Time‐lapse systems for embryo incubation and assessment in assisted reproduction
Abstract
Background
Embryo incubation and assessment is a vital step in assisted reproductive technology (ART). Traditionally, embryo assessment has been achieved by removing embryos from a conventional incubator daily for quality assessment by an embryologist, under a light microscope. Over recent years time‐lapse systems have been developed which can take digital images of embryos at frequent time intervals. This allows embryologists, with or without the assistance of embryo selection software, to assess the quality of the embryos without physically removing them from the incubator.
The potential advantages of a time‐lapse system (TLS) include the ability to maintain a stable culture environment, therefore limiting the exposure of embryos to changes in gas composition, temperature and movement. A TLS has the potential advantage of improving embryo selection for ART treatment by utilising additional information gained through continuously monitoring embryo development. Use of a TLS often adds significant extra cost onto an in vitro fertilisation (IVF) cycle.
Objectives
To determine the effect of a TLS compared to conventional embryo incubation and assessment on clinical outcomes in couples undergoing ART.
Search methods
We used standard methodology recommended by Cochrane. We searched the Cochrane Gynaecology and Fertility (CGF) Group trials register, CENTRAL, MEDLINE, Embase, CINAHL and two trials registers on 2 August 2017.
Selection criteria
We included randomised controlled trials (RCTs) in the following comparisons: comparing a TLS, with or without embryo selection software, versus conventional incubation with morphological assessment; and TLS with embryo selection software versus TLS without embryo selection software among couples undergoing ART.
Data collection and analysis
We used standard methodological procedures recommended by Cochrane. The primary review outcomes were live birth, miscarriage and stillbirth. Secondary outcomes were clinical pregnancy and cumulative clinical pregnancy. We reported quality of the evidence for important outcomes using GRADE methodology. We made the following comparisons.
TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment
TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images
TLS utilising embryo selection software versus conventional incubation and assessment
Main results
We included eight RCTs (N = 2303 women). The quality of the evidence ranged from very low to moderate. The main limitations were imprecision and risk of bias associated with lack of blinding of participants and researchers, and indirectness secondary to significant heterogeneity between interventions in some studies. There were no data on cumulative clinical pregnancy.
TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment
There is no evidence of a difference between the interventions in terms of live birth rates (odds ratio (OR) 0.73, 95% CI 0.47 to 1.13, 2 RCTs, N = 440, I2 = 11% , moderate‐quality evidence) and may also be no evidence of difference in miscarriage rates (OR 2.25, 95% CI 0.84 to 6.02, 2 RCTs, N = 440, I2 = 44%, low‐quality evidence). The evidence suggests that if the live birth rate associated with conventional incubation and assessment is 33%, the rate with use of TLS with conventional morphological assessment of still TLS images is between 19% and 36%; and that if the miscarriage rate with conventional incubation is 3%, the rate associated with conventional morphological assessment of still TLS images would be between 3% and 18%. There is no evidence of a difference between the interventions in the stillbirth rate (OR 1.00, 95% CI 0.13 to 7.49, 1 RCT, N = 76, low‐quality evidence). There is no evidence of a difference between the interventions in clinical pregnancy rates (OR 0.88, 95% CI 0.58 to 1.33, 3 RCTs, N = 489, I2 = 0%, moderate‐quality evidence).
TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images
No data were available on live birth or stillbirth. We are uncertain whether TLS utilising embryo selection software influences miscarriage rates (OR 1.39, 95% CI 0.64 to 3.01, 2 RCTs, N = 463, I2 = 0%, very low‐quality evidence) and there may be no difference in clinical pregnancy rates (OR 0.97, 95% CI 0.67 to 1.42, 2 RCTs, N = 463, I2 = 0%, low‐quality evidence). The evidence suggests that if the miscarriage rate associated with assessment of still TLS images is 5%, the rate with embryo selection software would be between 3% and 14%.
TLS utilising embryo selection software versus conventional incubation and assessment
There is no evidence of a difference between TLS utilising embryo selection software and conventional incubation improving live birth rates (OR 1.21, 95% CI 0.96 to 1.54, 2 RCTs, N = 1017, I2 = 0%, very low‐quality evidence). We are uncertain whether TLS influences miscarriage rates (OR 0.73, 95% CI 0.49 to 1.08, 3 RCTs, N = 1351, I2 = 0%, very low‐quality evidence). The evidence suggests that if the live birth rate associated with no TLS is 38%, the rate with use of conventional incubation would be between 36% and 58%, and that if miscarriage rate with conventional incubation is 9%, the rate associated with TLS would be between 4% and 10%. No data on stillbirths were available. It was uncertain whether the intervention influenced clinical pregnancy rates (OR 1.17, 95% CI 0.94 to 1.45, 3 RCTs, N = 1351, I2 = 42%, very low‐quality evidence).
Authors' conclusions
There is insufficient evidence of differences in live birth, miscarriage, stillbirth or clinical pregnancy to choose between TLS, with or without embryo selection software, and conventional incubation. The studies were at high risk of bias for randomisation and allocation concealment, the result should be interpreted with extreme caution.
PICOs
Plain language summary
Time‐lapse systems for embryo incubation and embryo assessment for couples undergoing IVF and ICSI
Review question
We wanted to determine whether a time‐lapse system (TLS) would improve the chances of a pregnancy and liveborn baby, and reduce the risk of miscarriage and stillbirth.
Background
In vitro fertilisation (IVF) and intracytoplasmic sperm injection (ICSI) are processes whereby a woman's eggs and a man's sperm are combined to achieve fertilisation outside of the body. Embryos are stored in an incubator and replaced into the woman between day 2 and 5 of development. Usually, embryos are removed from a conventional incubator for assessment, under a microscope, of their quality and stage of development. A TLS can take images of embryos at frequent time intervals, which allows assessment without removing the embryos from the incubator. A TLS can also apply a software programme that assists the embryologist in selecting the best quality embryo for replacement, potentially improving the chance of a liveborn baby.
Study Characteristics
The evidence is current to August 2017. We included eight studies (randomised controlled trials) of 2303 women undergoing IVF or ICSI. There were three different study designs: 1) TLS with conventional assessment of still TLS images versus conventional incubation and assessment, 2) TLS utilising embryo selection software versus TLS with conventional assessment of still TLS images, and 3) TLS utilising embryo selection software versus conventional incubation and assessment.
Trials included women undergoing IVF, and ICSI; some trials involve frozen embryo transfer and others fresh; one trial includes women using donor eggs, and the remainder use the woman's own eggs; the day of embryo transfer differs between trials; and in some only one embryo is replaced whereas in others, multiple embryos are replaced. We have taken account of these differences when assessing quality of the evidence. These differences should be seen as reflecting 'real world' practices, where there are variations in practice.
What the review found
TLS with conventional assessment of still TLS images versus conventional incubation and assessment
There is probably no difference between these interventions in live birth rates or pregnancy rates (moderate‐quality evidence), miscarriage rates or stillbirth rates (low‐quality evidence). The evidence suggests that if the live birth rate associated with conventional incubation and assessment is 33%, the rate with use of TLS with conventional morphological assessment of still TLS images is between 19% and 36%.
TLS utilising embryo selection software versus TLS with conventional assessment of still TLS images
No data were available on live birth or stillbirth. We are uncertain whether TLS utilising embryo selection software influences miscarriage rates, compared with TLS with conventional morphological assessment of still TLS images (very low‐quality evidence) and clinical pregnancy rates (low‐quality evidence). The evidence suggests that if the miscarriage rate associated with assessment of still TLS images is 5%, the rate with embryo selection software would be between 3% and 14%.
TLS utilising embryo selection software versus conventional incubation and assessment
There is no evidence from well designed studies that TLS utilising embryo selection software improves live birth or pregnancy rates compared to no TLS (very low‐quality evidence) or reduces miscarriages (very low‐quality evidence). The evidence suggests that if the live birth rate associated with no TLS is 38%, the rate with use of conventional incubation would be between 36% and 58%.
Patients need to be aware that there is no good evidence that TLS is more effective than conventional methods of embryo incubation. Women may wish to take part in RCTs on TLS in order to add to the existing evidence base, and help guide ART patients in the future.
Quality of the evidence
The quality of the evidence ranged from very low to moderate.
