in the platinum-based treatment, but the results were not consistent14,15,16, 19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39.

Meta-analysis rather than a single study can provide more comprehensive and compelling conclusions by systematically summarizing and analyzing previous data6. There were two meta-analyses which reported the inconsistent results for evaluating the associations between GSTP1 and GSTM1 polymorphisms and response to platinum-based chemotherapy in lung cancer17, 40. These two meta-analyses have not enrolled update studies and just analyzed a few studies, and thus may have biased conclusions. Moreover, up to now there have been no meta-analysis concerning in GSTT1 deletion polymorphism and the response to platinum-based chemotherapy in lung cancer. Therefore, after combining all available data and derived more precise and comprehensive assessment, we have updated this system review and meta-analysis to find out the reliable associations of GSTP1 (Ile105Val), GSTM1 (null/present) and GSTT1 (null/present) variants with the efficacy and clinical outcomes of NSCLC patients treated with platinum-based chemotherapy.

Results

Study selection and characteristics of studies

A total of 254 publications were found after excluding the duplication studies from 1185 publications. We excluded 162 irrelevant studies, 25 meta-analyses, 3 case reports, and 28 basic studies. Thirty-six studies were included for further review. After excluding no-detail data for meta-analysis, 31 papers including 5712 patients were enrolled in the final analysis. Among them, 29 studies were involved in GSTP1 Ile105Val; 16 studies were involved in GSTM1 (null/present); and 11 studies were involved in GSTT1 (null/present). The enrolled details and CONSORT diagram are shown in Fig. 1. The characteristics of first author name, publishing year, country, ethnicity, age, smoking percentage, clinical stage, method of detect polymorphisms, quality score (QS), and the number of patients were shown in Table 1. Twenty-two of the included studies were conducted on Asian patients and 9 were on Caucasian patients. The information of objective response rate (ORR), overall survival (OS) and hazard ratios (HRs), median survival time (MST), the median time to progression (TTP) and the median progression-free survival (PFS) in each study are shown in Tables 2 and 3.

Figure 1
figure 1

Procedure of article selection.

Full size image
Table 1 Characteristics of studies included in meta-analysis.
Full size table
Table 2 Association between the GSTP1 IIe105Val polymorphism and objective response rate, median survival time, median time to progression and median progression-free survival of platinum-based chemotherapy in NSCLC patients.
Full size table
Table 3 Association between the GSTM1 and GSTT1 polymorphisms and objective response rate, median survival time, median time to progression and median progression-free survival of platinum-based chemotherapy in NSCLC patients.
Full size table

Objective response rate of GSTP1 Ile105Val genetic polymorphism

There were 21 publications including a total of 3200 patients enrolled for comparing the ORR in GSTP1 Ile105Val different genotypic patients. The results showed that there was a statistically significant association between the GSTP1 Ile105Val polymorphism and the ORR under dominant model (IIe/Val + Val/Val vs. IIe/IIe: odds ratio (OR) = 1.437, 95% confidence intervals (CIs), 1.019–2.027, P = 0.039). Subgroup analyses by ethnicity suggested that, for the Asian group, the association was significant (OR = 1.592 (1.087–2.332), P = 0.017); for the Caucasian group, the association was not significant (OR = 0.767 (0.479–1.228), P = 0.269) (Table 4 and Fig. 2). Moreover, we also carried out the subgroup analyses based on the evaluation criterion, genotyping method, and quality score. The results were shown in Figure S1 and Table S1. It implied that the contribution of GSTP1 Ile105Val genetic polymorphism to the ORR of platinum-based chemotherapy has a manner of racial differences. Asian NSCLC patients (but not Caucasian NSCLC patients) bearing the favorable GSTP1 IIe105Val + Val105Val genotypes were more likely to have better response rates to platinum-based chemotherapy compared to those with the unfavorable IIe105IIe genotype.

Table 4 Meta-analysis of the association between GSTP1 IIe105Val polymorphism and platinum-based chemotherapy in objective response rate, overall survival and median progression-free survival for NSCLC patients.
Full size table
Figure 2
figure 2

Forest plots of ORR in NSCLC patients treated with platinum-based chemotherapy by the GSTP1 IIe105Val polymorphism (IIe/Val + Val/Val vs. IIe/IIe). Odds ratios (ORs) (and its 95% confidence interval (CI)) of objective response rate (ORR) stratified by ethnicity for GSTP1 IIe105Val IIe/Val + Val/Val vs. IIe/IIe.

