Results of the search

The following databases were systematically searched by Marian Showell, the Information Specialist at Cochrane Gynaecology and Fertility, on 15 October 2019: Cochrane Gynaecology and Fertility Specialised Register, CENTRAL, MEDLINE, Embase, and CINAHL. In addition, the 2019 European Society of Human Reproduction and Embryology (ESHRE) and American Society for Reproductive Medicine (ASRM) conference abstracts were handsearched by review author SA on 28 October 2019. In addition, a Google search using the terms 'GM‐CSF, culture media, RCT, and live birth' was undertaken on 28 October 2019.

The search returned 452 records, 151 of which were duplicates. This left 301 titles and abstracts for screening, which was undertaken by all co‐authors using the online software Covidence (Covidence). Each record was screened by two review authors at every stage. We considered 28 papers to be eligible for full‐text screening. We excluded 23 full texts for the following reasons: 11 were the wrong study design; seven were trial registry information only, without data; three were duplicates of included studies; one was an animal study; and one was the wrong intervention. We considered five studies to be eligible for inclusion in the review (Rose 2020; Sbracia 2014; Zafardoust 2017; Zavvar 2016; Ziebe 2013), of which three could be used in meta‐analysis (Rose 2020; Sbracia 2014; Ziebe 2013). The two studies that were not included in meta‐analysis were conference abstracts, and the data could not be extracted reliably without further information from the study authors (Zafardoust 2017; Zavvar 2016); unfortunately we were unable to contact the authors of these studies to obtain the needed clarification. The PRISMA figure illustrates the flow of studies through the review (Figure 1).

Figure 1

---

---

PRISMA study flow diagram.

Included studies

Five studies were eligible for inclusion in the review. Two of these were conference abstracts that could not be included in meta‐analysis because data could not be reliably extracted (Zafardoust 2017; Zavvar 2016). The remaining three studies included two fully published and peer‐reviewed papers, Rose 2020; Ziebe 2013, and one conference abstract (Sbracia 2014).

The largest study was undertaken in Europe; the trial was co‐ordinated from the Netherlands, and participants were recruited from 14 fertility clinics in Sweden and Denmark (Ziebe 2013). Ziebe 2013 included a total of 1332 participants, of whom 654 were randomised to the intervention arm and 678 were randomised to the control arm. The study was sponsored, co‐ordinated, and authored by the worldwide market‐leading manufacturer of GM‐CSF‐supplemented culture media. Women in the intervention arm had all of their embryos cultured in GM‐CSF‐supplemented culture medium at a concentration of 2 ng/mL from fertilisation through to embryo transfer. Women randomised to the control arm of the study had all of their embryos cultured in an IVF culture medium that did not contain GM‐CSF from fertilisation through to embryo transfer. Both IVF and ICSI were undertaken, and a maximum of two embryos were transferred on day 3 in a fresh embryo transfer cycle. The inclusion criteria for the study were as follows: women aged 25 to 39 years, women who had a regular menstrual cycle of 21 to 35 days, women treated with a standard gonadotropin‐releasing hormone (GnRH) agonist or antagonist protocol, and women with three or more follicles with a diameter of 14 mm on the day of human chorionic gonadotropin (hCG) administration, including a leading follicle of 17 mm. The exclusion criteria were: previous participation in the study; use of assisted hatching; use of non‐ejaculated sperm; medical conditions or genetic disorders prohibiting IVF/ICSI or interfering with the interpretation of results; use of investigational drugs within 30 days before oocyte retrieval; severe chronic disease of relevance for reproduction; and oocyte donation.

Rose 2020 was a smaller, single‐centre study undertaken in a fertility clinic in Australia. A total of 100 women were randomised, 50 to the intervention arm and 50 to the control arm of the study. There were no dropouts. Rose 2020 was a cross‐over RCT, but the published data were from the first phase of the trial prior to cross‐over. The study was sponsored by the worldwide market‐leading manufacturer of GM‐CSF‐supplemented culture media. The same company also funded two co‐authors of the study for statistical support. The women in this study underwent fresh embryo transfer following IVF or ICSI. The women in the intervention arm had all of their embryos cultured in GM‐CSF‐supplemented culture medium from fertilisation through to embryo transfer on day 5. The concentration of GM‐CSF in the intervention culture medium was 2 ng/mL, and the medium was changed on day 3 following observation, scoring, and washing, to the next phase of sequential fresh culture medium with the same concentration of GM‐CSF. The control culture medium did not contain GM‐CSF, and similarly, day 3 embryos were observed, scored, washed and then transferred to a fresh sequential culture medium. All trial participants had a day 5 embryo transfer, apart from one woman in the control arm and two in the intervention arm who underwent day 3 embryo transfer. Participants underwent single‐embryo transfer, except four women in the control arm and six women in the intervention arm, who underwent double‐embryo transfer.