Authors' conclusions
Summary of findings
| TLS with conventional morphological assessment of still TLS images compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing assisted reproductive technology | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with conventional incubation and assessment | Risk with TLS with conventional morphological assessment of still TLS images | |||||
| Live birth | 333 per 1,000 | 267 per 1,000 | OR 0.73 | 440 | ⊕⊕⊕⊝ | |
| Miscarriage | 37 per 1,000 | 83 per 1,000 | OR 2.25 | 440 | ⊕⊕⊝⊝ | |
| Stillbirth | 53 per 1,000 | 53 per 1,000 | OR 1.00 | 76 | ⊕⊕⊝⊝ | |
| Clinical pregnancy | 353 per 1,000 | 310 per 1,000 | OR 0.88 | 489 | ⊕⊕⊕⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the evidence for live birth once for imprecision due to there only being two trials, one in good prognosis patients and the other in poorer prognosis patients, totalling 440 women. | ||||||
| TLS utilising embryo selection software compared to TLS with conventional morphological assessment of still TLS images for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing assisted reproductive technology | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with TLS with conventional morphological assessment of still TLS images (trial design 2) | Risk with TLS utilizing embryo selection software | |||||
| Live birth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Miscarriage | 54 per 1,000 | 74 per 1,000 | OR 1.39 | 463 | ⊕⊝⊝⊝ | |
| Stillbirth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Clinical pregnancy | 537 per 1,000 | 529 per 1,000 | OR 0.97 | 463 | ⊕⊕⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the quality of evidence for miscarriage three times: once for risk of bias, once for indirectness and once for imprecision. The risk of bias is secondary to performance bias due to varying days and numbers of embryos transferred, decided upon by an unblinded embryologist. There is heterogeneity between the study designs leading to indirectness; one included study involved removing embryos for bench‐top microscopy daily in both the intervention and control arms, whereas the other left embryos in the intervention and control arms undisturbed. The imprecision is secondary to broad confidence intervals. | ||||||
| TLS utilising embryo selection software compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing ART | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with conventional incubation and assessment | Risk with TLS utilising embryo selection software | |||||
| Live birth | 381 per 1,000 | 461 per 1,000 | OR 1.21 | 1017 | ⊕⊝⊝⊝ | |
| Miscarriage | 94 per 1,000 | 70 per 1,000 | OR 0.73 | 1351 | ⊕⊝⊝⊝ | |
| Stillbirth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Clinical pregnancy | 545 per 1,000 | 584 per 1,000 | OR 1.17 | 1351 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the quality of evidence for live birth twice for risk of bias and once for indirectness. All included studies are at high risk of performance bias owing to lack of blinding or incomplete blinding. There was also high risk of selection bias. In one study, the randomisation of participants was undertaken by the principal investigator and allocation concealment was not described. In another study, some patients could request the intervention and this request was granted. The indirectness was due to one included study undertaking multiple embryo transfers per woman, and included women receiving donor oocytes from younger women. | ||||||
Background
Description of the condition
Embryo incubation is a critical step in all in vitro fertilisation (IVF) procedures. Embryo development within media in culture dishes in an incubator is a dynamic process, moving through the fertilisation stage to cleavage stage and then to the blastocyst stage in some cases. Throughout the incubation period, embryos are usually inspected at specific time points to provide a brief 'snap shot' assessment of the way the embryo is developing (morphological features). Embryologists apply a tiered grading system based on the morphology of the embryo in order to predict the potential for implantation and a successful pregnancy (Cummins 1986; Neuber 2003; Scott 2003; Scott 2003a; Shoukir 1997). A consensus on the minimum data set required for the accurate description of embryo morphology was established by Alpha Scientists in Reproductive Medicine and European Society of Human Reproduction and Embryology (ESHRE) Special Interest Group of Embryology (Alpha & ESHRE SIG 2011). A consensus on timings of observation of fertilised oocytes and embryos was established and deemed critical to the ability to compare results between different laboratories. The recommended checks, in hours, following insemination are:
-
a fertilisation check at 17 hours, a syngamy (fusion of gametes) check at 23 hours;
-
an early cleavage check at 26 hours post‐intracytoplasmic sperm injection (ICSI) or 28 hours post‐IVF;
-
day 2 embryo assessment at 44 hours;
-
day 3 embryo assessment at 68 hours;
-
day 4 embryo assessment at 92 hours;
-
day 5 embryo assessment at 116 hours.
Traditionally the checks have been achieved by physically removing embryos from the controlled environment of the incubator to analyse them under a light microscope for assessment of embryo development and quality. This practice exposes the embryos to the potentially suboptimal conditions of the environment outside of the incubator and human handling (Meseguer 2012a). Time‐lapse systems (TLSs) have evolved over recent years to increase the frequency of morphological observations whilst minimising the impact of the external environment and human handling on embryo development.
Description of the intervention
A TLS is a device which takes digital images of embryos at set time intervals, for example every 5 to 15 minutes. The system can be installed into an existing embryo incubator or can exist as a combined time‐lapse incubation system. The images are compiled using specialist software to create a time‐lapse sequence of embryo development. Images can be digitally displayed as a time‐lapse sequence on an external monitor to allow embryologists to assess the dynamic morphology of embryos thus negating the need for the embryologist to remove embryos from the incubator. Some TLSs also utilise computer‐assisted assessment of developmental milestones of embryos, also known as morphokinetic parameters, to offer a semi‐quantitative process of embryo evaluation (Conaghan 2013). These cell‐tracking software algorithms utilise data such as the timing of embryonic development events, and have evolved as a non‐invasive, non‐subjective way of attempting to improve the selection of embryos with the highest implantation potential. Some clinics have developed their own algorithms to adapt the standardised one that comes with the TLS device (Petersen 2016).
There are a number of commercially available TLSs developed by various manufacturers. TLSs are available as devices that can be placed within existing conventional incubators, and some exist with an integrated incubator. The integrated TLS combines both the time‐lapse cameras and the incubator in one device.
How the intervention might work
There are two potential benefits of a TLS. Firstly, an advantage may lie with the undisturbed nature of the culture conditions, whereby image for embryo assessment can be obtained without removing embryos from the incubator environment for conventional bench‐top light microscopy (which usually includes heated microscope stages). This minimises the exposure of embryos to both human handling and changes in air temperature and gas composition. This may lead to improved culture conditions.
A second potential advantage may be owing to the ability of a TLS to accumulate detailed time‐lapse images of embryo development at regular time intervals. This includes the timing of cell divisions, intervals between cell cycles, and other development factors (eg. dynamic pronuclei patterns, presence of multinucleation and fragmentation, and blastomere symmetry). Many of these features which are transient events may be missed by using standard morphological assessment at set time intervals. These detailed time‐lapse sequences can be utilised with or without cell‐tracking software algorithms as an adjunct to standard morphological assessment, to select the embryo with the highest implantation potential for transfer. This is important because there is a clear correlation between embryo morphology and viability (Finn 2010; Neuber 2006). The ability to select the highest quality embryo at an optimal stage of development for replacement first in an assisted reproductive technology (ART) cycle may lead to a reduction in time to pregnancy, and reduced need for subsequent embryo transfers. It is worth noting that the different makes of TLS follow the same basic principles but vary in technical detail such as gas mixture, temperature, group or single culture, dark or light field microscopy.
In order to assess the potential advantage of TLSs (i.e. the stable culture environment, or the time‐lapse sequence of images which can be assessed with cell‐tracking algorithms, or both), studies can be grouped into the following three designs.
Trial design 1: TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment
-
These studies control for how the embryos are selected for transfer, but the incubation differs. This will help to establish whether the culture conditions of the TLS potentially impact on favourable outcomes such as pregnancy and live birth
Trial design 2: TLS utilizing embryo selection software versus TLS with conventional morphological assessment of still TLS images
-
These studies control for the culture environment, with both arms of the trial being incubated in a TLS, but the way in which embryos are selected for transfer is tested. This study design will help to establish whether embryo selection software improves the selection of top‐quality embryos, and increases the pregnancy and live birth rate
Trial design 3: TLS utilizing embryo selection software versus conventional incubation and assessment
-
These studies aim to establish whether a combination of both the stable culture environment and the embryo selection software, is superior to conventional embryo incubation and assessment at improving pregnancy and live birth.
