First, we are very sorry that we did not include in our review Coleman et al.'s review, which was published in early 2011. We should have searched and reviewed previous literature on this issue more thoroughly.
Second, we agree with them that there is ‘insufficient’ evidence to support the use of pharmacotherapy for smoking cessation in pregnant smokers because no significant efficacy was found in the three placebo-controlled trials, whereas a significant efficacy was found only in the four non placebo-controlled trials. We already described this in the limitations section of the Discussion. We concluded that there ‘may’ be clinical evidence to support the use of pharmacotherapy for smoking cessation among pregnant smokers.
Third, when we reviewed their previous meta-analysis and then compared it with ours, we found that Coleman et al. used a random-effects model in all the analyses, whereas we used both fixed-effects and random-effects models according to the results of the heterogeneity test (i.e. fixed-effects model for data showing less heterogeneity and random-effects model for data showing heterogeneity). Coleman et al. reported the pooled risk ratio (RR) and 95% confidence interval (95% CI) for smoking cessation in later pregnancy after using nicotine replacement therapy (NRT) was 1.63 (0.85–3.14) using a random-effects model. When we performed a ‘fixed-effects’ meta-analysis using data shown in Figure 2 of Coleman et al.'s review (I2 = 45%; in our review, when I2 is <50%, we used a fixed-effects model), NRT was efficacious (pooled RR, 1.47; 95% CI 1.05–2.05; I2 = 44.9%). Hence, the main reason why the findings and conclusions are not consistent between the two meta-analyses, although similar data were used, is that different models were used, respectively. If Coleman et al. applied a fixed-effects model when I2 is <50%, their conclusion would also be that NRT is efficacious, as in our review.
Last, the remaining issue would be that we should present more conservative findings when the pooled effect sizes are different in the meta-analyses by using between fixed-effects and random-effects models in case of less heterogeneity. We, basically, think that using a fixed-effects model is reasonable although the findings are different between the two models if the heterogeneity test shows less heterogeneity (i.e. I2 < 50%). However, we want to hear other experts’ opinions on this issue.