The inclusion criteria were: patients must have previously had consecutive transfer of two or more embryos without a positive pregnancy outcome OR have had a history of at least one previous pregnancy loss OR a previous history of poor embryo development (< 20% of embryos developing on the time at day 3 or no blastocysts above grade 2 on day 5). Other additional inclusion parameters included a maternal age between 25 and 41 years, the use of a standard GnRH agonist or antagonist protocol, and three or more follicles of > 14 mm as seen by transvaginal ultrasound before the day of hCG administration. Exclusion criteria included: a need for surgical sperm retrieval (except in cases of previous vasectomy), the use of another investigational drug within 30 days of oocyte retrieval, and/or the presence of a severe chronic disease that could impact the IVF cycle or reproductive outcomes.

Sbracia 2014 was another small, single‐centre RCT, undertaken in a fertility clinic in Italy. The study was written as an abstract for an international conference. The authors reported that there was no funding for the study. A total of 100 women were randomised, 50 to the intervention arm and 50 to the control arm of the study. The women in the intervention arm had all of their embryos cultured in GM‐CSF‐supplemented culture media at a concentration of 2 ng/mL from fertilisation through to embryo transfer. Women in the control arm of the study had all of their embryos cultured in a medium not containing GM‐CSF from fertilisation to embryo transfer. The brand name of the control culture medium was not disclosed in the paper. Fresh embryo transfer of up to a maximum of three embryos following ICSI was undertaken in all cycles in both the intervention and control arms of the study. The inclusion criteria were: women with recurrent implantation failure, three or more consecutive failed IVF cycles with a total of at least 8 good embryos replaced in the uterus, and women aged 40 or less. The exclusion criteria were: women aged over 40, chromosomal defects in the couple, metabolic diseases (diabetes, etc.), and other genetic diseases (thalassaemia, cystic fibrosis, etc.).

Both Zafardoust 2017 and Zavvar 2016 were eligible for inclusion, however data could not be reliably extracted for meta‐analysis. Zafardoust 2017 was a conference abstract which outlined that it was a single‐centre RCT undertaken at a fertility clinic in Iran. The study included couples undergoing frozen embryo transfer following an ICSI cycle with their own gametes. Couples were randomised to either have their frozen embryos thawed and cultured in a test medium containing 2 ng/mL of GM‐CSF, or a control medium not containing GM‐CSF. Couples were eligible for inclusion in the study if the female partner was < 40 years old, had at least four good‐quality embryos after thawing (grade A), and had not had more than one previous embryo transfer. Couples were excluded from entering the study if they needed ICSI cycles requiring pre‐implantation genetic diagnosis, if the female partner had an anatomic disorder of the uterus, or one or more hydrosalpinges. The abstract outlines that 90 women were randomised, and 10 were excluded from the final analysis due to various reasons, however the original numbers of women randomised to each group are not disclosed. The outcome of interest reported by the study was clinical pregnancy, which is reported as two percentages, alongside a P value. However, it was not clear which percentage belonged to which arm of the study, therefore it was impossible to extract any meaningful data for meta‐analysis. Review author SA attempted to contact two of the authors of this study by email on three separate occasions for clarification of these issues, but unfortunately no response was forthcoming.

Zavvar 2016 was also a conference abstract, which outlined a single‐centre RCT undertaken in a fertility clinic in Iran. Zavvar 2016 sought to compare the outcomes of women undergoing ICSI who were randomised to receive an embryo culture medium containing 2 ng/mL GM‐CSF or to a culture medium not containing GM‐CSF. The inclusion criterion was women who produced only immature oocytes in spite of stimulation with gonadotropins. No exclusion criteria were described. The day of embryo transfer and length of time embryos were exposed to the intervention or control media were not described. The outcome of interest, the clinical pregnancy rate, was reported as percentages alongside a P value. However, it was not possible to identify which result was associated with which arm of the study, therefore we could not reliably include data from this study. Review author SA attempted to contact the authors of this study on two separate occasions by email, but unfortunately no response was received.

