FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W4328103745> ?p ?o ?g. }

• W4328103745 endingPage "912" @default.
• W4328103745 startingPage "910" @default.
• W4328103745 abstract "Randomized controlled trials (RCTs) may provide excellent evidence of whether infertility interventions improve (or worsen) patient outcomes but only if they are designed and analyzed in an appropriate fashion. Assisted reproductive technologies (ARTs) present distinctive opportunities for error in statistical analyses due to the multistage nature of the treatments involved (1Wilkinson J. Roberts S.A. Showell M. Brison D.R. Vail A. No common denominator: a review of outcome measures in IVF RCTs.Hum Reprod. 2016; 31: 2714-2722Crossref PubMed Scopus (40) Google Scholar, 2Wilkinson J. Stocking K. Study design flaws and statistical challenges in evaluating fertility treatments.Reprod Fertil. 2021; 2: C9-C21Crossref PubMed Scopus (6) Google Scholar). Study participants must typically pass through a series of milestones for successful treatment, including egg retrieval, successful fertilization and culture of embryos, embryo transfer, implantation, and pregnancy. One consequence of this is that analyses may be performed using a variety of different denominators, excluding participants who did not reach a certain milestone. Hence, live births may be calculated per cycle started, per egg collection, per transfer, or using some other choice of denominator. Unfortunately, some commonly used denominators do not represent statistically sound options, and as a result, many RCTs in the field are analyzed in a manner that discards the benefits of randomization altogether. We will motivate the discussion using a recent RCT that evaluated the effect of PGT-A on cumulative live births in women with ≥3 good-quality blastocysts (3Yan J. Qin Y. Zhao H. Sun Y. Gong F. Li R. et al.Live birth with or without preimplantation genetic testing for aneuploidy.N Engl J Med. 2021; 385: 2047-2058Crossref PubMed Scopus (85) Google Scholar). As is typical for large ART RCTs, the trial generated much discussion around its design features, including the limit on the number of embryos subjected to PGT-A and the decision not to transfer mosaic embryos (4Wong A.J. DeCherney A.H. Live birth with or without preimplantation genetic testing for aneuploidy.N Engl J Med. 2022; 386: 703Crossref PubMed Scopus (0) Google Scholar). These points are interesting but have no bearing on the present discussion. Instead, we consider the choice of denominator in the study. The investigators used cumulative live birth per woman randomized as the primary outcome, wherein randomization was conducted on day 5 of embryo culture (3Yan J. Qin Y. Zhao H. Sun Y. Gong F. Li R. et al.Live birth with or without preimplantation genetic testing for aneuploidy.N Engl J Med. 2021; 385: 2047-2058Crossref PubMed Scopus (85) Google Scholar). In a subsequent letter to the editor, it was suggested that “live birth per embryo transfer would have been a more clinically relevant metric” (4Wong A.J. DeCherney A.H. Live birth with or without preimplantation genetic testing for aneuploidy.N Engl J Med. 2022; 386: 703Crossref PubMed Scopus (0) Google Scholar). We suspect that this view is commonly held because this was identified as the most common denominator used for the calculation of live birth rates in a review of ART RCTs (1Wilkinson J. Roberts S.A. Showell M. Brison D.R. Vail A. No common denominator: a review of outcome measures in IVF RCTs.Hum Reprod. 2016; 31: 2714-2722Crossref PubMed Scopus (40) Google Scholar). The premise of PGT-A is that it can reduce miscarriages by preventing the transfer of nonviable embryos. Of course, we could eliminate miscarriages altogether by performing no transfers at all; however, it is difficult to imagine that this would be a popular approach among patients. Hence, for any reduction in miscarriages to be beneficial, it must not come at the expense of substantially compromising the live birth rate, as would happen if PGT-A frequently returned an incorrect test result, preventing the transfer of viable embryos. For some patients, a reduction in live birth rate may be seen as an acceptable trade-off to avoid the experience of miscarriage. This decision can only be made if the trade-off is well understood, however, meaning that the impact on both miscarriage and live birth needs to be considered in any evaluation of the intervention. To evaluate the potential benefit of PGT-A, we must, therefore, answer the following question: does PGT-A reduce miscarriages, and what is the impact on live birth rates? Let us consider the implications of this for the choice of denominator to be used in the analysis. Figure 1 (adapted from the review by Wilkinson and Stocking (2Wilkinson J. Stocking K. Study design flaws and statistical challenges in evaluating fertility treatments.Reprod Fertil. 2021; 2: C9-C21Crossref PubMed Scopus (6) Google Scholar)) shows a schematic for a hypothetical RCT, similar in design to our motivating example (3Yan J. Qin Y. Zhao H. Sun Y. Gong F. Li R. et al.Live birth with or without preimplantation genetic testing for aneuploidy.N Engl J Med. 2021; 385: 2047-2058Crossref PubMed Scopus (85) Google Scholar). For illustration, we considered only the outcome of the first embryo transfer. Participants were randomly allocated to either the PGT-A or control arm on day 5 of embryo transfer. Ignoring the categorization of the participants according to prognosis to begin with, it is apparent that more participants proceeded to transfer without PGT-A than those who proceeded to transfer with PGT-A because in the PGT-A arm, testing led to the conclusion for some patients that there were no transferrable embryos. Indeed, in the schematic, we can see that more women in the control arm did experience miscarriage as a result. However, the implication in the schematic is that this reduction in miscarriages came at a cost. It appears that some of the women who did not proceed to transfer to the PGT-A arm would have had a successful outcome if