Birthweight is a basic characteristic that defines an individual; the weight and sex of an infant are key themes in discussion following a birth. Similarly, when considering pregnancy outcome and its associations with the subsequent health of the infant, birthweight is critical. Much of the focus on birthweight is on infants who are SGA because of the association of being SGA with perinatal mortality. The diagnostic effectiveness of ultrasound in this context was the subject of a Cochrane review of diagnostic effectiveness,23 and this is discussed extensively in Chapter 9. However, being born LGA is also a predictor of adverse outcomes, including perinatal mortality and morbidity arising from traumatic delivery, which is the focus of this chapter.
Ultrasonic EFW was first described > 40 years ago.100 The most widely employed equation for EFW was published by Hadlock et al.5 in 1985, and a reference range for EFW was published in 1991.6 A subsequent multicountry study by the World Health Organization7 derived very similar EFW percentiles, as described by Hadlock in Houston, Texas, USA, in the early 1990s. Hence, the diagnostic tools have been available for many years to identify SGA and LGA fetuses. Moreover, a RCT101 has indicated that routine IOL in the presence of suspected macrosomia may prevent shoulder dystocia, which is one of the key adverse outcomes associated with an infant being LGA.
Despite the widely available diagnostic tools, it is still not clear whether or not screening and intervention for suspected fetal macrosomia is clinically effective. The Health Technology Assessment (HTA) programme is currently funding a RCT of intervention in women diagnosed with a LGA infant (‘Induction of labour for predicted macrosomia: the Big Baby trial’; ISRCTN18229892). However, as universal ultrasound in late pregnancy is not recommended in the UK, these women will have received a clinically indicated scan. Hence the results of the study may not be applicable to low-risk women, because the diagnostic effectiveness of the test will vary between women who are scanned routinely and those scanned for a clinical indication. Hence, the aim of the present study was to quantify the diagnostic effectiveness of universal ultrasound in late pregnancy in predicting delivery of a large infant and one of the major associated complications, namely shoulder dystocia.
Methods
Sources of meta-analysis
A systematic search was performed in MEDLINE, EMBASE, CDSR and CENTRAL. The search was carried out on 22 October 2018. No restrictions on language or geographical location were applied. The protocol for the review was designed a priori and registered with the International Prospective Register of Systematic Reviews PROSPERO (registration number CRD42017064093). The studies were identified using a combination of words related to ‘ultrasound’, ‘pregnancy’, ‘estimated fetal weight’, ‘EFW’, ‘birthweight’, ‘macrosomia’, ‘large for gestational age’, ‘shoulder dystocia’ and ‘brachial plexus injury’.
Study selection
Selection criteria allowed the inclusion of cohort or cross-sectional studies involving singleton pregnancies in which an ultrasound scan was performed at ≥ 24 weeks’ gestation. We included all studies in which the ultrasound was performed as part of universal screening, studies that used low-risk populations only and studies with mixed-risk populations. We excluded studies that were focused on high-risk patients, such as patients with pre-existing or gestational diabetes, and studies in which the ultrasound was performed intrapartum. We included studies regardless of the formula and threshold they used to define macrosomia. We also included studies regardless of whether the result was blinded to clinicians. We included studies that reported the following outcomes: LGA (defined as birthweight > 4000 g or > 90th centile) and severe LGA (birthweight > 4500 g or > 97th centile); shoulder dystocia; and adverse neonatal outcomes, such as neonatal unit admission, 5-minute Apgar score of < 7 and neonatal metabolic acidosis.
Study quality assessment and data extraction
The literature search, study selection and analysis ware performed independently by two authors (AM and NS) using Review Manager 5.3. Any differences were resolved in discussion with the senior author (GS). The risk of bias in each included study was assessed using the QUADAS-2 tool as outlined in the Cochrane Handbook of Diagnostic Test Accuracy Studies.37 This tool assesses the included studies for potential bias in four domains: patient selection, index test, reference standard, and flow and timing. We assessed the risk for flow and timing from the perspective of universal ultrasound screening at about 36 weeks’ gestation. We used a predesigned data extraction form to extract information on study characteristics (i.e. year of publication, country, setting, study design and blinding), patient characteristics (i.e. inclusion and exclusion criteria, and sample size), the index test (i.e. gestational age at scan, formula and cut-off values used) and reference standard (i.e. pregnancy outcome, gestational age at delivery and interval from scan to delivery). We also collected information, such as inclusion or exclusion of patients with pre-existing or gestational diabetes.