Excluded studies

We excluded 23 studies following full‐text screening. Eleven studies were the wrong study design, and were excluded for the following reasons: Agerholm 2010 was a phase I safety study and did not replace embryos; Fawzy 2019, Shapiro 2003, Sjoblom 1998, Sjoblom 1999, Sjoblom 1999a, and Sjoblom 2001 randomised oocytes opposed to women or couples; Kinoshita 2019 was a retrospective study; Min 2017 and Sfontouris 2013 were observational studies; and Siristatidis 2013 was a systematic review. Scarpellini 2011 was excluded because it did not study the intervention we were interested in, and Siqueira 2016 was excluded because it was an animal study. Six studies were duplicate references of included studies: Rose 2020, Sbracia 2014, Zafardoust 2017, and Zavvar 2016. Four studies are awaiting classification because the nature of the study and whether women or oocytes was randomised was unclear (ISRCTN94726536; NCT01689428; NCT01689454; NCT02651285).

Allocation

We assessed the risk of selection bias for the three studies included in meta‐analysis to be low (Figure 2) (Rose 2020; Sbracia 2014; Ziebe 2013), as random sequence generation was described in detail and considered to be adequate to achieve a truly random sequence. Rose 2020 described how 50 cards with 'control' and 50 with 'BlastGen' written on them were placed in sealed envelopes by a person unrelated to the trial. They were shuffled several times, and the envelopes were then numbered and opened in consecutive order by the embryologist when an eligible participant was scheduled for egg retrieval. Sbracia 2014 described how participants were randomised using a computer‐generated number sequence; however, allocation concealment was not described, therefore we deemed this study to be at unclear risk. Ziebe 2013 described how they used a computer‐generated randomisation list in blocks of four for each individual clinic in order to maintain balance between the treatment groups at each site. Allocation concealment was described in detail and considered to be at low risk on the basis that each study site received a list of study‐specific consecutive patient ID numbers and a corresponding number of identical‐looking randomised bottles of test and control media that were individually labelled with the corresponding study‐specific ID numbers. On site, the lowest number on the list was always allocated to any new patient recruited at the time of informed consent signature. Consequently, the clinician, embryologist, and participant were all blinded to the allocation.

Figure 2

---

---

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Zafardoust 2017 and Zavvar 2016 could not be included in meta‐analysis, but were considered to be at unclear risk of selection bias because no description of randomisation or allocation concealment was provided.

Blinding

We considered Rose 2020 and Ziebe 2013 to be at low risk for performance and detection bias. Rose 2020 described how clinicians, sonographers, statisticians, and participants were completely blinded to the intervention. Ziebe 2013 described how participants and investigators, including clinicians and embryologists, were blinded to treatment allocation. Following email correspondence, we established that the clinicians performing the ultrasound scans were blinded to the treatment allocation at all times.

Sbracia 2014 did not provide any description of blinding of participants, personnel, or outcome assessors, and was therefore assessed as at unclear risk of bias of performance and detection bias. We judged Zafardoust 2017 and Zavvar 2016 to be at unclear risk of performance and detection bias because there was no description of who, if anyone, was blinded.

Incomplete outcome data

We assessed Rose 2020 and Sbracia 2014 as at low risk of attrition bias. Both studies reported no dropouts. We considered Ziebe 2013 to be at high risk of attrition bias because despite all dropouts being accounted for, the reasons given were not included within the predefined exclusion criteria. For example, no oocytes retrieved, no semen sample, no fertilisation, no embryo transfer, and "non‐includable after randomisation" were given as reasons for exclusion after randomisation, however none of these were listed as exclusion criteria. We contacted the authors to obtain accurate intention‐to‐treat (ITT) data for both arms of the study, which they were able to provide.

We considered Zafardoust 2017 to be at high risk of attrition bias because 10 women were not included in the final analysis, with no reasons provided. We considered this to be a high rate of attrition in a small study. Unfortunately we were unable to use data from this study as the data could not be reliably extracted. We considered Zavvar 2016 to be at unclear risk of attrition bias as dropouts were not described. We could not include data from this study because it was unclear how many participants were included in the analysis.

