We searched the above 11 databases and obtained 677 literatures. After browsing the titles and abstracts, we downloaded 21 full texts. Finally, four studies [25-28] were included. Three were case-control studies and the remaining one was a cohort study, all of which were published in English. The flow chart is shown in Fig. 1.
Characteristics of the included studies
A total of 12912 participants were included in four studies [25-28]. Two of the four case-control studies [25, 27] observed male with abnormal sperm, such as dyspermia, azoospermia or oligospermia, and the other two involved female patients with primary infertility or minimal/mild endometriosis. The only cohort study  observed nulliparous female without a history of infertility. The exposures of two studies [25, 27] were coffee, and the other two were caffeine containing beverage (CCB), including caffeinated soft drinks, coffee, tea, cocoa or cola et al. For two of the studies [26, 28], exposure intensity were analyzed according to the daily caffeine intake (mg/d), and for the other two studies [25, 27] caffeine intake was calculated according to the corresponding conversion relationship of caffeine consumption according to each kind of CCB.
In three case-control studies [25-27], odds ratio (OR) were used to indicate the risk of infertility caused by coffee or caffeine intake. The cohort study  used hazard ratio (HR) to indicate the risk of caffeine intake. HR is a risk ratio that takes into account the time factors, that is, when the number of events is included in the dynamic cohort, it also takes into account the loss of follow-up or other exposure factors to cause the event to occur. In this review, OR and HR cannot be directly converted. The basic characteristics of the four studies included are shown in Table 1.
Risk of bias of included studies
The full score of NOS is 9 points. Only one case-control study  used blinded investigation, and the other three studies [26-28] used unblinded investigations or case records as a way to determine exposures or outcomes. Only one case-control study  used community controls, and the rest were hospital controls. One case-control study  did not describe non-response rate and did not adjust for confounding factors. Three case-control studies gained 6, 6 and 7 points respectively. The cohort study achieved 9 points. The methodological quality of the case-control studies included in this review is general, but the cohort study is good. The methodological quality of included studies is shown in Fig. 2.
Estimate relationship between caffeine intake and infertility
For the effect of caffeine on male factor infertility
In Buiatti's study , males with azoospermia or oligospermia (i.e., sperms/ml < 20, 000, 000) were used as the case group, and normal routine examinations and sperm counts > 20, 000, 000/ml were used as the control group in the same period. The study showed that drinking coffee seems to have no correlation with sperm abnormalities (RR 0.91, 95%CI 0.46-1.82). Although it was mentioned that the coffee intake was divided into different doses for further analysis, relevant data was not provided and the corresponding relative risk were not calculated based on them.
Prazzini's study  had three groups, in which infertile men with dyspermia were used as the case group, and normospermic infertile men and fertile men with unknown semen quality as the control groups. Our study included only the case group and the fertile control group. With the number of cups of coffee per day, the risk of dyspermia increased. Compared to less than 100mg/d caffeine intake those who had 200-300mg/d caffeine intake may have an OR of 1.7 (95%CI 0.8-3.7), and those who had more than 400mg/d caffeine intake may have even higher risk of infertility (OR 3.4, 95%CI 2.4-12.6).
For the effect of caffeine on female factors infertility
In Grodstein's study , primary infertility was used as the case group and those with a history of delivery in the past two years as the control group. It reported the number of infertile patients with different doses of caffeine intake in different etiology. After adjusting for the relevant confounding factors, the risk of infertility caused by tubal disease (RR 1.5, 95%CI 1.1-2.0) increased significantly when caffeine consumption more than 7g/m (233mg/d), and the risk of cervical factors (OR 1.3, 95%CI 0.7-2.4, for 5.1-7 g/m and OR 1.4, 95%CI 0.9-2.3, for >7 g/m) or endometriosis (OR 1.9, 95%CI 1.2-2.9, for 5.1-7 g/m and OR 1.6, 95%CI 1.1-2.4, for >7 g/m) increased significantly when caffeine intake greater than 5g/m (167mg/d).We added the number of case together regardless the cause of infertility and re-analyzed the odds ratio. The results showed with 101-167mg/d caffeine intake may not increase the risk of infertility compared to less 100mg/d caffeine intake (OR 1.00, 95%CI 0.79-1.25), however, higher dose of caffeine intake may increase the risk (OR 1.16, 95%CI 0.94-1.42, for 168-233mg/d; OR 1.36, 95%CI 1.16-1.60, for >233mg/d).
The only cohort study included  was a retrospective cohort study. It was divided into exposure and non-exposure group according to whether the participants drink coffee or tea. For different doses of total caffeine consumption (from coffee and tea), the risk of infertility for consumers was similar to that of never consumers. Only 100 mg/d of caffeine consumed did not affect the risk of primary infertility in the consumers (HR 1.00, 95% CI 0.98-1.02).
Details of the results from each single study was also shown in Table 1.
Considering the clinical heterogeneity among each study, we chose to present the single study’s results with bubble plot first. The X-axis represents the effect value (OR), the Y-axis represents the methodological quality score (according to NOS), and the bubble size reflects the size of the sample size. Fig. 3(A) showed the risk of caffeine intake (more than 100mg/d) for infertility from 3 studies, all of them found no difference of incidence rate of infertility between low dose and no caffeine intake (OR varied from 0.77 to 1). Fig. 3(B) showed the risk of caffeine intake (more than 200mg/d) for infertility also from 3 studies. Both case-control studies found that the incidence of infertility was higher in the medium dose than no caffeine intake, but the cohort study did not show a different incidence of infertility between the two groups (OR varied from 0.67 to 1.64). Fig. 3(C) showed the risk of caffeine intake (more than 400mg/d) for infertility from 2 studies. The case-control study (OR 4.99) found that the incidence of infertility was higher in the high dose than no caffeine intake, but the cohort study did not show a different incidence of infertility between the two groups (OR 0.74).
Meta-analysis was also conducted (Fig. 4). Subgroup was classified according to the type of the study (case-control and cohort study). The results also showed caffeine intake may not increase the risk of infertility (lose dose: OR 0.95, 95%CI 0.78-1.16, M-H Fixed, I2=0%, P=0.64, 4627 participants, 3 studies; medium dose: OR 1.14, 95%CI 0.69-1.86, M-H Random, I2=72%, P=0.0006, 5353 participants, 3 studies; high dose: OR 1.86, 95%CI 0.28-12.22, M-H Random,I2=94%, P<0.0001, 2435 participants, 2 studies). However, due to the obvious statistical heterogeneity when pooling the data for high dose of caffeine intake assessment (I2=94%,), results of this meta-analysis was only used for GRADE evaluation.
Quality of evidence
We used the GRADE method to evaluate the certainty of the evidence. For the three different doses of formed evidence bodies, no matter from which aspects, including large effect, dose-response gradient and plausible confounding, cannot be upgraded. The overall quality of the three evidence is all low (Table2).