Table 3 Specific biases, confounders and limitations present and/or declared in the available studies

Qin et al., 2016 [19]

Meta-analysis of 50 cohort studies

Little evidence of publication bias.

22 % of the included studies did not adjust and/or match any factors (i.e. maternal age, education, parity, race, occupation, smoking during pregnancy, socioeconomic status, etc.) when estimating the effect of ART singletons on obstetric outcomes.

- Patients who achieved a pregnancy with OI and IUI have been considered in the NC category.

- Substantial heterogeneity among studies for association between ART singleton pregnancies and obstetric risks.

- A number of the outcomes, especially for pregnancy-related complications, relied on between 2 and 7 of the 50 total studies.

Cromi et al., 2016 [21]

Case–control study

The study reports the experience of a tertiary referral center; therefore, this factor may have inflated the observed rates of peripartum hysterectomy.

The control populations have not been separated into two groups (normal fertile couples and infertile couples who conceive without treatment) to determine the degree to which observed associations are specifically related to the ART procedures vs. infertility per se.

- Detailed information on the infertility treatments was not available.

- Small study number.

Pandey et al., 2012 [18]

Review of 20 matched cohort studies and 10 unmatched cohort studies

Ascertainment bias with the findings of increased complications with IVF/ICSI; i.e. women may be more anxious following fertility treatment and therefore more likely to report problems.

The quality of most of the studies was high and they have adjusted for most important confounders of age and parity.

- Some authors excluded pregnancies resulting from ovulation induction whereas others were not able to identify them from all non-IVF/ICSI conceptions.

- The review cannot determine whether the increased risk is due to the inherent infertility itself or the process of ovarian stimulation and/or embryo culture.

Qin et al., 2015 [22]

Meta-analysis of 39 cohort studies

- No evidence of publication bias among studies of ART and risk of adverse outcomes.

- All the included original studies used a cohort study design, which minimizes recall and selection biases.

- Some included studies have considered pregnancies arising after OI and IUI to be in the spontaneously generated category.

- A number of the outcomes, especially for pregnancy-related complications, relied on between 1 and 8 of the 39 total studies.

- The study population consisted of monochorionic and dichorionic twins, and monochorionic twins are known to be at high risk than dichorionic twins.

- Residual confounding is a concern, although restricting analysis to studies that have matched or adjusted confounding factors did not materially alter the combined risk estimate.

Qin et al., 2016 [23]

Meta-analysis of 15 cohort studies

No evidence of publication bias.

26.7 % of the studies did not adjust and/or match any factors

when estimating the effect of ART on obstetric outcomes in

dichorionic twin pregnancies.

- More than half of the included studies had a small sample size.

- Most of included studies belonged to retrospective cohort design.

- There was substantial heterogeneity among studies for association between ART and obstetric risks in dichorionic

twin pregnancies.

- A number of the outcomes, especially for maternal complications, relied on between 2 and 7 of the 15 total studies.

Pinborg et al., 2013 [9]

Meta-analysis of 3 studies (for the considered outcome: PTB in SET vs. DET)

Subfertility per se is a bias and it cannot be prevented directly.

Vanishing twin pregnancies involve about 10 % of pregnancies with a DET-only strategy, leading to growth disturbances and to non-optimal perinatal outcomes among ART singletons.

Marton et al., 2016 [35]

Longitudinal, retrospective cohort study

The focus was not on procedure specifics, even if each artificial procedure has profound effect on the splitting of the zygote, which represents a bias in the comparison of spontaneous and IVF–ICSI VT-pregnancies.

The relatively low incidence of VTS denotes the low power of the statistical analyses.

Nakashima et al., 2013 [43]

Retrospective study based on national registry

The effect of the subfertility has not been prevented.

The dataset had information on few confounders.

Detailed information on the infertility treatments was not available.

Sunkara et al., 2015 [49]

Prospective cohort study

The effect of the different gonadotropin dosages has not been excluded.

The dataset had no information on important confounders such as smoking, BMI and the medical history of women during pregnancy.

- The dataset did not allow specific identification of women with PCOS and its anonymized nature did not make it permissible to analyze one cycle per woman.

- Individual women would have contributed to more than one cycle and outcome in the data set which means that the true sample size is unknown.

Kalra et al., 2012 [61]

Retrospective cohort study

To attempt to control selection bias, subanalyses were performed.

- Adjusted analyses were performed and included variables that were considered clinically important, because they are associated with adverse outcome.

- Data not adjusted for gonadotropin dose.

Mäkinen et al., 2013 [65]

Retrospective cross-sectional cohort study

The effect of the subfertility has not been prevented.

- The study was not adjusted for smoking and for gonadotropin doses.

- Because of insufficient perinatal data, the authors were no able to control additional factors known to affect pregnancy outcome such as PIH, PE and GDM.

Lack of control of the duration of infertility.