Selective reporting

We rated Rose 2020, Sbracia 2014, and Ziebe 2013 as being at low risk of reporting bias because the study authors confirmed via email correspondence that they had reported all outcomes as per their prospective clinical trials registrations (NCT02305420, NCT01718210, and NCT00565747, respectively).

We rated Zafardoust 2017 as being at high risk of reporting bias. The available abstract did not report data on miscarriage, multiple pregnancy, and beta human chorionic gonadotropin (BHCG) levels, which are secondary outcomes noted on the prospective clinical trials register (Zafardoust 2017). We attempted to contact the study authors to establish if further trial data were available, but received no response. We rated Zavvar 2016 as being at unclear risk of reporting bias. We had no access to a protocol or an online clinical trial registry.

Other potential sources of bias

We have been in extensive contact with the authors of Ziebe 2013 via email to clarify various numbers from their published study. The co‐authors of this study have been very forthcoming in answering all of our queries and have offered clear explanations of how various numbers are reached in their paper. However, we have assessed Ziebe 2013 as at high risk of bias for this domain because the numbers published in the paper differ from those published in this review, that is we discovered through correspondence that some participants were inaccurately described as miscarriages opposed to biochemical pregnancy losses. We also asked for individual participant data in relation to those babies that suffered aneuploidy or birth defects, or both. On reviewing the data, we discovered that some women underwent termination of pregnancy in light of aneuploidy or birth defects, which had not been included in their aneuploidy or birth defect data. We also discovered that one baby had been classified as having a birth defects, when in fact it was reported as having immature lungs secondary to prematurity. In addition, the reporting of multiple pregnancies in the paper is very confusing. We clarified all of these issues through correspondence, which is summarised in Appendix 13. One co‐author of this review has written a letter, which has been published, outlining the concerns regarding the statistical analysis presented in the paper (Farquhar 2015). Examples of concerns include the adjustment of sample size and the increase of concentration of human serum albumin following interim analysis, and the reporting of 'ongoing implantation rate' as number of transferred embryos opposed to per woman.

GM‐CSF‐supplemented culture media versus culture media not supplemented with GM‐CSF for women undergoing assisted reproduction

A total of five studies undertook this comparison. Three of these studies (1532 participants) reported data that could be included in meta‐analysis (Rose 2020; Sbracia 2014; Ziebe 2013).

Primary outcomes

1.1 Live birth

Two studies (N = 1432) provided live‐birth data (Rose 2020; Ziebe 2013). We obtained ITT live‐birth data following correspondence with the authors of Ziebe 2013 (see Appendix 13). There were 179 live births reported amongst the 704 women randomised to the GM‐CSF arm, and 162 live births amongst the 728 women randomised to the control arm. No studies reported ongoing pregnancy as a proxy to live birth.

We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the live‐birth rate when compared to using conventional culture media not supplemented with GM‐CSF (odds ratio (OR) 1.19, 95% confidence interval (CI) 0.93 to 1.52, 2 RCTs, N = 1432, I² = 69%, low‐quality evidence) (Analysis 1.1; Figure 3).

Figure 3

---

---

Forest plot of comparison: 1 GM‐CSF‐supplemented culture medium versus culture medium not supplemented with GM‐CSF, outcome: 1.1 Live birth.

The evidence suggests that if the rate of live birth associated with conventional culture media not supplemented with GM‐CSF was 22%, the rate with the use of GM‐CSF‐supplemented culture media would be between 21% and 30%.

1.2 Miscarriage

Two RCTs (N = 1432) provided miscarriage data (Rose 2020; Ziebe 2013). The authors of both studies were able to clarify that the miscarriages were of clinical pregnancies. Based on correspondence, we were able to remove terminations of pregnancy that had been classified as miscarriage in these two studies (Appendix 13). Terminations of pregnancy as a result of aneuploidy or birth defect are accounted for in Analysis 1.6 and Analysis 1.7. All miscarriages were first‐trimester losses, apart from one in the control arm of Rose 2020, which was a midtrimester loss at 17 weeks' gestation. There were 19 miscarriages amongst the 704 women randomised to the GM‐CSF arm, and 26 miscarriages amongst the 728 women randomised to the control arm.