Full size image

Overall survival of GSTP1 Ile105Val genetic polymorphism

There were 15 publications including a total of 4276 patients enrolled for comparing the overall survival rates in GSTP1 Ile105Val different genotypic patients. The results showed that there were no statistically significant associations between the GSTP1 Ile105Val polymorphism and OS under any genetic models (IIe/Val vs. IIe/IIe: OR = 0.972 (0.798–1.184), P = 0.78; Val/Val vs. IIe/IIe: OR = 0.772 (0.504–1.182), P = 0.234; Val/Val + IIe/Val vs. IIe/IIe: OR = 0.833 (0.296–2.347), P = 0.729) (Table 4). For the Asian group or the Caucasian group, there were no significant associations between the GSTP1 Ile105Val polymorphism and OS under any genetic models (Table 4).

Median progression-free survival of GSTP1 genetic polymorphisms

There were 2 publications including a total of 430 patients enrolled for comparing the median progression-free survival rates in GSTP1 Ile105Val different genotypic patients. The results showed that there were no statistically significant associations between the GSTP1 Ile105Val polymorphism and PFS under any genetic models (IIe/Val vs. IIe/IIe: OR = 0.728 (0.207–2.566), P = 0.622; Val/Val vs. IIe/IIe: OR = 0.511 (0.049–5.317), P = 0.574) (Table 4).

Objective response rate of GSTM1 and GSTT1 null or present genetic polymorphism

There were 10 publications including a total of 1638 patients enrolled for comparing the objective response rates in the GSTM1 null or present genotypic patients. The results showed that there were statistically significant associations between the GSTM1 null or present polymorphism and ORR (null vs. present: OR = 1.478 (1.200–1.820), P < 0.001). Subgroup analyses by ethnicity suggested that, for the Asian group, the association was significant (OR = 1.493 (1.192–1.870), P < 0.001); for the Caucasian group, the association was not significant (OR = 1.393 (0.806–2.408), P = 0.236) (Table 5 and Fig. 3). The Asian NSCLC patients bearing the favorable GSTM1 null genotype were more likely to have better response rates to platinum-based chemotherapy compared to those with the unfavorable GSTM1 present genotype.

Table 5 Meta-analysis of the association between GSTM1 and GSTT1 polymorphisms and platinum-based chemotherapy in objective response rate, overall survival for NSCLC patients.
Full size table
Figure 3
figure 3

Forest plots of ORR in NSCLC patients treated with platinum-based chemotherapy by the GSTM1 null or present polymorphism (null vs. present). Odds ratios (ORs) (and its 95% confidence interval (CI)) of objective response rate (ORR) stratified by ethnicity for the GSTM1 null or present polymorphism IIe/Val + Val/Val vs. IIe/IIe.

Full size image

There were 5 publications including a total of 1162 patients enrolled for comparing the objective response rate in the GSTT1 null or present genotypic patients. The results showed that there were no statistically significant associations between the GSTT1 null or present polymorphism and ORR (null vs. present: OR = 1.035 (0.805–1.331), P = 0.80). Subgroup analyses by ethnicity suggested that, for the Asian group, the association was also not significant (OR = 1.033 (0.802–1.332), P = 0.91); for the Caucasian group, the association was not significant (OR = 1.106 (0.193–6.342), P = 0.79) (Table 5).

Overall survival and median progression-free survival of GSTM1 and GSTT1 null or present genetic polymorphism

There were 12 publications including a total of 2638 patients enrolled for comparing the overall survival rates in the GSTM1 null or present genotypic patients. The results showed that there were no statistically significant associations between the GSTM1 null or present and OS (null vs. present: OR = 1.054 (0.870–1.277), P = 0.593). For the Asian group or Caucasian group, there were no significant associations between the GSTM1 null or present genotype and OS (Table 5).

There were 2 publications including a total of 430 patients enrolled for comparing the median progression-free survival rates in the GSTM1 null or present genotypic patients. The results showed that there were no statistically significant associations between the GSTM1 null or present and PFS (null vs. present: OR = 0.912 (0.680–1.224), P = 0.539) (Table 5).