Why it is important to do this review
New interventions, such as TLSs, should be evaluated by randomised controlled trials to establish their safety, clinical effectiveness and cost‐effectiveness (Campbell 2000; Harper 2012). Countering the potential benefits outlined above, a TLS involves exposing embryos to light during image acquisition, at predetermined intervals. Furthermore, the authorities responsible for the regulation of fertility clinics and research involving human embryos have a responsibility to provide impartial and authoritative information to prospective and current patients on fertility treatments to aid them in making informed decisions on their care (ACART; HFEA). Therefore it is vital that up‐to‐date and thorough systematic reviews, accessible to patients and healthcare workers, are published on the topic. This will enable information on the technology's success rates in terms of live birth or ongoing pregnancy rate, and safety in terms of adverse events, to be accessible and help guide informed decision making.
This is an update of a Cochrane review published under the same title in 2015. The original review included two completed RCTs and interim data from one ongoing RCT. The results of the original review showed that there was insufficient evidence of differences in live birth, miscarriage, stillbirth or clinical pregnancy to choose between TLS and conventional incubation.
This updated review is aimed at establishing whether there is evidence of any overall benefit of culturing embryos in a TLS with or without embryo selection software, over current conventional embryo incubation and assessment.
Objectives
To determine the effect of a time‐lapse system (TLS) compared to conventional embryo incubation and assessment on clinical outcomes in couples undergoing assisted reproductive technology (ART).
Methods
Criteria for considering studies for this review
Types of studies
Inclusions: any randomised controlled trial (RCT), whether published or not, which in principle could answer questions regarding clinical (post‐implantation) outcomes.
Exclusions: quasi‐randomised and other concurrently controlled studies were excluded. We excluded trials that randomised oocytes or embryos as it would not be possible to compare clinical outcomes. We excluded cross‐over trials as the design is not valid in this context.
Types of participants
Couples of any age undergoing assisted reproduction where embryo incubation was required.
Types of interventions
-
Time‐lapse system (TLS) with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1)
-
TLS utilizing embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2)
-
TLS utilizing embryo selection software versus conventional incubation and assessment (trial design 3)
Any type of TLS , using any type of embryo selection software and any type of conventional incubator, was eligible.
Types of outcome measures
Primary outcomes
1. Live birth rate per couple randomly assigned
2. Miscarriage, and stillbirth
Secondary outcomes
3. Clinical pregnancy, defined as evidence of a gestational sac, confirmed by ultrasound, per couple randomly assigned
4. Cumulative clinical pregnancy rate, per couple randomly assigned
Search methods for identification of studies
Two review authors (SA and PB) searched, from the inception of the databases to 2 August 2017, for all published and unpublished RCTs of time‐lapse systems, without language restrictions and in consultation with the Cochrane Gynaecology and Fertility Group (CGFG) Information Specialist. We used both electronic searches of bibliographic databases and handsearching as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).
Electronic searches
We searched in the following electronic databases, trial registers and websites.
-
Cochrane Gynaecology and Fertility Group specialised register, PROCITE platform (searched 2 August 2017) (Appendix 1)
-
Cochrane Central Register of Studies Online (CRSO), web platform (searched 2 August 2017) (Appendix 2)
-
MEDLINE® In‐Process & Other Non‐Indexed Citations, OVID platform (searched from 1946 to 2 August 2017) (Appendix 3)
-
Embase, OVID platform (searched from 1980 to 2 August 2017) (Appendix 4)
-
Cumulative Index to Nursing and Allied Health (CINAHL), EBSCO platform (searched from 1961 to 2 August 2017) (Appendix 5)
For MEDLINE, we used the Cochrane highly sensitive search strategy for identifying RCTs: sensitivity and precision maximizing version (2008 revision), Ovid format (Higgins 2011). The LILACS search strategy was combined with the RCT filter of the IAHx interface.
Other electronic sources of trials (web platforms, all searched August 2017) included the following.
-
Trial registers for ongoing and registered trials; the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) portal (www.apps.who.int/trialsearch/) and ClinicalTrials.gov (www.clinicaltrials.gov)
-
The Web of Knowledge (wokinfo.com/)
-
Proquest Dissertations and Theses (search.proquest.com)
-
Grey literature through the System for Information on Grey Literature in Europe 'OpenGrey' (www.opengrey.eu/).
Searching other resources
We attempted to identify additional relevant RCTs by using the following methods:
-
contact with authors of all RCTs identified by other methods;
-
contact with manufacturers of TLSs;
-
handsearching of selected journals in obstetrics, gynaecology and reproductive medicine, as well as conference proceedings (for abstracts) of the European Society for Human Reproduction and Embryology (ESHRE) and the American Society for Reproductive Medicine (ASRM);
-
contacting known experts and personal contacts regarding unpublished materials;
-
searching the citation lists of all identified articles for any relevant reference.
Data collection and analysis
Selection of studies
Two authors (SA and PB) independently scanned the titles and abstracts of the articles retrieved by the search. We then obtained full texts of potentially eligible studies and examined these independently for their suitability according to the inclusion criteria. In the case of doubt between the two authors, a third author (CF) was consulted to gain consensus on whether to include the trial or not. We documented the selection process with a Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) flow chart.
Data extraction and management
Two authors (SA and PB) independently obtained and extracted data. In the case of disagreement between the two authors, they consulted a third author to achieve consensus (CF). They extracted data using a data extraction form designed and piloted by the authors. If studies were reported in multiple publications, we extracted data from the different publications and then combined these into a single data extraction form so no data were omitted. The following characteristics of included studies were included in the data extraction form:
-
methods;
-
participants;
-
interventions;
-
outcomes, including adverse events;
-
funding source for studies.
Assessment of risk of bias in included studies
Two review authors (SA and PB) independently assessed the risk of bias in included studies using the Cochrane 'Risk of bias' assessment tool. We evaluated all included studies for the following: adequacy of sequence generation and allocation concealment; adequacy of blinding of couples, providers and outcome assessors; completeness of outcome data; risk of selective outcome reporting; and risk of other potential sources of bias (Higgins 2011).
Disagreements between authors were resolved by consensus and consulting a third reviewer (VJ). The results of the assessment of risk of bias are presented in the 'Characteristics of included studies' table and a 'Summary of findings' table.
Measures of treatment effect
For dichotomous data (for example, live birth or not), we calculated Mantel‐Haenszel odds ratios (ORs) and the 95% confidence intervals (CIs).
Unit of analysis issues
The data were analysed per couple randomised. Studies randomising oocytes or embryos were excluded.
Dealing with missing data
If relevant data were missing from an included study, we contacted the original investigators of the trial to request the missing data. All original investigators were contacted. In particular, we obtained clinical pregnancy and live birth data from Park 2015, miscarriage and clinical pregnancy data per woman randomised for Goodman 2016, and live birth and stillbirth data from Kahraman 2013. If participants were described as 'lost to follow up' without a specified reason, we assumed the participant did not experience the event or outcome (that is, did not become pregnant).
Assessment of heterogeneity
We considered whether the clinical and methodological characteristics of the included studies were sufficiently similar for meta‐analysis to provide a clinically meaningful summary. We assessed statistical heterogeneity by measuring the I² statistic. We assumed that there was substantial heterogeneity when I² was calculated to be greater than 50% (Higgins 2011).
Assessment of reporting biases
In view of the difficulty of detecting and correcting for publication bias and other reporting biases, the authors aimed to minimise their potential impact by ensuring a comprehensive search for eligible studies and by being alert to duplication of data. Within‐study reporting bias was assessed, and assessed as low risk if all of the study's prespecified primary outcomes were reported as outlined in the study's protocol.
Data synthesis
Where sufficient data were available, we combined the data for the primary outcomes by using a fixed‐effect model in the following comparisons.
-
TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1)
-
TLS utilizing embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2)
-
TLS utilizing embryo selection software versus conventional incubation and assessment (trial design 3)
Subgroup analysis and investigation of heterogeneity
Where sufficient data were available, we aimed to conduct the following subgroup analyses to determine the potential causes of heterogeneity for the live birth and clinical pregnancy outcomes:
-
donor oocytes (from donors of any age) versus autologous oocytes (from women of any age);
-
fresh cycles (where embryos were replaced either at cleavage stage (day 3) or blastocyst (day 5)) versus frozen cycles (where frozen embryos were replaced in an assisted reproductive technology cycle).