It is unclear whether use of GM‐CSF‐supplemented culture media makes any difference to miscarriage rate when compared to conventional culture media not supplemented with GM‐CSF (OR 0.75, 95% CI 0.41 to 1.36, 2 RCTs, N = 1432, I² = 0%, low‐quality evidence) (Analysis 1.2; Figure 4).

Figure 4

---

---

Forest plot of comparison: 1 GM‐CSF‐supplemented culture medium versus culture medium not supplemented with GM‐CSF, outcome: 1.2 Miscarriage.

The evidence suggests that if the miscarriage rate associated with conventional culture media not supplemented with GM‐CSF was 4%, the rate with the use of GM‐CSF‐supplemented culture media would be between 2% and 5%. It is worth noting that these figures are per woman randomised, hence the apparently very low miscarriage rates. They do not include miscarriages that occurred before the diagnosis of a clinical pregnancy on ultrasound scan, otherwise known as biochemical pregnancy losses.

Secondary outcomes

1.3 Clinical pregnancy

Three studies (N = 1532) reported clinical pregnancy rates (Rose 2020; Sbracia 2014; Ziebe 2013). Both Rose 2020 and Ziebe 2013 describe how an ultrasound scan was performed at seven weeks’ gestation in order to diagnose clinical pregnancy. Information on the methods of Sbracia 2014 was limited, as the study is only available as a conference abstract, and we received no response to our emails to the authors of the study. The authors of Sbracia 2014 describe pregnancy rate as their primary outcome, however there are no further details as to what stage pregnancy was diagnosed, and whether they were clinical pregnancies diagnosed with ultrasound. The authors report an "implantation rate", which we have taken to mean a biochemical pregnancy rate. Consequently, for the purposes of this review, we have assumed the pregnancy rate in Sbracia 2014 to be clinical.

There were 221 clinical pregnancies amongst the 754 women randomised to the GM‐CSF arm, and 205 clinical pregnancies amongst the 778 women randomised to the control arm.

We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the clinical pregnancy rate when compared to using a conventional culture medium not supplemented with GM‐CSF (OR 1.16, 95% CI 0.93 to 1.45, 3 RCTs, N = 1532 women, I² = 67%, low‐quality evidence) (Analysis 1.3; Figure 5).

Figure 5

---

---

Forest plot of comparison: 1 GM‐CSF‐supplemented culture medium versus culture medium not supplemented with GM‐CSF, outcome: 1.3 Clinical pregnancy.

The evidence suggests that if the clinical pregnancy rate associated with conventional culture media not supplemented with GM‐CSF was 26%, the rate with the use of GM‐CSF‐supplemented culture media would be between 25% and 34%.

1.4 Multiple gestation

Two studies (N = 1432) reported multiple gestation rate (Rose 2020; Ziebe 2013). The multiple gestation rate was clarified following correspondence with authors of both studies (Appendix 13). The authors of Ziebe 2013 also detail the incidence of monozygotic and dizygotic twins, but in this review we did not differentiate between types of twins. The authors of Rose 2020 report single‐embryo transfer as standard, but explain that four women in the control arm and six women in the intervention arm received double‐embryo transfer (Appendix 13). The authors of Ziebe 2013 describe how a maximum of two embryos were replaced per woman with a mean embryo transfer rate of 1.51 for the control arm and 1.49 for the GM‐CSF arm. There was one triplet pregnancy in the intervention arm of the study by Ziebe 2013. The remaining multiple gestations reported here were all twins.

There were 31 women with a multiple gestation amongst the 704 women randomised to the GM‐CSF arm, and 205 women with a multiple pregnancy amongst the 728 women randomised to the control arm.

We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the multiple pregnancy rate when compared to use of a conventional culture medium not supplemented with GM‐CSF (OR 1.24, 95% CI 0.73 to 2.10, 2 RCTs, N = 1432, I² = 35%, very low‐quality evidence) (Analysis 1.4).

The evidence suggests that if the multiple gestation rate associated with conventional culture media not supplemented with GM‐CSF was 4%, the rate with the use of GM‐CSF‐supplemented culture media would be between 3% and 7%.