There were 10 publications including a total of 2275 patients enrolled for comparing the overall survival rates in the GSTT1 null or present genotypic patients The results showed that there were no statistically significant associations between the GSTT1 null or present and OS (null vs. present: OR = 1.076 (0.899–1.288), P = 0.424). For the Asian group, there was no significant association (null vs. present: OR = 0.867 (0.683–1.101), P = 0.242). However, for the Caucasian group, there was significant association between the GSTT1 null or present genotype and OS (null vs. present: OR = 1.423 (1.084–1.869), P = 0.011) (Table 5 and Fig. 4). The results suggested that the Caucasian lung cancer patients bearing the GSTT1 null genotype might be more closely associated with shorter survival time and higher risks of death than the GSTT1 present patients.

Figure 4
figure 4

Forest plots of OS in NSCLC patients treated with platinum-based chemotherapy by the GSTM1 null or present polymorphism (null vs. present). Hazard ratios (HRs) (and its 95% confidence interval (CI)) of overall survival (OS) stratified by ethnicity for the GSTM1 null or present polymorphism null vs. present.

Full size image

Publication bias and sensitivity analysis

Publication bias was examined by Egger’s test and Begg’s test. As shown in Figure S1, Begg’s funnel plots and Egger’s funnel plots under the GSTP1 IIe105Val dominant model (IIe/Val + Val/Val vs. IIe/IIe) appear approximately symmetrical and show no publication bias (P = 0.833, P = 0.467, respectively). As for the GSTM1 null or present genetic polymorphism, the shapes of the Begg’s funnel plots and Egger’s funnel plots seem approximately symmetrical and show no publication bias (P = 0.592, P = 0.399, respectively). The shapes of the Begg’s funnel plots and Egger’s funnel plots of GSTT1 null or present genetic polymorphisms seem not symmetrical and show publication bias (P = 0.007, P = 0.002, respectively, Figure S3). After being divided into two groups according to ethnicity, the shapes of the Begg’s funnel plots and Egger’s funnel plots of GSTT1 null or present genetic polymorphism in the Asian population have publication bias (P = 0.027, P = 0.002, respectively, Figure S3). However, there is no publication bias of GSTT1 null or present genetic polymorphism in the Caucasian population (P = 0.221, P = 0.385, respectively, Figure S3). Sensitivity analysis results show that changing the effect models had no significant effects on the pooled OR, HR and the final strength of the association between GSTP1 IIe105Val, GSTM1 and GSTT1 null or present genetic polymorphisms and the clinical outcome of platinum-based chemotherapy to NSCLC patients. Moreover, Fig. 5 show the results of sensitivity analysis regarding ORR of GSTP1 IIe105Val dominant model (IIe/Val + Val/Val vs. IIe/IIe) in overall population or Asian population. We found that excluded studies did not influence the overall effective size in the Asian population.

Figure 5
figure 5

The sensitivity analysis of pooling ORs of ORR in NSCLC patients treated with platinum-based chemotherapy by the GSTP1 IIe105Val polymorphism (IIe/Val + Val/Val vs. IIe/IIe). (a) in overall patients, (b) in Asian patients.

Full size image

Discussion

In this meta-analysis, results show that GSTP1 IIe105Val IIe/Val and Val/Val genotypic Asian NSCLC patients were more likely to have better response rates compared to GSTP1 IIe105Val IIe/IIe patients. The Asian NSCLC patients bearing the favorable GSTM1 null genotype were more likely to have better response rates to platinum-based chemotherapy compared to those patients with the unfavorable GSTM1 present genotype. Caucasian NSCLC patients bearing GSTT1 null genotype might be more closely associated with shorter survival time and higher risks of death than the GSTT1 present patients. Herein, we suggested that the GSTP1 IIe105Val, GSTM1 and GSTT1 null or present genetic polymorphisms might be predictive factors for the efficacy of platinum-based chemotherapy to NSCLC patients.