If we detected substantial heterogeneity we hoped to explore this by employing the random‐effects model. We aimed to take any statistical heterogeneity into account when interpreting the results, especially if there was any variation in the direction of effect.
Sensitivity analysis
We planned to undertake sensitivity analyses for the review outcomes to determine whether the results were robust to decisions made during the review process. These analyses would have included consideration of whether the review conclusions would have differed if:
-
the summary effect measure was relative risk rather than odds ratio;
-
eligibility was restricted to studies with low risk of bias for randomisation and allocation concealment.
Overall quality of the body of evidence: 'Summary of findings' table
We prepared 'Summary of findings' tables using GRADEpro GDT (www.gradepro.orgGRADEproGDT 2015) and Cochrane methods in March 2018. These tables evaluate the overall quality of the body of evidence for the main review outcomes (live birth, miscarriage, stillbirth and clinical pregnancy) for the review comparisons:
-
TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1);
-
TLS utilizing embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2); and
-
TLS utilizing embryo selection software versus conventional incubation and assessment (trial design 3).
We assessed the quality of the evidence using GRADE criteria: risk of bias, consistency of effect, imprecision, indirectness and publication bias. Judgements about evidence quality (high, moderate, low or very low) was made by two review authors who worked independently (SA and PB), and resolved disagreements by discussion. Judgements were justified, documented, and incorporated into reporting of results for each outcome.
Results
Description of studies
Results of the search
The first iteration of this review included three parallel‐design randomised controlled trials (RCTs) from a search which retrieved 33 articles in total (new studies added were Kahraman 2013; Kovacs 2013; Rubio 2014). Two further searches in 2016 and 2017 retrieved 82 and 293 articles respectively. We retrieved a further four articles through handsearching. We screened 266 articles after removing duplicates. Twenty‐five full‐text articles were potentially eligible and we retrieved these in full text. We identified five new studies which met our inclusion criteria (Goodman 2016; Kaser 2017; Park 2015; Wu 2016; Yang 2017). We excluded the remaining 20 studies for the following reasons: three studies were not RCTs; three were systematic reviews; two were letters; nine randomised embryos or oocytes; two were pseudo‐randomised; and in one study we were unable to determine the nature of the control group despite attempts at contacting the authors. (Figure 1, Excluded studies). In total, we included eight RCTs in the quantitative synthesis.
Study flow diagram.
Included studies
Study design and setting
Eight RCTs are included in this review. The largest study is a multi‐centre RCT conducted in Spain, which was included in the first iteration of this review (Rubio 2014). Three new single‐centre studies added to this review were conducted in the USA (Goodman 2016; Kaser 2017; Wu 2016). Two further new single‐centre studies were added; one completed study was undertaken in Sweden (Park 2015), and another ongoing study was undertaken in China, from which we have interim results (Yang 2017). The final two studies were included in the first iteration of this review, one of which was a single‐centre RCT conducted in Turkey (Kahraman 2013), and the other is the completed results of a single‐centre RCT in Hungary (Kovacs 2013).
Participants
The studies included 2303 infertile couples undergoing assisted reproductive technology (ART). Three studies included couples undergoing intracytoplasmic sperm injection (ICSI) alone (Kahraman 2013; Rubio 2014; Park 2015). Two studies included couples undergoing in vitro fertilisation (IVF) (Goodman 2016; Kovacs 2013). The remaining studies describe including couples undergoing both IVF and ICSI (Kaser 2017; Wu 2016; Yang 2017).
The largest study is Rubio 2014, with 856 participants. The second largest study has 364 participants (Park 2015), followed by the interim results of Yang 2017, with 334 participants. The fourth largest study has 300 participants (Goodman 2016), followed by Kaser 2017, with 163. The sixth largest study is the completed results of Kovacs 2013, with 161 participants. The remaining two studies are relatively small, with 76 and 49 participants (Kahraman 2013; Wu 2016, respectively).
All studies utilised the autologous oocytes of the women randomised into their study with the exception of Rubio 2014, which included couples undergoing ART with autologous or donor oocytes. The proportion of couples receiving donor oocytes in this study is unknown. Most donor oocytes in this study were used in fresh cycles, however some donor oocytes were obtained from an oocyte bank and were therefore vitrified.
All studies included women undergoing fresh embryo transfer, hence no cumulative cycle results are available. The majority of studies undertook single embryo transfer (Kahraman 2013; Kaser 2017; Kovacs 2013; Park 2015; Yang 2017). One study describes replacing between one and three embryos based on published American Society for Reproductive Medicine (ASRM) committee guidance and patient preferences (Goodman 2016). Another study undertook multiple embryo transfer (Rubio 2014), and another did not disclose the number of embryos transferred (Wu 2016).
The reported causes of infertility varied between studies. Some studies specifically described their participants as 'good prognosis patients' (e.g. Rubio 2014; Yang 2017). One study specifically described their participants as 'poor prognosis patients', but gave no further information (Wu 2016). One study described 'tubo‐peritoneal factor' as the cause of infertility (Kahraman 2013), and another described male‐factor infertility being present in more than 99% of participants in both arms and female‐factor infertility being present in approximately 20% of participants in both arms (Park 2015). In Kovacs 2013, various causes of infertility in participants was described ("male, tubal, unexplained etc."). One study described "a combination of anovulation, diminished ovarian reserve, endometriosis, male factor, tubal, unknown, and uterine" as causes of infertility (Kaser 2017). Finally, in Goodman 2016, a range of infertility diagnoses was described, from "unexplained, ovulatory dysfunction, male factor, tubal factor, low ovarian reserve, AMA, endometriosis, mixed factors and other".
Interventions
We have sought to divide studies into three comparisons depending on the nature of the intervention and the control, in order to truly assess if, and where, the benefit of a time‐lapse system (TLS) lies.
1) TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1)
Three studies undertook this comparison (Kahraman 2013; Park 2015; Wu 2016). All three studies utilised an integrated TLS, and all three had two arms. Embryo transfer (ET) was undertaken at blastocyst in Kahraman 2013, day three in Wu 2016, and day two in Park 2015. It was confirmed on correspondence with the authors of one study that no embryo selection software was utilised in the intervention arm (Kahraman 2013). Embryos were left undisturbed in the TLS in the intervention arm in all three studies. In the control arm, embryos in all studies were assessed by conventional morphology under a benchtop microscope.
2) TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2)
Two studies undertook this comparison (Goodman 2016; Kaser 2017). One study utilised an integrated TLS (Goodman 2016), and the other utilised a TLS which was placed inside a conventional incubator (Kaser 2017). The embryos in the intervention arms were selected for transfer according to the information obtained from the embryo selection software, however the embryos of the women randomised to the intervention arm in one study were removed from the incubator for conventional benchtop morphology in addition to TLS selection (Kaser 2017). In addition, the embryos in the control arm of this study were assessed with conventional morphological assessment under a benchtop microscope. TLS images were not utilised for the selection of embryos for replacement in the control arm.
One study was a three‐arm study (Kaser 2017). There were two intervention arms; both were TLS utilising embryo selection software, but one arm undertook ET on day three and the other undertook ET on day five. The control arm undertook ET on day five. The other study had two arms, and ET was undertaken on day three or day five (Goodman 2016).
We conducted in‐depth discussions with the authors of Kaser 2017, and it was decided that trial design 2 was the most appropriate comparison, given that embryo selection software was utilised and the trial design tested the embryo‐selection element of the TLS software.
3) TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3)
Three studies undertook this comparison (Kovacs 2013; Rubio 2014; Yang 2017). Two of these utilised a TLS which was placed inside a conventional incubator (Kovacs 2013; Yang 2017); the other study utilised an integrated TLS (Rubio 2014). In Rubio 2014, ET was undertaken on days three and five in both arms; in Kovacs 2013, blastocyst transfer was undertaken in both arms. One study undertook ET on day three in the intervention arm and day five (blastocyst) in the control arm (Yang 2017). We took methodological advice on Yang 2017, and made the decision to keep the study in this review despite the differing days of ET. We gave this study a high 'Risk of bias' rating due to this within‐study imbalance.
