The not‐so‐simple question of when or if to induce a term pregnancy

In this issue of Paediatric and Perinatal Epidemiology, Berman and colleagues1 revisited one of the most complex questions in perinatal epidemiology: what is optimal pregnancy length? To tackle the question, Berman et al.1 adopted a lifetable approach to examine perinatal mortality by gestational week among more than 300,000 term births between 2009 and 2019 in Western Australia. The authors computed the perinatal risk index, a summary measure of the probability of stillbirth and neonatal mortality at any given gestational week that is not commonly used in perinatal epidemiology. The authors found that the perinatal risk index was lowest at 39 weeks, regardless of Aboriginal status, parity, and obstetrical risk. They concluded that induction at 39 weeks is safe while cautioning that not all patients should be induced at this point. We found their recommendation prudent. This commentary aims to draw attention to the potential biases underlying research on optimal pregnancy length. Identifying the optimal gestational age to deliver an uncomplicated pregnancy is one of the most methodologically daunting tasks in obstetrics. The risk of perinatal mortality follows a U-shaped curve, peaking in the preterm period and then descending to a trough as the pregnancy approaches term gestations before increasing again. Foetal and neonatal mortality is elevated before 37 weeks as foetuses are immature, especially at very low gestational ages. While neonatal mortality decreases after 37 weeks, the reduction is offset by an increase in foetal mortality as pregnancy progresses. The risk of neurological morbidity is also U-shaped, decreasing early term before increasing again post-term.2 Balancing these risks and finding the optimal trough is no small feat. The perinatal risk index is one way to balance weekly mortality risks. The perinatal risk index is the product of antepartum stillbirth, intrapartum stillbirth, and neonatal mortality rates.3 The denominator for each component differs and includes ongoing pregnancies for antepartum stillbirth rates, deliveries for intrapartum stillbirth rates, and live births for neonatal mortality rates. While the perinatal risk index is easy to interpret for physicians and researchers, this indicator is limited by the type of data used for calculations. Limitations may be particularly pronounced when the data are observational. Patients are not randomised to deliver at a specific gestational week in observational data. Instead, other factors influence the delivery week, many of which are unmeasurable.2 As a result, patients included in the denominator of each week have unique characteristics that may be associated with mortality. Obesity and low educational attainment are a few examples, as patients with these characteristics are more likely to deliver early term than patients without these risk factors.4 Intensity of care is another factor that can contribute, as patients who deliver at 38 or 39 weeks may be monitored more closely to prevent perinatal death, especially if risk factors are present. While Berman et al. account for Aboriginal status, parity, and obstetric risk, unmeasured confounders associated with the gestational age of birth and the risk of perinatal mortality likely persist. The problem caused by a lack of randomisation is illustrated when stratifying patients on obstetric risk. Patients with low obstetric risk had higher perinatal mortality at 40 weeks than patients with high obstetric risk. The pattern flipped at 39 weeks, at which point patients with high obstetric risk had higher mortality than patients with low obstetric risk. The most likely explanation is that patients with high obstetric risk were selected to be induced at earlier gestational ages, leaving only healthier patients at lower risk of perinatal mortality in later groups. This nonrandom selection process occurs every gestational week among patients with high obstetric risk. Nonrandom selection also occurs among patients with low obstetric risk. The degree to which selection happens is unknown but may lead to paradoxical patterns in mortality rates. Berman et al. acknowledge this issue but could not resolve the problem as the perinatal risk index is not designed to account for unmeasured confounders or selection bias. Researchers have proposed alternative methods to reduce selection bias in observational data, such as foetuses-at-risk or counterfactual approaches. However, these methods have their own shortcomings. The foetuses-at-risk method relies on adapting the denominator of perinatal mortality rates to include ongoing pregnancies to minimise bias caused by gestational age stratification. Its application to postnatal outcomes has been debated, as events like neonatal death are not possible until foetuses are born.5 Counterfactual methods address unmeasured confounding by mimicking randomised control trials. However, these methods rely on conceptualising a hypothetical population representative of a randomised population, which is not always intuitive.6 Randomised trials are currently the gold standard to test the safety of induction at term, but even this approach has limitations.2, 7 The ARRIVE trial, for example, investigated perinatal outcomes following labour induction in low-risk nulliparous patients.8 Patients were randomised to induction at 39 weeks or expectant management. The trial found that induction at 39 weeks was not associated with an increased risk of neonatal mortality. Although patients were randomised, the trial was critiqued because induction inherently shortens the length of pregnancy in the intervention group.9 The longer length of pregnancies managed expectantly results in a mortal time bias where these pregnancies can disproportionately accumulate perinatal deaths. Although mortal time bias has the potential to explain some of the findings in the ARRIVE trial, the trial subsequently led to an increase in inductions at 39 weeks in the general population.9 It does not help that other studies, both observational and randomised, have found slightly different answers.2, 7 An analysis of Swedish data found that the risk of stillbirth or infant mortality was not lower for births at 39 weeks compared with 40 weeks or later.2 A Cochrane meta-analysis of randomised trials similarly found that induction before 40 weeks was not associated with lower perinatal mortality than induction at 40 or 41 weeks.7 Furthermore, studies on neurodevelopmental outcomes are limited, complex, and at times contradictory.2, 10 These findings cast doubt on the measurable benefit of induction at 39 weeks in the absence of clinical indications. In closing, Berman and colleagues tackled a difficult question that has no correct answer at this time. Their study is a reminder of the complex methodological issues surrounding the optimal timing of delivery. The results from such studies, including randomised trials, should be interpreted cautiously due to the possibility of bias. For now, we agree with Berman et al.'s conclusion that induction at 39 weeks is likely safe but may not benefit all patients. Nathalie Auger is a physician-epidemiologist at the University of Montreal Hospital Centre and a full clinical professor of epidemiology in the School of Public Health at the University of Montreal. Dr. Auger has nearly 20 years of experience using administrative health data for studies of maternal and child health. Dr. Auger serves on Paediatric and Perinatal Epidemiology's editorial board. Jessica Healy-Profitós is an epidemiologist at the University of Montreal Hospital Research Centre. Her work focuses on maternal-child health surveillance, pregnancy complications and their impact on long-term health, perinatal maternal mental health, and congenital anomalies. Ms. Healy-Profitós earned her master's in public health from The Ohio State University. N Auger and J Healy-Profitós contributed equally to the writing of this commentary. None. The authors declare no conflicts of interest. This work was funded by the Canadian Institutes of Health Research (grant number PJT-162300) and the Fonds de recherche du Québec-Santé (grant number 296785). Not applicable.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.