they had, in fact, done so. Accordingly, we see more live births in the control arm than in the PGT-A arm, both numerically and as a proportion of all the women randomized to each respective group. Let us return to the question posed in the previous section: does PGT-A reduce miscarriages, and what is the impact on live birth rates? By assessing miscarriages and live births per woman randomized, we get an answer to this question. There were fewer miscarriages per woman randomized in the PGT-A group than in the control group (1/7 vs. 3/7, respectively). However, there were also fewer live births (2/7 vs. 3/7, respectively). By evaluating the impact on both the outcomes per woman randomized, patients can make an informed choice based on their own preferences. Is the reduction in live birth worthwhile? In contrast, if we consider only participants who underwent an embryo transfer, PGT-A does, indeed, appear to have a superior live birth rate (2/3 vs. 3/6). Note, however, that this analysis cannot answer our question about the potential trade-offs between miscarriage and live birth. This analysis, in fact, attempts to answer a hypothetical question: what would be the effect of PGT-A if it was always correct when it was found that there were no transferrable embryos? This is an interesting thought experiment; however, in practice, the possibility that PGT-A might sometimes be wrong is part of the effect that we are trying to assess. This poses a challenge to the suggestion that an analysis per embryo transfer would be “more clinically relevant” because this analysis refers to a purely hypothetical scenario, which real patients do not experience. More generally, this shows how careful consideration needs to be given to the implications of statistical analysis for the interpretation and clinical relevance of the result. A framework for explicitly aligning the statistical analysis with the specific research question in the study has recently been elaborated and included in the guidelines adopted by the European Medicines Agency and Food and Drug Administration (5Food and Drug Administration. E9 (R1) statistical principles for clinical trials: addendum: estimands and sensitivity analysis in clinical trials: guidance for industry. Available at: https://www.fda.gov/regulatory-information/search-fda-guidance-documents/e9r1-statistical-principles-clinical-trials-addendum-estimands-and-sensitivity-analysis-clinical. Accessed April 11, 2023.Google Scholar). A less obvious, but nonetheless critical, problem is the fact that the use of a postrandomization denominator, as happens when we evaluate outcomes “per embryo transfer,” nullifies the benefits conferred by randomization. Returning to Figure 1, we see that the participants are categorized according to their prognosis. Randomization ensures that any imbalances in baseline characteristics are random rather than systematic. It is worth noting that the purpose is not to ensure that the groups are identical in terms of their characteristics, as has been erroneously suggested by some commentators. Provided that imbalances are random, a direct comparison of outcomes between the groups allows for valid inference about the effect of an intervention using the standard statistical apparatus of P values and confidence intervals. This is why we randomize. In Figure 1, random allocation ensured that any imbalances in prognostic baseline characteristics between the PGT-A and no-PGT-A arms are not systematic, meaning that there is no confounding between the study arms. However, the benefit conferred by random allocation only holds when we base our analysis on the complete randomized cohort. If we restrict our analysis to the subgroup of women who underwent embryo transfer, these benefits are lost: there is no longer any expectation that the participants in the 2 study arms will not systematically differ in terms of their prognostic characteristics. In fact, given that the use of PGT-A influences the likelihood of embryo transfer, the characteristics of these participants are likely to differ between the 2 arms. In Figure 1, we see that many participants with worse prognoses did not proceed to embryo transfer in the PGT-A arm. Consequently, the comparison of outcomes in women who underwent embryo transfer ends up being highly imbalanced. The remaining participants in the PGT-A arm have generally better prognosis than those remaining in the no-PGT-A arm. Differences in live birth, therefore, reflect not only the effects of treatment (should there be any) but also differences in patient characteristics: there is confounding (or more accurately, selection bias). As such, this analysis transforms an RCT into a biased, observational study. It is unclear why we would go to the trouble of conducting an RCT at all if we are going to undermine the rationale for doing so at the analysis stage. An RCT requires valid statistical analysis to yield useful information. Here, we have described a common statistical error in the analysis of RCTs: the use of a postrandomization denominator for analysis using the analysis “per embryo transfer” as an illustration. We discussed the outcome of the first transfer, rather than the more relevant cumulative outcome over several transfer attempts, for simplicity of presentation. We stress that similar considerations would apply in relation to cumulative live birth: analysis per transfer would disguise any reduction in live births due to reduced transfers while introducing confounding between the study arms. Although we used the example of PGT-A here, this error is frequently found in infertility RCTs of various interventions (1Wilkinson J. Roberts S.A. Showell M. Brison D.R. Vail A. No common denominator: a review of outcome measures in IVF RCTs.Hum Reprod. 2016; 31: 2714-2722Crossref PubMed Scopus (40) Google Scholar). Using a postrandomization denominator for analysis fails to answer the question of practical relevance and undermines random allocation. This is statistics as reversed alchemy: it turns gold into a base metal." @default.