Outcomes
All eight studies reported clinical pregnancy rates per couple. Miscarriage data were available for all included studies except for Wu 2016. In the case of Yang 2017, the miscarriage rate was calculated by us using ongoing pregnancy data minus clinical pregnancy data. Miscarriage data are confirmed to be loss of a clinical pregnancy (not biochemical) in the studies by Kahraman 2013; Kaser 2017; Kovacs 2013; Park 2015; and Yang 2017. In two studies the miscarriage data were a mixture of biochemical and clinical pregnancy losses (Goodman 2016; Rubio 2014). Unfortunately the authors of these studies were unable to provide only miscarriage data from clinical pregnancies. In these cases we have taken the pragmatic view to include these data as the majority of the pregnancy losses in these studies are from clinical pregnancies, according to the authors.
We obtained live birth data for three studies following communication with the authors (Kahraman 2013; Kovacs 2013; Park 2015). For Rubio 2014, we obtained data from a related publication and conference abstract pertaining to the same study (Insua 2017; Insua 2015). We obtained stillbirth data from two studies following communication with the authors (Kahraman 2013; Park 2015).
Excluded studies
We excluded 20 studies from the review for the following reasons.
-
Three were not RCTs
-
Three were systematic reviews
-
Two were letters
-
Nine randomised embryos or oocytes opposed to women or couples
-
In one study we were unable to determine the nature of the control group despite attempts at contacting the authors
-
Two studies were pseudo‐randomised
Risk of bias in included studies
For details of the 'Risk of bias' assessments, see Figure 2 and Figure 3.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Allocation
Sequence generation
Seven of the eight studies were at low risk of selection bias related to sequence generation. Five used a computer‐generated randomisation list (Goodman 2016; Kahraman 2013; Kaser 2017; Park 2015; Wu 2016). One study utilised a random number table (Yang 2017). One study undertook paired randomisation whereby two envelopes containing time‐lapse or control group assignments were prepared and the first patient was randomly assigned to one of the groups and the next patient received the other assignment (Kovacs 2013). This was repeated with patient numbers three and four, and so on.
We deemed one study to have high risk of bias in this domain because, although it undertook adequate random sequence generation, some women were able to request the intervention, and in some cases this request was granted (Rubio 2014). The authors of this study assured us that this preferential allocation occurred on a minority of occasions and the vast majority of participants were truly randomised, therefore we have maintained that this is a randomised controlled trial (RCT).
Allocation concealment
Five studies described methods of allocation concealment which were at low risk of selection bias (Goodman 2016; Kahraman 2013; Kaser 2017; Park 2015; Yang 2017). In each of these studies, the randomisation list or numbered opaque sealed envelopes were held and administered by personnel not directly involved in the recruitment of participants.
We deemed two studies to be at high risk of bias for this domain (Kovacs 2013; Rubio 2014). In the case of Kovacs 2013, the randomisation was carried out by the principal investigator who was involved in the study. In the case of Rubio 2014, it was described that in some cases the allocation was non‐random.
We judged one study to be at unclear risk of bias in this domain due to the limited description of randomisation (Wu 2016). We understand it was undertaken by a member of the team not associated with the treatment cycle and then subsequently the designation was reported to the embryology staff who processed the participant's oocytes/embryos. However, it is unclear how the randomisation list was stored, at what point the participants were randomised, and whether the person undertaking randomisation was responsible for recruitment.
Blinding
Blinding of participants and personnel (performance bias)
Three studies blinded their couples and this blinding was not broken unless participants withdrew from the study (Goodman 2016; Kahraman 2013; Park 2015). Clinicians involved in the study were also blinded until after embryo transfer. One study described blinding the embryologist to the Eeva rating for the morphological assessment of embryos (Kaser 2017). The participants and physicians were all blinded to the time‐lapse system (TLS) ratings. In addition, the sonographer was blinded in Goodman 2016, and the statistician was blinded in Park 2015.
Three studies did not blind or maintain blinding of their participating couples (Kovacs 2013; Rubio 2014; Yang 2017). In two of these studies the clinical staff were not blinded either (Kovacs 2013; Yang 2017). The gynaecologist and the statistician were blinded in Rubio 2014. We assessed these three studies as being at high risk of this bias.
We deemed one study as having high risk of performance bias as the blinding was not described and it would have been impossible to blind the embryologist (Wu 2016). We have been unable to contact the authors for further clarification.
None of the included studies blinded the embryologists, but this would have been impossible. We considered a lack of blinding of embryologists as a reason for high risk of performance bias. This renders all included studies as having a high risk of performance bias. In some studies, the lack of blinding may have influenced the number or day of transfer. In addition, it is impossible to remove the risk of performance bias when the person selecting the embryo for transfer is unblinded.
Blinding of outcome assessors (detection bias)
We judged all eight studies to be at low risk of detection bias because the outcomes (live birth, clinical pregnancy, miscarriage and stillbirth) are objective, and cannot be influenced by the knowledge of the intervention. Two studies described how those staff undertaking the ultrasounds were blinded to the intervention (Goodman 2016; Rubio 2014). The remaining studies did not blind their outcome assessors, however we still deemed these studies as having low risk of bias due to the reason described above.
Incomplete outcome data
We deemed the following studies to be at low risk of attrition bias:
-
Goodman 2016, because we were able to obtain the outcome data from the five women excluded after randomisation;
-
Kahraman 2013, because the 12 couples who dropped out after randomisation were accounted for, and the reasons were clearly stated;
-
Kaser 2017, because all data were presented in their paper as intention‐to‐treat;
-
Park 2015, because there was only one woman excluded from analysis due to having been accidentally randomised twice;
-
Wu 2016, because the small number of patients excluded were accounted for according to pre‐determined grounds for exclusion; and
-
Rubio 2014, because the 13 couples who were excluded following randomisation were accounted for and were a very small proportion of the total number of couples randomised.
We judged one study to be at high risk of attrition bias because a large proportion of the couples recruited were excluded from the trial (22 out of 161 couples randomised) (Kovacs 2013). On communication with the author it was made clear that these excluded couples were 'dropouts'. Reasons for dropouts were provided, however not all had reasons which tallied with pre‐determined exclusion criteria and, with such a high attrition rate, this study is at high risk of attrition bias.
We deemed one study to have unclear risk of attrition bias (Yang 2017). Attrition was mentioned, but reasons were not provided.
We undertook an intention‐to‐treat analysis on all dichotomous outcomes, using data from those women excluded post‐randomisation where possible.
Selective reporting
We considered six studies to be at low risk of reporting bias because they reported and published all outcomes they set out to investigate (Goodman 2016; Kahraman 2013; Kaser 2017; Kovacs 2013; Park 2015; Rubio 2014). This was confirmed on communication with authors and by referencing against online trials registers if they were available.
We considered one study to be at unclear risk of reporting bias because we had no access to their protocol and we couldn't contact the authors to ask if they published all outcomes they set out to assess (Wu 2016).
We deemed one study to be at high risk of reporting bias because on communication with authors, they mentioned a series of outcomes, including implantation rates, twin pregnancy rate (monozygotic twins), and ectopic pregnancy which were never published (Yang 2017). We concede that this is only an interim report and that such reports do not always include all secondary outcomes. However, the fact that this publication is the interim analysis of a full study, and it is not clear from communication with authors that this was a planned analysis, means the study is at risk of selective reporting.
Other potential sources of bias
We found no potential sources of within‐study bias in Goodman 2016, Kahraman 2013, Kaser 2017, Park 2015, Rubio 2014, and Wu 2016. We assessed these studies as having low risk of this bias.
We deemed one study to have an unclear risk of within‐study bias (Kovacs 2013). Data included in this review were obtained from the author directly, however these have not be published. Interim reporting and analysis of results from this study are available in various published sources, with differing results.
We assessed one study, Yang 2017, as having a high risk of within‐study bias. This is due to the difference in day of embryo transfer between arms of study (day three for intervention and day five for control). This difference in maturity of the embryo may have had an impact on the likelihood of an ongoing pregnancy.
Effects of interventions
See: Summary of findings for the main comparison TLS with conventional morphological assessment of still TLS images compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction; Summary of findings 2 TLS utilising embryo selection software compared to TLS with conventional morphological assessment of still TLS images for embryo incubation and assessment in assisted reproduction; Summary of findings 3 TLS utilising embryo selection software compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction
1. Time‐lapse system (TLS) with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1)
Three studies undertook this comparison (Kahraman 2013; Park 2015; Wu 2016), with a total of 489 participants.