Statistical and meta-analysis methods
The statistical and meta-analysis methods employed are described in Chapter 4.
Results
The literature search flow chart is presented in Appendix 5, . We identified 40 studies102–141 that met our inclusion criteria, which involved a total of 66,187 patients. The study characteristics are presented in Appendix 5, Table 24. Five studies105,114,120,123,138 (n = 8088) included unselected pregnancies, nine studies110,116,118,119,122,129,131,139,140 (n = 6436) included only low-risk pregnancies and 26 studies102–104,106–109,111–113,115,117,121,124–128,130,132–137,141 (n = 51,663) included mixed-risk pregnancies.
The assessment of study quality was performed using the QUADAS-2 tool and is summarised in Appendix 5, . The main risk of bias was for reference standard because only two studies116,138 blinded the results to the clinicians. The second most common risk of bias was for flow and timing. This is because six studies106,111,123,125,133,142 had a very short interval between ultrasound and delivery (the ultrasound was carried out either prior to IOL or < 72 hours from delivery), two studies105,114 had a long interval (the ultrasound was carried out prior to 33 weeks’ gestation) and two studies104,107 did not specify the gestational age at delivery. Finally, three studies110,134,140 included prolonged (> 41 weeks’ gestation) pregnancies only, which were classified as having ‘high applicability concerns because of patient selection’.37
The most commonly reported outcomes were birthweight of > 4000 g (29 studies103–106,110–113,118–123,125–135,138–141) followed by birthweight > 90th centile (seven studies102,107,109,114,115,124,138), both of which we classified as LGA. We defined severe LGA as a birthweight of > 4500 g (six studies113,117,131,137,138,141) or > 95th or 97th centiles (two studies114,138). Shoulder dystocia was reported in six studies.108,112,116,136,138,141 Finally, neonatal morbidity (any related outcomes) was reported in only two studies,112,138 and consequently we could not produce summary results for this outcome. The most commonly used formulas for EFW were those described by Hadlock et al.,5 followed by Shepard et al.143 The most common thresholds for suspected LGA on scan were 4000 g (21 studies103,104,106,108,110,118,119,121,125,126,128–135,139–141) and 90th centile for the gestational age (nine studies). The abdominal circumference was used in nine studies,102,105,107,109,111,112,114,115,138 with the most common threshold applied being 36 cm (five studies122,123,125,127,137).
We present the summary diagnostic performance in . An estimated EFW of > 4000 g or > 90th centile had > 50% sensitivity for predicting LGA at birth and this was similar regardless of the formula used. The positive LR ranged between 7.5 and 12 for the Hadlock formulas5,6 and was about 5 for the Shepard formula.143 The abdominal circumference (AC) had similar performance with the EFW. Suspected LGA also had about 70% sensitivity at predicting severe LGA at birth. Finally, an EFW of > 4000 g or 90th centile had 22% sensitivity at predicting shoulder dystocia with a statistically significant positive LR of 2.1.
Summary diagnostic performance of suspected LGA in predicting LGA at birth and shoulder dystocia
The summary ROC curves for LGA and shoulder dystocia are presented in . We also present the pooling of the DORs (). Finally, we used Deeks’ funnel plot asymmetry test to assess the risk of publication bias using the outcome of LGA for the analysis (see Appendix 5, ). The test showed potentially significant risk of publication bias (p = 0.02).
Summary ROC curves for the diagnostic performance of EFW > 4000 g (or 90th centile) at predicting (a) LGA at birth (birthweight > 4000 g or > 90th centile); and (b) shoulder dystocia.