Maheshwari et al., 2012 [81]

Systematic review and meta-analysis of 11 cohort studies

Patients who have had fresh cycle may be different from those who had frozen replacement cycles.

Data not adjusted for confounders such as age, smoking, parity, duration of infertility, and pre-existing medical illness due to varied design of the studies

- This review is limited to data from observational studies

- There is inconsistency in definition of outcomes, such as antepartum hemorrhage, congenital anomalies, and perinatal mortality. In addition, not all outcomes have been reported by all studies.

- There is clinical heterogeneity in terms of the population sampled, design of studies, method of freezing, and regimens in replacement cycles.

- There is uncertainty as to whether method of thawing and protocol used (natural or hormonally mediated cycle) for replacement has any bearing on different obstetric and perinatal outcomes.

Ishihara et al., 2014 [89]

Retrospective study based on national registry

The different protocols and criteria for the use of frozen ET and blastocyst transfer potentially could bias the data.

The Japanese registry is cycle based with complete anonymity, therefore, it is impossible to know the detailed background of the patients who underwent ART, e.g., gravidity, parity, previous uterine surgery, etc.

- Wide variability of data compiled from almost 600 clinics that are different in size, location, and other characteristics.

- Lack of a national registry of perinatal outcomes (incomplete data on maternal and neonatal outcomes for the final analysis).

Pinborg et al., 2014 [98]

National register-based controlled cohort study with meta-analysis

A bias is very unlikely as data coding was based on national registers

The data were not adjusted for confounding factors, such as maternal BMI and GDM.

The size of the frozen ET/fresh sibling cohort was limited.

Cobo et al., 2014 [107]

Retrospective cohort study

To avoid any selection bias, the study included the entire population of women from the two analyzed cohorts as were originally present in the Clinic.

- The study analyzed all the births for which there was notification, and not the whole series of IVF/ICSI pregnancies achieved in the Institution during the study period.

- The conclusions are based on retrospective data that were partially obtained through medical questionnaires.

- Information on pregnancy losses before 24 weeks is lacking (i.e., ectopic pregnancies, early and late miscarriages, and terminations of pregnancy due to fetal abnormalities).

- The statistical power may be limited to detect a minor increase in the incidence of negative rare outcomes (i.e., major congenital malformations)

Li et al., 2014 [115]

A population-based cohort study

The effect of the subfertility has not been prevented.

The study used each treatment cycle as the unit of analysis where one woman could be included in both fresh and thaw cycles.

- Lack of information available on clinical-specific cryopreservation protocols and processes for slow freezing-thaw and vitrification-warm of blastocysts and the potential impact on outcomes. The lack of consistent cryopreservation protocols and comparison of embryo qualities might over-estimate the successful rate of vitrification and under-estimate the successful rate of slow freezing of blastocysts.

- The data are observational and hence conclusions concerning the biological effects of vitrification and slow freezing cannot be drawn from our study

Buckett et al., 2007 [123]

Observational study

Risk of an ascertainment bias.

The database includes women with PCOS and the effect on birth weight may be a result of the inherent PCOS, rather than as a direct result of the treatment modality.

- Retrospective design.

- Small sample size.

Jing et al., 2016 [136]

Retrospective cohort study

The effect of the subfertility has not been prevented.

To reduce the significant differences in genetic disorders, number of transferred embryos, methods of genetic testing, and vanishing twin between the two groups a logistic regression was applied in the study.

- Retrospective design.

- Small sample size.

- The study focused on obstetrics and neonatal out- comes and only included patients who were > 28 weeks pregnant.

Storgaard et al., 2016 [155]

Systematic review with meta-analysis of 22 cohort studies and 13 annual report of ASRM

The effect of the subfertility has not been tested.

- OD patients are very heterogeneous regarding age, inheritance and infertility history.

- Oocyte donors also constitute a heterogeneous group. This affects pregnancy rates, but it is not known whether it influences obstetric and neonatal outcomes.

- To ensure reliable results only studies of high and medium quality were included in the meta-analyses (of the 21 included cohort studies comparing an OD group to a control group only two were of high quality and 11 were of medium quality).

- Some outcomes were poorly defined, e.g. only three out of 11 studies included a strict definition of gestational diabetes

Salha et al., 1999 [159]

Retrospective cohort study

The effect of the subfertility has not been prevented.

To limit the confounding variables, women who conceived with donated gametes were compared to age- and parity-matched controls from similar demographic backgrounds who conceived with their own gametes.

- Retrospective design.

- Small sample size.

Söderström-Anttila et al., 2016 [162]

Systematic review of observational studies (cohort studies and case-series)

Cohort studies, but not case series, were assessed for methodological quality, in terms of risk of bias.

- Lack of high quality studies.

- Most studies have small sample size, lack of controls and a low response rate.

- Gestational and traditional surrogacy was not always separated in the studies.