Primary outcomes
1.1 Live birth
Two studies provided live birth data following correspondence with their authors (Kahraman 2013; Park 2015; N = 440). Sixty‐eight live births were reported in the TLS arm from the 278 women randomised to that arm. There were 54 live births from the 162 women randomised to the control arm (conventional incubation and embryo assessment).
There is probably no difference between the interventions in live birth rates (odds ratio (OR) 0.73, 95% confidence interval (CI) 0.47 to 1.13, 2 RCTs, N = 440, I2 = 11% , moderate‐quality evidence, Analysis 1.1; Figure 4). The evidence suggests that if the live birth rate associated with conventional incubation and assessment is 33%, the rate with use of TLS with conventional morphological assessment of still TLS images is between 19% and 36%.
Forest plot of comparison: 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), outcome: 1.1 Livebirth.
1.2 Miscarriage and stillbirth
Two studies provided both miscarriage and stillbirth data (Kahraman 2013; Park 2015; N = 440). The stillbirth data were available following communication with the authors of Park 2015.
There also may be no difference between the interventions in miscarriage rates. Out of 278 women randomised to the intervention arm, 19 women experienced a miscarriage; out of the 162 randomised to the control arm, 6 experienced miscarriage (OR 2.25, 95% CI 0.84 to 6.02, 2 RCTs, N = 440, I2 = 44%, low‐quality evidence, Analysis 1.2; Figure 5). The evidence suggests that if the miscarriage rate with conventional incubation is 3%, the rate associated with TLS with conventional morphological assessment of still TLS images would be between 3% and 18%.
Forest plot of comparison: 2 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment: miscarriage.
Regarding stillbirth, there were two stillbirths out of 38 women randomised to the intervention arm and two out of the 38 women randomised to the control arm in the study by Kahraman 2013. There were no stillbirths recorded in either arm of the study in Park 2015, meaning that its result is inestimable. In accordance with Cochrane methodological guidance, we have removed Park 2015 from meta‐analysis. Results from this solitary study (not meta‐analysis) suggest that there may be no difference between the interventions in rates of stillbirth (OR 1.00, 95% CI 0.13 to 7.49, 1 RCT, N = 76, low‐quality evidence, Analysis 1.3).
Secondary outcomes
1.3 Clinical pregnancy
All three studies provided clinical pregnancy data (Kahraman 2013; Park 2015; Wu 2016; N = 489). Of the 302 women randomised to the intervention arm there were 92 clinical pregnancies, and from the 187 women randomised to the control arm there were 66. The moderate‐quality evidence suggests that there is probably no difference between TLS with conventional morphological assessment of still TLS images and conventional incubation and assessment (OR 0.88, 95% CI 0.58 to 1.33, 3 RCTs, N = 489, I2 = 0%, moderate‐quality evidence Analysis 1.4).
2. TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2)
Two studies undertook this comparison (Goodman 2016; Kaser 2017), with a total of 463 participants. It is worth noting that in Kaser 2017, there were two intervention groups; one involved day three embryo transfer, and the other involved day five embryo transfer. The two intervention groups are represented as separate entities at meta‐analysis and the single control group has been split to share between the two intervention groups in order to avoid artificially doubling the effect of the control group.
Primary outcomes
2.1 Live birth
Live birth data were not collected by either study. This was confirmed on correspondence with the authors of both studies.
2.2 Miscarriage and stillbirth
Stillbirth data were not collected by either study.
We obtained miscarriage data for all women randomised following correspondence with the authors of both studies. For Goodman 2016, the miscarriage data include a combination of biochemical and clinical pregnancy losses. Unfortunately these data could not be separated for this review. For Kaser 2017, the data include miscarriages from clinical pregnancy losses.
There were 18 miscarriages out of 260 women randomised to the intervention arm, and 11 out of 203 women randomised to the control arm. We are uncertain whether TLS utilising embryo selection software influences miscarriage rates (OR 1.39, 95% CI 0.64 to 3.01, 2 RCTs, N = 463, I2 = 0%, very low‐quality evidence, Analysis 2.1). The evidence suggests that if the miscarriage rate associated with assessment of still TLS images is 5%, the rate with embryo selection software would be between 3% and 14%.
Secondary outcomes
2.3 Clinical pregnancy
Both studies reported this outcome. There were 132 clinical pregnancies from the 260 women randomised to the intervention group and 109 pregnancies from the 203 women randomised to the control group. There may be no difference between the interventions in clinical pregnancy rates (OR 0.97, 95% CI 0.67 to 1.42, 2 RCTs, N = 463, I2 = 0%, low‐quality evidence, Analysis 2.2).
3. TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3)
Three studies undertook this comparison (Kovacs 2013; Rubio 2014; Yang 2017), with a total of 1351 participants.
Primary outcomes
3.1 Live birth
Live birth data were available for two studies (Kovacs 2013; Rubio 2014). For Kovacs 2013, live birth data were provided following a request via correspondence. For Rubio 2014, we obtained data from a recently published paper and a published conference abstract (the references for these are provided as sub‐references under Rubio 2014).
There were 250 live births from the 524 women randomised to the intervention arm, and 188 live births from the 493 women randomised to the control arm. There is very low‐quality evidence that TLS utilising embryo selection software may improve live birth rates compared to conventional incubation and assessment (OR 1.21, 95% CI 0.96 to 1.56, 2 RCTs, N = 1017, I2 = 0%, Analysis 3.1Figure 6). The evidence suggests that if the live birth rate associated with no TLS is 38%, the rate with use of conventional incubation would be between 44% and 73%.
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.1 Livebirth.
3.2 Miscarriage and stillbirth
Stillbirth data were not collected by any study.
Miscarriage data are losses of clinical pregnancies in two studies (Kovacs 2013; Yang 2017). The other study has a combination of biochemical and clinical pregnancy losses (Rubio 2014).
There were 50 miscarriages from 691 women randomised to the intervention arm, and 62 miscarriages from 660 women randomised to the control arm. We are uncertain whether TLS utilising embryo selection software influences miscarriage rates (OR 0.73, 95% CI 0.49 to 1.08, 3 RCTs, N = 1351, I2 = 0%, very low‐quality evidence, Analysis 3.2; Figure 7). The evidence suggests that if miscarriage rate with conventional incubation is 9%, the rate associated with TLS would be between 4% and 10%.
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.2 Miscarriage.
Secondary outcomes
3.3 Clinical pregnancy
All three studies reported this outcome. There were 404 clinical pregnancies from the 691 women randomised to the intervention arm, and 360 pregnancies from the 660 women randomised to the control arm. We are uncertain whether TLS utilising embryo selection software influences clinical pregnancy rates (OR 1.17, 95% CI 0.94 to 1.45, 3 RCTs, N = 1351, I2 = 42%, very low‐quality evidence, Analysis 3.3Figure 8).
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.3 Clinical pregnancy.
It is worth noting that one study, Yang 2017, is very different in design to the other two included studies owing to the fact that it has differing days of embryo transfer in the intervention and the control arm of the study. When we removed this study in a sensitivity analysis, the pooled effect changed, revealing very low‐quality evidence of an improvement in clinical pregnancy rates (OR 1.30, 95% CI 1.02 to 1.67, 2 RCTs, N = 1017, I2=0%, very low‐quality evidence).
Subgroup and sensitivity analysis
We did not perform any planned subgroup or sensitivity analyses as there were insufficient number of included studies within the meta‐analyses.
Discussion
Summary of main results
Trial design 1
The comparison 'Time‐lapse system (TLS) with conventional morphological assessment of still TLS images versus conventional incubation and assessment', aims to assess the potential advantages of a stable incubator environment. The embryo selection software is not utilised and the embryos are left undisturbed until transfer. The three relevant studies included participants with a variety of infertility diagnoses. One described its participants as 'poor prognosis', with no further details (Wu 2016). Another described women with 'tubo‐peritoneal factor' (Kahraman 2013), and the third described over 99% male‐factor infertility, with 20% female‐factor in both arms (Park 2015). This variety adds to the broad applicability of results to common clinical practice. Two studies undertook embryo transfer at day two or three (Park 2015; Wu 2016), whereas the third study undertook blastocyst transfer (Kahraman 2013). All oocytes were autologous.