The platinum-based chemotherapy is the standard first-line and effective therapies for NSCLC patients, especially for advanced cancer. However, the efficacy of the platinum-based chemotherapy varies wildly among patients. Previous studies provide the evidences that genetic variants of genes involved in the detoxification and DNA repair pathways including GSTP1, GSTM1, GSTT1, ERCC1, XPD, XPG, XRCC1 may influence the anti-cancer efficacy of platinum-based chemotherapy14, 29, 38, 41,42,43,44,45. However, their results were inconsistent and need meta-analysis and further confirmation. GSTP1, GSTM1, and GSTT1 are three genes of human glutathione S-transferases (GSTs) super family members, which have crucial roles in metabolizing most cytotoxic cancer chemotherapeutic agents such as the platinum detoxification46, 47. One nonsynonymous polymorphism occurring in GSTP1 (IIe105Val) in exon 5 and allelic deletions in the GSTM1 and GSTT1 variants are associated with the lower substrate specific catalytic activity and the reduced enzyme activity, thus lowering the intracellular concentration of chemotherapeutic agents16, 48. Therefore, the patients who suffer the favorable GSTP1 IIe105Val and Val105Val genotypes may display a reduced ability to detoxify drug metabolites, thus promoting better response rates to platinum-based chemotherapy. Allelic deletions in the GSTM1 and GSTT1 genotypes are associated with reduced enzyme activity thus they could be predictive factors of the efficacy of platinum-based chemotherapy.

Published data have indicated that the GSTP1 IIe105Val variant might be associated with the efficacy of platinum-based chemotherapy in lung cancer patient16, 19,20,21, 23, 24, 27, 29. However, there were also some negative results about GSTP1 IIe105Val variant15, 22, 28, 31, 49. In our meta-analysis, we found the significant association between GSTP1 IIe105Val dominant model and ORR of Asian NSCLC patients treated with platinum-based chemotherapy (Table 4). Because of the heterogeneity, random model were used to pool the OR of ORR of overall patients and Asian patients. Moreover, we carried out the sensitivity analysis and results showed that changing the effect models had no significant effects on the pooled OR of ORR and the final strength of the association between GSTP1 IIe105Val and ORR of Asian NSCLC patients treated with platinum-based chemotherapy. Moreover, excluded studies did not influence the overall effective size in Asian population (Fig. 5). Lung cancer is a kind of complicated illness and different ethnicities have different genetic backgrounds, which may affect the anti-cancer therapeutic outcome of platinum-based chemotherapy. Herein, we evaluated the relationship of GSTP1, GSTM1, and GSTT1 variants and the efficacy of platinum-based chemotherapy stratified by different ethnicities. There were two ethnicities enrolled in our studies: Asian and Caucasian. Our analysis found no significant association on GSTP1 IIe105Val variant and clinical outcome of platinum-based chemotherapy in Caucasian patients (Table 4). It implies that the ethnic difference also influence the contribution of the GSTP1 IIe105Val variant to the variation of clinical outcomes of platinum-based chemotherapy. Therefore, the ethnic factor should be considered and weighed with caution when we drawn the conclusions from our meta-analysis. Ethnic individual platinum-based chemotherapy treatment for NSCLC patients should be conducted in the future.

GSTM1 and GSTT1 are located on chromosome 1p13.3 and 22q11.2. Homozygous of GSTM1 and GSTT1 null genotypes lead to an absence of enzymatic activity50,51,52. The relationships of GSTM1 and GSTT1 genotypes and the survival rates in lung cancer are revealed to be quite conflicting also. Several studies have not found significant associations21, 23, 49, 53, 54, while others have found significant associations20, 22, 25, 34. Our meta-analysis suggests that the lung cancer patients bearing the favorable GSTM1 null genotype were more likely to have better response rates to platinum-based chemotherapy compared to those with the unfavorable GSTM1 present genotype in Asian patients, but not in Caucasian patients (Table 5 and Fig. 3). In Caucasian group, there was significant association between the GSTT1 null or present genotype and overall survival (null vs. present: OR = 1.423 (1.084–1.869), P = 0.011) (Table 5 and Fig. 4).