• W4328103745 created "2023-03-22" @default.
• W4328103745 creator A5091167688 @default.
• W4328103745 date "2023-06-01" @default.
• W4328103745 modified "2023-10-14" @default.
• W4328103745 title "Neither relevant nor randomized: the use of “per embryo transfer” in the analysis of preimplantation genetic testing for aneuploidy trials" @default.
• W4328103745 cites W2523330271 @default.
• W4328103745 cites W3164863188 @default.
• W4328103745 cites W4225509520 @default.
• W4328103745 cites W4226106794 @default.
• W4328103745 doi "https://doi.org/10.1016/j.fertnstert.2023.03.020" @default.
• W4328103745 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/36948442" @default.
• W4328103745 hasPublicationYear "2023" @default.
• W4328103745 type Work @default.
• W4328103745 citedByCount "2" @default.
• W4328103745 countsByYear W43281037452023 @default.
• W4328103745 crossrefType "journal-article" @default.
• W4328103745 hasAuthorship W4328103745A5091167688 @default.
• W4328103745 hasBestOaLocation W43281037451 @default.
• W4328103745 hasConcept C104317684 @default.
• W4328103745 hasConcept C160099875 @default.
• W4328103745 hasConcept C16685009 @default.
• W4328103745 hasConcept C196843134 @default.
• W4328103745 hasConcept C2779672484 @default.
• W4328103745 hasConcept C2780673598 @default.
• W4328103745 hasConcept C30481170 @default.
• W4328103745 hasConcept C54355233 @default.
• W4328103745 hasConcept C71924100 @default.
• W4328103745 hasConcept C86803240 @default.
• W4328103745 hasConceptScore W4328103745C104317684 @default.
• W4328103745 hasConceptScore W4328103745C160099875 @default.
• W4328103745 hasConceptScore W4328103745C16685009 @default.
• W4328103745 hasConceptScore W4328103745C196843134 @default.
• W4328103745 hasConceptScore W4328103745C2779672484 @default.
• W4328103745 hasConceptScore W4328103745C2780673598 @default.
• W4328103745 hasConceptScore W4328103745C30481170 @default.
• W4328103745 hasConceptScore W4328103745C54355233 @default.
• W4328103745 hasConceptScore W4328103745C71924100 @default.
• W4328103745 hasConceptScore W4328103745C86803240 @default.
• W4328103745 hasIssue "6" @default.
• W4328103745 hasLocation W43281037451 @default.
• W4328103745 hasLocation W43281037452 @default.
• W4328103745 hasOpenAccess W4328103745 @default.
• W4328103745 hasPrimaryLocation W43281037451 @default.
• W4328103745 hasRelatedWork W2183685009 @default.
• W4328103745 hasRelatedWork W2353446234 @default.
• W4328103745 hasRelatedWork W2354292814 @default.
• W4328103745 hasRelatedWork W2509202055 @default.
• W4328103745 hasRelatedWork W2513703644 @default.
• W4328103745 hasRelatedWork W2775574441 @default.
• W4328103745 hasRelatedWork W2922412025 @default.
• W4328103745 hasRelatedWork W318902238 @default.
• W4328103745 hasRelatedWork W4220678844 @default.
• W4328103745 hasRelatedWork W4319601417 @default.
• W4328103745 hasVolume "119" @default.
• W4328103745 isParatext "false" @default.
• W4328103745 isRetracted "false" @default.
• W4328103745 workType "article" @default.