The diagnostic odds ratios for the diagnostic performance of EFW > 4000 g (or > 90th centile) at predicting (a) LGA at birth (birthweight > 4000 g or > 90th centile); and (more...)
Discussion
The key findings of the present study are that suspicion of fetal macrosomia on ultrasound scan is strongly predictive of the risk of delivering a large infant, but it is only weakly, albeit statistically significantly, predictive of the risk of shoulder dystocia. In the case of delivering a LGA infant as defined by the Hadlock formula, the positive LRs were quite strong, in the region of 7–12, whereas in relation to the diagnosis of shoulder dystocia the positive LR was ≈ 2. The forest plot of DORs indicates significant heterogeneity between the studies in their ability to predict a LGA infant. The source of this heterogeneity is unclear but it could relate to differences in the quality of the performance of the diagnostic test, such as the quality of the imaging equipment, the skill and training of the sonographers, and the characteristics of the population.
In this chapter and in Chapters 4 and 7 we have focused analysis on data from the POP study, as these data are particularly applicable to the research question addressed in this report, given that late-pregnancy ultrasound was performed in a large number of nulliparous women using contemporary equipment and staff trained using the standards of NHS England. The POP study analysis of a 36 weeks’ gestation scan in the diagnosis of macrosomia had previously been published138 and this was incorporated into the meta-analysis. Interestingly, the DOR from the POP study was 17.1 (95% CI 12.0 to 24.3) and this was virtually identical to the summary estimate from all of the other studies, which was also 17.1 but with a slightly narrower 95% CI (13.3 to 22.0). These data suggest that the results from the POP study are likely to be generalisable.
A recurrent theme in all chapters has been the lack of blinding in studies of the diagnostic effectiveness of ultrasound of pregnancy screening research. Hence, generally, the POP study has been unique as a contemporary study of late pregnancy in nulliparous women. However, in this analysis there is a second comparable study: the Genesis study. This was a prospective cohort study of 2772 nulliparous pregnant women recruited across seven centres in Ireland between 2012 and 2015. Women had the ultrasound scan between ≥ 39 weeks’ gestation and < 41 weeks’ gestation (i.e. ≈ 3–4 weeks later than in the POP study). Although the scan was carried out slightly later than stated in the research question of the current report, the design makes the study particularly useful.
The analysis of fetal macrosomia from the Genesis study has been published in abstract form only. It did not report the diagnostic effectiveness of EFW as a predictor of LGA birthweight, but it did report shoulder dystocia. Interestingly, the POP study and the Genesis study were the only two large studies (i.e. comprising > 1000 women) not to demonstrate a statistically significant association between macrosomic EFW and the risk of shoulder dystocia. Overall, the meta-analysis indicated that ultrasound may be weakly predictive of shoulder dystocia. However, as with other analyses in Chapters 4–7, these findings could be explained by ascertainment bias. Specifically, if a scan is performed and the fetus is suspected to be macrosomic, the clinical staff attending the birth may be more likely to institute manoeuvres for shoulder dystocia in the event of any delay, or to document a given delay as being due to shoulder dystocia. The potential for such biases may explain why the studies with blinded ultrasound were not significantly associated with shoulder dystocia and why the meta-analysis as a whole was only weakly predictive of shoulder dystocia, whereas it was strongly predictive for macrosomia. A weak association between ultrasonic EFW and the risk of shoulder dystocia is not surprising given that the actual birthweight of the infant is not strongly predictive of shoulder dystocia and that the majority of cases of shoulder dystocia do not involve a macrosomic infant.144
Finally, the relationship between fetal macrosomia and pregnancy outcome is an area where there is good evidence that revealing the scan result changes the experience of complications of women who are false positives. Multiple studies have demonstrated that a false-positive diagnosis of fetal macrosomia is an independent risk factor for emergency caesarean delivery.145–147 These observations underline the potential of screening low-risk women to cause harm and that designing a study where the results are revealed to the attending physician could lead to an association that is iatrogenic (because the knowledge of the result may change clinical decision-making) rather than because of a true prediction.