1.5 Preterm birth

Two studies (N = 1432) reported the preterm birth rate, defined as the birth of a baby (or babies in the case of multiple pregnancy) under 37 weeks' gestation, per woman randomised (Rose 2020; Ziebe 2013). Preterm birth was detailed in the published study by Ziebe 2013. For singletons, the preterm birth data were easily extractable. For women with multiple gestations, the authors of Ziebe 2013 report gestational age at birth with a standard deviation, therefore we clarified these data with the study authors to establish the number of preterm births (Appendix 13). We sought preterm birth data through correspondence with the authors of Rose 2020 (Appendix 13). We counted twins and triplets that were born before 37 weeks as one event for this outcome, as we undertook ITT analysis.

There were 30 women with a preterm birth amongst the 704 women randomised to the GM‐CSF arm, and 26 women with a preterm birth amongst the 728 women randomised to the control arm.

We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the preterm birth rate when compared to using a conventional culture medium not supplemented by GM‐CSF (OR 1.20, 95% CI 0.70 to 2.04, 2 RCTs, N = 1432, I² = 76%, very low‐quality evidence) (Analysis 1.5).

The evidence suggests that if the preterm birth rate associated with conventional culture media not supplemented with GM‐CSF was 4%, the rate with the use of GM‐CSF‐supplemented culture media would be between 3% and 7%.

1.6 Birth defects

The authors of two studies (N = 1432) were able to provide details on birth defects following correspondence (Appendix 13) (Rose 2020; Ziebe 2013). The authors of Rose 2020 explained that there was one baby with multiple birth defects, which was detected antenatally (this participant was classified as experiencing a miscarriage in the published study, but we have clarified that this was a termination of pregnancy, and it has therefore been removed from the miscarriage group in this review). The authors of Ziebe 2013 provided details on 22 infants who were born with defects (three participants underwent termination of pregnancy for birth defects). We did not count any infants as having both birth defects and aneuploidy, but rather divided them into one of the two groups. We are not aware of twins within the birth defects group, and have assumed all data to be per woman randomised.

Thirteen women had a baby with a birth defect amongst the 704 women randomised to the GM‐CSF arm, and 10 women had a baby with a birth defect amongst the 728 women randomised to the control arm.

We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the rate of birth defects when compared to using a conventional culture medium not supplemented by GM‐CSF (OR 1.33, 95% CI 0.59 to 3.01, I² = 0%, 2 RCTs, N = 1432, low‐quality evidence) (Analysis 1.6).

The evidence suggests that if the birth defect rate associated with conventional culture media not supplemented with GM‐CSF was 1%, the rate with the use of GM‐CSF‐supplemented culture media would be between 1% and 4%.

1.7 Aneuploidy

Two studies (N = 1432) provided data regarding aneuploidy after we inquired about this outcome (Rose 2020; Ziebe 2013). The authors of Rose 2020 described how one baby had a trisomy, which was detected antenatally, and Ziebe 2013 reported that one baby had a trisomy detected antenatally.

No women had a baby with aneuploidy amongst the 704 women randomised to the GM‐CSF arm, and two women had a baby with aneuploidy amongst the 728 women randomised to the control arm. We are uncertain whether GM‐CSF‐supplemented culture media makes any difference to the rate of aneuploidy when compared to using a conventional culture medium not supplemented by GM‐CSF (OR 0.34, 95% CI 0.03 to 3.26, I² = 0%, 2 RCTs, N = 1432, low‐quality evidence) (Analysis 1.7).

The evidence suggests that if the aneuploidy rate associated with conventional culture media not supplemented with GM‐CSF was 0.3%, the rate with the use of GM‐CSF‐supplemented culture media would be between 0% and 0.9%.

1.8 Stillbirth

Two studies (N = 1432) reported stillbirth (Rose 2020; Ziebe 2013). Following correspondence, the authors of Rose 2020 provided data on stillbirth that were not published (Appendix 13). There were no stillbirths reported in either arm of the study, hence the OR was not estimable (Analysis 1.8). The average rate of stillbirth ranges from approximately 4 per 1000 total births in high‐income countries to approximately 28 per 1000 total births in low‐income countries such as sub‐Saharan Africa (Lawn 2016), therefore the stillbirth rate in this review is better than average for the high income countries.