Important things that cannot be ignored in meta-analysis are heterogeneity and publication bias. We carried out the Q test and I2 statistics to test the significance of heterogeneity. There were obvious heterogeneities in pooled ORR, OS and PFS of GSTP1 IIe105Val variant patients (Table 4). Therefore the random model was used. In order to find out the source of heterogeneity, we conducted subgroup analysis by ethnicity. However, after the subgroup analysis by ethnicity, there were still heterogeneities in Asian group even when clinical outcome were pooled (Table 4). In contrast, in the Caucasian group, there were no heterogeneities, indicating the heterogeneity could be partly accounted for by the genetic distribution in different ethnicities. In addition, the inconsistency of these studies about Asian patients may be due to the source of the patients, disease condition, publication qualities or other clinical issues. Further large sample multi-center studies are needed. In order to draw more cautious conclusion on GSTP1 IIe105Val, we also carried out the sensitivity analysis. Results showed that changing the effect models had no significant effects and excluding some studies did not influence the overall effective size in pooled OR value of ORR in the Asian population (Fig. 5). We used Egger’s test and Begg’s test to analyze publication bias. There was no publication bias in GSTP1 IIe105Val and GSTM1 null or present genetic polymorphism on clinical outcome of platinum-based chemotherapy (Figure S2). We have seen the publication bias in GSTT1 null or present variant on clinical outcome of platinum-based chemotherapy (Egger’s test P = 0.002, Begg’s test P = 0.007, Figure S3). After the subgroup analysis by race, the publication bias has disappeared in the Caucasian group (Egger’s test P = 0.221, Begg’s test P = 0.385) but not in the Asian population (Egger’s test P = 0.027, Begg’s test P = 0.002, Figure S3). Herein, we could drawn the conclusion that the Caucasian lung cancer patients bearing GSTT1 null genotype might be more closely associated with shorter survival time and higher risks of death than the GSTT1 present patients and there was no publication bias in this meta-analysis about GSTT1 null genotype and survival time in the Caucasian population.

Our meta-analysis pooled ORR, OS and PFS of NSCLC patients treated with platinum-based chemotherapy harboring different GSTM1 and GSTT1 null genotypes.

After our precise and comprehensive assessment of the update system review and meta-analysis, pooled ORR, OS and PFS enrolled a total of 5712 NSCLC patients treated with platinum-based chemotherapy in our comprehensive and systematic evaluation of efficacy. We found that GSTP1 IIe105Val, GSTM1 and GSTT1 null genetic polymorphisms might be predictive factors for the efficacy of platinum-based chemotherapy to NSCLC patients.

Previously, there were two meta-analyses that revealed the GSTP1 IIe105Val, GSTM1 null genetic polymorphisms and the efficacy of platinum-based chemotherapy in NSCLC patients and no meta-analysis about GSTT1 null genetic polymorphisms and the efficacy of platinum-based chemotherapy in NSCLC patients17, 40. These results from the two meta-analyses seem conflicting rather than conclusive for each other. The different studies enrolled in their analysis may possibly bias the conclusions. In our meta-analysis, we systematically enrolled all available up-to-date studies related with GSTP1 IIe105Val, GSTM1 and GSTT1 null genetic polymorphisms and the efficacy of platinum-based chemotherapy to NSCLC patients.

Our updated meta-analysis enrolled 29 publications including 5414 NSCLC patients harboring GSTP1 IIe105Val variant, 16 publications including 3008 NSCLC patients harboring GSTM1 null or present variant, 11 publications including 2356 NSCLC patients harboring GSTT1 null or present variant, which are several times more than the previous two meta-analyses. Therefore, our meta-analysis is more precise and reliable in predicting the role of GSTP1, GSTM1 and GSTT1 polymorphisms on the clinical outcome of platinum-based chemotherapy in NSCLC patients.

Despite our efforts to conduct a comprehensive and accurate meta-analysis, it still has several limitations, which should be taken into account in interpreting the existing results. First of all, the sample sizes and numbers of enrolled studies in our meta-analysis are still limited, especially in the subgroup analysis and single studies (range from 59 to 420). Only 9 publications of patients were of Caucasian populations, which also limited the generalizability to other ethnic populations. Some indicators such as TTP or PFS may have been undervalued in analysis because of the limited numbers of enrolled studies. The second limitation is the significant heterogeneity between studies in pooled analysis for GSTP1, although it is unlikely to influence the final conclusion after other analyses are carried out, such as stratified analyses by race, sensitivity analysis, and the changes of analysis models. Thirdly, the variation in the patients’ characteristics in each study, such as age, gender percentage, ethnicity, TNM staging, smoking history, specific anti-cancer drugs, chemotherapy regimens, test methods, may also influence the heterogeneity of studies and the final conclusions. Moreover, the quality of publications is still in need of further accurate and precise improvement.