There is moderate‐quality evidence that there is probably no difference between the interventions in live birth rate or clinical pregnancy rates. There is also low‐quality evidence of no difference between the interventions in the rates of miscarriage or stillbirth per couple randomly assigned.
Trial design 2
The comparison 'TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images' aims to assess the potential advantages of the embryo selection software over conventional morphology. In this comparison, both arms of the study are housed in a TLS, but the embryo selection software is only utilised in one arm. Therefore the incubator environment is identical in both arms. Two studies were eligible for this comparison. One had two intervention arms; embryo transfer on day three, and embryo transfer on day five (Kaser 2017). The control arm had embryo transfer on day five only. The other study, Goodman 2016, undertook a combination of embryo transfer on day three or five. It is worth noting that the embryos were left undisturbed in Goodman 2016, however in Kaser 2017, the embryos in both the intervention arms and the control arms underwent daily conventional morphological assessment in addition to the application of embryo selection software in the intervention arms. There was a broad variety of infertility diagnoses in both studies, which adds to the overall applicability of results to broad clinical practice in reality.
We are uncertain whether there is a difference between the interventions in terms of miscarriage rates as the evidence is very low‐quality. There is low‐quality evidence suggesting that there may be no difference in clinical pregnancy rates. No evidence for live birth or stillbirth was available.
Trial design 3
The comparison 'TLS utilising embryo selection software versus conventional incubation and assessment' aims to assess the potential advantages of a combination of the stable incubator environmentand the embryo selection software versus conventional incubation and assessment. Three studies undertook this comparison. One of these utilised a combination of autologous and donor oocytes, the proportion of each are unknown (Rubio 2014). The remaining two studies used autologous oocytes. One study undertook embryo transfer on day three in the intervention group and day five in the control group (Yang 2017). Another study undertook transfer on day five (Kovacs 2013). In Rubio 2014, there was a combination of transfer on day three and day five. A variety of infertility diagnoses were recorded in the women in these studies. Two studies described their participants as 'good prognosis' (Rubio 2014; Yang 2017).
The meta‐analysis revealed very low‐quality evidence of an improvement in live birth rate when TLS utilising embryo selection software was used versus conventional incubation and assessment, however it is important to reflect on reasons to be cautious when accepting this result. Firstly, one study, Yang 2017, is an incomplete study and has not yet contributed live birth rates to the analysis. It is highly likely that the live birth result will change when these data are added, given that they observed lower clinical pregnancy rates and higher miscarriage rates in the intervention arm. Secondly, it is worth reflecting on the inconsistency of results across the three trial designs. The pooled estimates from trial design 1 report lower success with the incubator aspect of TLS; likewise trial design 2 report lower success with the cell tracking software element of TLS; however trial design 3 reports higher success when using both. From a scientific point of view, it is difficult to reason why there should be this discrepancy in findings, especially between trial design 2 and 3.
We are uncertain whether there is a difference between TLS utilising embryo selection software versus conventional incubation and assessment in miscarriage rates or clinical pregnancy rates, as the evidence is very low‐quality. Stillbirth was not examined by these studies.
Overall completeness and applicability of evidence
This updated systematic review on time‐lapse systems now includes eight RCTs, and includes additional data from two studies included in the first iteration of the review (Kovacs 2013; Rubio 2014). Data from 2303 women has gone towards formulating the findings of this review, however there are some comparisons which are better informed than others.
For example, approximately 59% of participants were included in trials that assessed TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3). This is mainly due to the largest trial undertaking this comparison (Rubio 2014). Trial designs 1 and 2 (TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment, and TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images) include the remaining 21% and 20% of participants respectively, but there are no women available to inform live birth findings in trial design 2, meaning there are profound gaps in evidence for TLS in this comparison. In addition, there were no stillbirth data for trial designs 2 and 3. This may be because stillbirth is so rare that it is not considered to be an important outcome, but it is important that future trials report this outcome, as it is a measure of safety.
Trial designs 1 and 2 included 489 and 463 women respectively. This is in comparison to the 1351 women who were included in trial design 3. Despite the additional information from previous and newly incorporated trials, the results of this review remain unclear. Further trials of each design are required to bolster participant numbers and to interrogate the robustness of the finding of an improvement of live birth with TLS versus control in trial design 3. The largest trial that informs trial design 3 has a number of biases, arising from the non‐randomised approach for some participants, the subsequent lack of blinding, the use of donor oocytes in a number of women, and the routine use of multiple embryo transfer.
There was heterogeneity between trials in the diagnosis of infertility, the day of embryo transfer, the use of IVF or ICSI and the make and model of TLS. All of these factors help to make the results of this review more applicable to clinical practice in the real world, where there is naturally this variation in clinical practices.
All studies excluded women who underwent frozen embryo transfer, except Kahraman 2013, whose investigators were able to provide data for these women. For Rubio 2014, the investigators were unable to provide data specifically for women who underwent donor oocyte IVF/ICSI. Therefore, in order to subgroup autologous, donor and frozen oocytes, future studies will need to present their results under these subgroups and state explicitly how many couples underwent these interventions.
Elective single embryo transfer was undertaken in most studies (Kahraman 2013; Kaser 2017; Kovacs 2013; Park 2015; Yang 2017), however two studies undertook multiple embryo transfers (Goodman 2016; Rubio 2014;). We were unable to obtain figures from the authors of Rubio 2014, on exactly what proportion of couples received multiple embryo transfer in each arm of the study. Given that this study contributed a large proportion of the data in trial design 3, it is important to recognise that the results presented here may reflect rates of clinical outcomes in keeping with multiple embryo transfer as opposed to single embryo transfer. One study did not disclose the number of embryos transferred per woman (Wu 2016).
Quality of the evidence
The quality of the evidence ranged from very low to moderate. The main limitations were risk of bias, imprecision and indirectness. Risk of bias was commonly associated with lack of blinding of participants or those involved in the study, attrition rates following randomisation, reporting of interim results and variation in number and day of embryos transferred between arms of the study.
The quality of the evidence for trial design 1 (TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment) is low to moderate. The evidence for live birth was downgraded in GRADE for imprecision owing to there only being 2 trials, totaling 440 women. The evidence for miscarriage and stillbirth was downgraded twice each for imprecision secondary to broad confidence intervals and a small number of events. The evidence for clinical pregnancy was downgraded once for risk of bias owing to unclear selection and performance bias (summary of findings Table for the main comparison).
The quality of evidence for trial design 2 (TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images) is very low to low. The quality of evidence for miscarriage was very low, and was downgraded once for risk of bias, once for indirectness and once for imprecision. This was owing to varying days and numbers of embryos transferred, decided upon by an unblinded embryologist, and secondary to heterogeneity between the study designs. One included study involved removing embryos for bench‐top microscopy daily in both the intervention and control arms, whereas the other left embryos in the intervention and control arms undisturbed. Also, there were broad confidence intervals of the two included studies which indicates imprecision. The quality of evidence for clinical pregnancy was low and was downgraded in GRADE once for the heterogeneity in study designs mentioned above (summary of findings Table 2).
The quality of the evidence for trial design 3 (TLS utilising embryo selection software versus conventional incubation and assessment) is of very low quality. Live birth was downgraded in GRADE twice for risk of bias and once for indirectness. All included studies are at high risk of performance bias owing to lack of blinding or incomplete blinding. There was also high risk of selection bias. In one study, the randomisation of participants was undertaken by the principal investigator and allocation concealment was not described. In another study, some patients could request the intervention and this request was granted. The indirectness was due to one included study undertaking multiple embryo transfers per woman, and included women receiving donor oocytes from younger women. Likewise, miscarriage was downgraded in GRADE twice for risk of bias as mentioned above and once for indirectness secondary to one included study including miscarriages of biochemical as well as clinical pregnancies. These miscarriage data could not be separated by the authors of the study. Finally, clinical pregnancy was downgraded twice for risk of bias and once for indirectness secondary to the day of embryo transfer being variable between studies. One study had blastocyst transfers, one had varied days of transfer and one had day‐three transfer for the intervention arm and day‐five transfer for the control arm. (summary of findings Table 3).
