Characteristics of the studies
According to the search strategy, we found a total of 3,714 articles (1,056 from PubMed, 823 from Web of Science, 463 from CNKI, 1150 from Wan Fang Data and 222 from Chongqing VIP). Among these articles, a total of 68 eligible articles[3, 4, 6-8, 20-82] were finally included for the synthetic analyses. All were published between 1981 and 2017. Among the 68 included articles, 33 were published in Chinese, and 35 were published in English. A flow diagram for the article selection is shown in Figure 1. Baseline characteristics of the 68 eligible articles are listed in additional file 1 Table 1. Among the included articles, there were 37 that reported MN frequency, 20 SCE frequency, 16 CA frequency, four OTM, five TM, seven TL, and four articles reporting T DNA%. The concentration of benzene exposure was less than 3.25mg/m3 in seven articles reporting MN frequency, three on CA frequency and two on TM.
Q statistics and I2 were used to test the heterogeneity of the genetic damage indicators, and the results are as follows: MN frequency (Q = 1126.55, I2 = 95.7%, P < 0.1), SCE frequency (Q = 228.21, I2 = 91.2%, P < 0.1), CA frequency(Q = 88.01, I2 = 81.8%, P < 0.1), OTM (Q = 20.09, I2 = 80.1%, P < 0.1), TM (Q = 114.24, I2 = 93.9%, P < 0.1), TL (Q = 254.58, I2 = 95.7%, P < 0.1), and T DNA% (Q = 83.13, I2 = 95.2%, P < 0.1). The results showed a high degree of heterogeneity among the studies, and we therefore used the random-effects model for analysis. The pooled estimates of effect SMD values on genetic damage indicators are shown in Table 2.
The meta-analysis results showed that occupational benzene exposure significantly increased MN frequency, SCE frequency, CA frequency, OTM, TM, TL, and T DNA% compared with the control group (P < 0.05), and the pooled effect value estimates were 1.36, 0.98, 0.76, 1.06, 0.96, 1.78, and 1.42, respectively. The study also evaluated the effects of an exposure concentration less than 3.25mg/m3. For low exposure, the pooled estimate of effect value for MN frequency was 0.46 (95% confidence interval (CI)(0.09-0.82), P < 0.05), for CA frequency was 0.26 (95% CI (−0.16-0.68), P > 0.05), and for TM was 0.59 (95% CI (−0.08-1.27), P > 0.05), indicating that a low concentration of benzene exposure can also cause genetic damage, mainly by affecting MN frequency. The respective forest plots are shown in Figures 2, 3 and 4.
To judge the stability of the analysis method, the fixed effect model and random effect model were used to calculate the combined SMD and 95% CI for each genetic damage index. The results of the fixed-effects model of MN frequency, SCE frequency, CA frequency, OTM, TM, TL, and T DNA% are as follows: 0.97 (95% CI, 0.91-1.02), 0.67 (95% CI, 0.56-0.78), 0.74 (95% CI, 0.63-0.85), 0.90 (95% CI, 0.71-1.08), 0.84(95% CI, 0.69-0.99), 1.50 (95% CI, 1.33-1.66), and 1.15 (95% CI, 0.98-1.31), respectively. The detailed results are shown in Table 3. Comparing the results of the two models, no significant difference was found in the combined effect value of each genetic damage index, which indicated that the results were stable and reliable.
Egger’s linear regression test and Begg’s test are used to demonstrate publication bias; if P < 0.05, we used the trim and fill method for corrections. Egger’s linear regression test and Begg’s tests both showed no publication bias for the genetic damage indicators CA frequency, OTM, TM, and TL (P > 0.05), while the Egger’s linear regression test and Begg’s tests both found publication bias (P < 0.05) for MN frequency and SCE frequency, Egger’s linear regression test showed publication bias for T DNA% (P = 0.039, 95% CI, 1.57-33.02). The results are shown in Table 4. Therefore, we corrected for publication bias using the trim and fill method. The results for MN frequency and SCE frequency showed that the number of missing articles was 0, and there was no publication bias. The T DNA% results showed that 1 article was missing; although there was publication bias, the combined effect values did not change significantly, and the original results were robust. The results are shown in Table 5.
The selected articles are regarded as a continuous study. This means that the research results must be added to the meta-analysis over time, as the publication of new original studies requires a repeated analysis. This approach can highlight the dynamic trend of the research conclusions and thus reflect the impact of each study on the overall conclusions. We performed cumulative meta-analysis of MN frequency, SCE frequency, and CA frequency according to publication time. The results showed that the stability of the pooled estimate of effect and 95% CI increased with publication time, becoming more stable and statistically significant (P < 0.05). The results are shown in Figures 5, 6, and 7.