Conclusions

In conclusion, our meta-analysis indicates that the GSTP1 IIe105Val, GSTM1 and GSTT1 null or present genetic polymorphisms might be predictive factors for the efficacy of platinum-based chemotherapy to NSCLC patients. GSTP1 IIe105Val IIe/Val and Val/Val genotypic NSCLC patients were more likely to have better response rates compared to those IIe/IIe genotypic Asian patients. The lung cancer patients bearing the favorable GSTM1 null genotype were more likely to have better response rates to platinum-based chemotherapy compared to those with the unfavorable GSTM1 present genotype, especially in Asian patients. Caucasian lung cancer patients bearing GSTT1 null genotype might be more closely associated with shorter survival time and higher risks of death than the GSTT1 present patients. In the future, well-designed pharmacogenetic studies with multi-center, multi-ethnic and large sample sizes are needed to draw a more accurate and robust conclusion.

Materials and Methods

Study review and selection

We reviewed the databases including PubMed, EMBASE, Web of Science, Wanfang and CNKI to 14 Oct. 2016. The searching strategy was “GSTP1 or GSTP1 glutathione S-transferase pi 1”, “GSTM1 or glutathione S-transferase mu 1”, “GSTT1 or glutathione S-transferase theta 1”, “lung cancer or carcinoma or tumor”, “SNPs or genetic polymorphisms or variations”, “pharmacogenomics”, “platinum or cisplatin or carboplatin or nedaplatin, lbaplatin, oxaliplatin” and “chemotherapy” Dr. Qiang Qu and Dr. Huan Ye reviewed all relevant articles to identify potential eligible studies.

Inclusion and exclusion criteria

The inclusion criteria are: (1) NSCLC patients; (2) At least having one of GSTP1 IIe105Val, GSTM1 and GSTT1 null or present genetic polymorphisms data; (3) At least having one clinical indicator (ORR, OS, PFS, TTP, OR and HR with corresponding to 95% CIs); (4) Treatments having platinum-based chemotherapy. A study was excluded if any of the following exclusion criteria applies: (1) having no relevance to cancer and clinical patients; (2) having no variants information or having no clinical indicators; (3) Involving just in animals or cells; or being a review, or being an abstract with no data. Different opinions on study selections were solved in a discussion by all authors.

Data collection and quality assessment

Two investigators (Dr. Qiang Qu and Dr. Huan Ye) independently extracted data from eligible studies. Different opinions on study selections were solved by all author’s discussion. The data were extracted as follows: authors’ names, sex, smoking status, ethnicities (Asian and Caucasian), clinical stage, evaluation criterion, genotyping methods, outcomes (ORR, OS, PFS, MST, TTP, OR and 95% CI), and the number of responders and non-responders in different genotypes. The QS for each study was also evaluated separately by two investigators (Dr. Jian Qu and Dr. Meiqin Shao) using previous methods6. According to the QS, every study has its score range from 0 to 24 reflecting cancer clinical stage, evaluation criteria, platinum combinations, genotyping methods, OS, PFS, MST, and sample size. The literature with QS ≤ 14 was considered low quality and the literature with QS > 14 was considered high quality.

Response evaluation criteria in solid tumors (RECISTC) or World Health Organization (WHO) were used to evaluate therapeutic efficacy including complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD). We evaluated the ORs and 95% CIs for the objective response rate (ORR) and no response after platinum-based chemotherapy (CR + PR vs. PD + SD). PRISMA checklist was used for our meta-analysis guideline55.

Statistical analysis

We used STATA version 12 (Stata Corp, College Station, TX, USA) to carry out the meta-analysis. Heterogeneity was assessed by the Cochrane’s Q-statistic test and I 2 test. Random effect model was used in the analysis if P < 0.05 and I 2 > 50%, otherwise, a fixed effect model was chosen56. The significance of the pooled ORs was estimated using the Z-test. Publication bias was analyzed by Egger’s test and Begg’s test. Tests were two-sided and statistical significance was accepted at P < 0.05.