It should be noted that despite all studies being at high risk of performance bias owing to the lack of blinding of embryologists, we have not downgraded any studies for this unless other aspects of performance bias were lacking, for example, if participants were unblinded, or if the day or number of embryos transferred was decided by the unblinded embryologist.
Potential biases in the review process
We aimed to identify all eligible studies for inclusion in this review, and we contacted the included study authors on many occasions to seek as much information for inclusion as possible. Authors of most studies have been forthcoming with further study information, which has helped us to accrue a full picture of the study outcomes, as well as providing information needed to assess and establish the risk of bias.
Agreements and disagreements with other studies or reviews
To date, there are four published systematic reviews which have included the same inclusion and exclusion criteria on the topic of TLS versus conventional incubation (Chen 2017; Polanski 2014; Kaser 2014; Pribenszky 2017). Two of these are now out of date and new studies have been published since then (Polanski 2014; Kaser 2014). Both reviews report no evidence of a difference between TLS and control.
One systematic review, Kaser 2014, included 13 eligible studies after systematic searching, however none of the studies were RCTs and the majority were retrospective cohort studies. This review concludes that there is currently limited evidence to support the routine clinical use of TLS for selection of human pre‐implantation embryos.
Six eligible studies were included in Chen 2017, but it missed out two further eligible RCTs that are included in this review. It does not include all the potential live birth data, including data from Kahraman 2013; Kovacs 2013; Park 2015. It concludes that there is currently "insufficient evidence to support that time‐lapse imaging is superior to conventional methods for embryo incubation and selection".
In Pribenszky 2017, the authors undertook a systematic review of TLS utilising TLS embryo selection software. They concluded that TLS using embryo selection software was associated with a significantly higher ongoing pregnancy rate, a significantly lower early pregnancy loss and a significantly higher live birth rate in comparison to control. However, we have detected a number of problems with this review which have been published as a letter (Armstrong 2018). The issues outlined are as follows.
-
They have combined trials with different intervention and control arms. For example, three of the five included trials are study design 3, but one is study design 1 and one is study design 2.
-
They have also included a trial that describes itself as a prospective cohort study, not an RCT. On closer investigation, this trial is pseudo‐randomised (randomisation based on patient record number). This is not considered methodologically sound for systematic reviews of RCTs.
-
The authors describe applying an intention‐to‐treat analysis (which is considered the gold standard in fertility research), however the early pregnancy loss, live birth and stillbirth data are analysed per woman that became pregnant. This is known to skew the results toward showing a larger intervention effect.
-
It appears that full data from the included trials have not been entered into the review. For example, live birth data is not included from Rubio 2014, despite being published as an abstract in 2015.
-
We note that all three authors declared in this review that they work for Vitrolife, a biotechnology company that manufactures and promotes TLS.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Forest plot of comparison: 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), outcome: 1.1 Livebirth.
Forest plot of comparison: 2 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment: miscarriage.
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.1 Livebirth.
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.2 Miscarriage.
Forest plot of comparison: 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), outcome: 3.3 Clinical pregnancy.
Comparison 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), Outcome 1 Livebirth.
Comparison 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), Outcome 2 Miscarriage.
Comparison 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), Outcome 3 Stillbirth.
Comparison 1 TLS with conventional morphological assessment of still TLS images versus conventional incubation and assessment (trial design 1), Outcome 4 Clinical pregnancy.
Comparison 2 TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2), Outcome 1 Miscarriage.
Comparison 2 TLS utilising embryo selection software versus TLS with conventional morphological assessment of still TLS images (trial design 2), Outcome 2 Clinical pregnancy.
Comparison 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), Outcome 1 Livebirth.
Comparison 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), Outcome 2 Miscarriage.
Comparison 3 TLS utilising embryo selection software versus conventional incubation and assessment (trial design 3), Outcome 3 Clinical pregnancy.
| TLS with conventional morphological assessment of still TLS images compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing assisted reproductive technology | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with conventional incubation and assessment | Risk with TLS with conventional morphological assessment of still TLS images | |||||
| Live birth | 333 per 1,000 | 267 per 1,000 | OR 0.73 | 440 | ⊕⊕⊕⊝ | |
| Miscarriage | 37 per 1,000 | 83 per 1,000 | OR 2.25 | 440 | ⊕⊕⊝⊝ | |
| Stillbirth | 53 per 1,000 | 53 per 1,000 | OR 1.00 | 76 | ⊕⊕⊝⊝ | |
| Clinical pregnancy | 353 per 1,000 | 310 per 1,000 | OR 0.88 | 489 | ⊕⊕⊕⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the evidence for live birth once for imprecision due to there only being two trials, one in good prognosis patients and the other in poorer prognosis patients, totalling 440 women. | ||||||
| TLS utilising embryo selection software compared to TLS with conventional morphological assessment of still TLS images for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing assisted reproductive technology | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with TLS with conventional morphological assessment of still TLS images (trial design 2) | Risk with TLS utilizing embryo selection software | |||||
| Live birth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Miscarriage | 54 per 1,000 | 74 per 1,000 | OR 1.39 | 463 | ⊕⊝⊝⊝ | |
| Stillbirth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Clinical pregnancy | 537 per 1,000 | 529 per 1,000 | OR 0.97 | 463 | ⊕⊕⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the quality of evidence for miscarriage three times: once for risk of bias, once for indirectness and once for imprecision. The risk of bias is secondary to performance bias due to varying days and numbers of embryos transferred, decided upon by an unblinded embryologist. There is heterogeneity between the study designs leading to indirectness; one included study involved removing embryos for bench‐top microscopy daily in both the intervention and control arms, whereas the other left embryos in the intervention and control arms undisturbed. The imprecision is secondary to broad confidence intervals. | ||||||
| TLS utilising embryo selection software compared to conventional incubation and assessment for embryo incubation and assessment in assisted reproduction | ||||||
| Patient or population: couples undergoing ART | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | No of participants | Certainty of the evidence | Comments | |
| Risk with conventional incubation and assessment | Risk with TLS utilising embryo selection software | |||||
| Live birth | 381 per 1,000 | 461 per 1,000 | OR 1.21 | 1017 | ⊕⊝⊝⊝ | |
| Miscarriage | 94 per 1,000 | 70 per 1,000 | OR 0.73 | 1351 | ⊕⊝⊝⊝ | |
| Stillbirth | 0 per 1000 | 0 per 1000 | not estimable | 0 RCTs | ||
| Clinical pregnancy | 545 per 1,000 | 584 per 1,000 | OR 1.17 | 1351 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the mean risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| a We downgraded our assessment of the quality of evidence for live birth twice for risk of bias and once for indirectness. All included studies are at high risk of performance bias owing to lack of blinding or incomplete blinding. There was also high risk of selection bias. In one study, the randomisation of participants was undertaken by the principal investigator and allocation concealment was not described. In another study, some patients could request the intervention and this request was granted. The indirectness was due to one included study undertaking multiple embryo transfers per woman, and included women receiving donor oocytes from younger women. | ||||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Livebirth Show forest plot | 2 | 440 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.73 [0.47, 1.13] |
| 2 Miscarriage Show forest plot | 2 | 440 | Odds Ratio (M‐H, Fixed, 95% CI) | 2.25 [0.84, 6.02] |
| 3 Stillbirth Show forest plot | 2 | 440 | Odds Ratio (M‐H, Fixed, 95% CI) | 1.0 [0.13, 7.49] |
| 4 Clinical pregnancy Show forest plot | 3 | 489 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.88 [0.58, 1.33] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Miscarriage Show forest plot | 2 | 463 | Odds Ratio (M‐H, Fixed, 95% CI) | 1.39 [0.64, 3.01] |
| 2 Clinical pregnancy Show forest plot | 2 | 463 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.97 [0.67, 1.42] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Livebirth Show forest plot | 2 | 1220 | Odds Ratio (M‐H, Fixed, 95% CI) | 1.21 [0.96, 1.54] |
| 2 Miscarriage Show forest plot | 3 | 1351 | Odds Ratio (M‐H, Fixed, 95% CI) | 0.73 [0.49, 1.08] |
| 3 Clinical pregnancy Show forest plot | 3 | 1351 | Odds Ratio (M‐H, Fixed, 95% CI) | 1.17 [0.94, 1.45] |
