Point: Setting realistic expectations for the evaluation of intrauterine growth charts

The question of which growth chart should be used for identifying fetuses and newborns at risk because of suboptimal growth was propelled into the scientific arena by the INTERGROWTH 21st and World Health Organization (WHO) projects to develop intrauterine growth standards (i.e. charts developed in low-risk populations with normal growth) in the mid-2010s.1, 2 Previously, despite broad consensus in the obstetric community on the use of birthweight percentile cut-offs to screen for fetuses or newborns with growth anomalies (below 10th percentile for small for gestational age (SGA) and above 90th percentile for large for gestational age (LGA), with more extreme thresholds for severe cases: <3rd, <5th, >95th, >97th),3 most recommendations did not specify which charts should be used to determine these thresholds. The publication of these new international charts, in tandem with the Intergrowth project's stated aim to produce growth standards for global use, led to a plethora of research studies seeking to establish whether universal thresholds of SGA and LGA exist and to compare the performance of different charts. This research, which uses a variety of evaluation strategies and health outcomes to assess performance, largely refuted the universal standard hypothesis but also revealed the limits of growth charts for predicting health risks. The study by John et al.4 entitled ‘The clinical performance and population health impact of birthweight-for-gestational age indices at term gestation’ published in this issue of Paediatric and Perinatal Epidemiology adds to this ongoing debate about the choice and performance of intrauterine growth charts. Their study explores the ability of percentile thresholds from different intrauterine growth charts to identify term singleton liveborn infants with severe neonatal morbidity. Three different charts are compared: internal charts derived from their study population (singleton term live births in the United States from 2003 to 2017), the INTERGROWTH 21st newborn charts1 and the WHO estimated fetal weight charts.2 Their results corroborate previous work by showing that the percentages of SGA and LGA newborns differ depending on the chart. Furthermore, in models of the association between percentile values and risks of adverse neonatal outcomes, they find that all charts perform poorly at predicting severe neonatal morbidity at an individual and population level. However, in interpreting these results to conclude about the limits of growth charts, it is important to consider whether expectations are too high. A first expectation is that intrauterine growth charts could be useful for predicting risks of all newborn morbidity. This study's neonatal morbidity outcome (5-minute Apgar score <4, neonatal seizures, need for assisted ventilation, and neonatal death) was chosen because of its relevance in describing morbidity among term live births. However, neonatal morbidity occurs among normally grown infants and can have no relationship with their growth, especially morbidity resulting from acute events, such as uterine rupture or cord prolapse. Including morbidities resulting from other aetiological mechanisms in the outcome will reduce estimates of a chart's predictive value, as growth percentiles are of no relevance for predicting their occurrence. Having a single composite outcome is also problematic because of differences in the causes and consequences of abnormal growth at the extremes of the percentile distribution. For instance, thresholds selected to predict morbidity resulting from growth restriction (<10th or <3rd percentiles) will not be of use in predicting morbidity from labour dystocia caused by macrosomia, typically defined as a birthweight over 4000 or 4500 g. This suggests the need for specific outcomes when evaluating the predictive value of low and high thresholds of the percentile distribution. Selecting an outcome that is appropriate for evaluating the performance of growth charts in predicting morbidities associated with fetal growth abnormalities is complex as most neonatal morbidities can have many causes. One solution that we used for evaluating the predictive value of estimated fetal weight charts to detect morbidity associated with fetal growth was to associate the occurrence of neonatal morbidity with low or high birthweight percentiles defined using several thresholds (3rd and 10th percentiles for SGA births and 90th and 97th percentiles for LGA births).5 This definition assumes that morbidities associated with restrictive and excessive growth are more likely to occur among newborns with birthweight below and above these thresholds, respectively. However, it can be criticised because appropriate for gestational age newborns with growth abnormalities will be excluded. New proposed definitions of fetal growth restriction that add clinical or biological criteria (Doppler velocimetry findings, antenatal diagnoses of fetal growth restriction, and placental pathology) and other anthropometric measurements to birthweight percentiles to better distinguish growth restriction from constitutional size could help in refining appropriate outcomes for prediction studies.6-8 A second expectation is that observational studies of birthweight can provide robust estimates of a chart's performance. Growth monitoring is a component of antenatal care in all high-income countries and leads to interventions to prevent stillbirth and neonatal morbidity. Antenatal suspicion of growth restriction or fetal overgrowth changes obstetric care and can influence gestational age and birthweight (the exposures) and morbidity (the outcome) in prediction studies. The complexity of investigating this question is exacerbated in the sub-population of term live births, as growth screening affects selection into the sample. Effective screening of growth abnormalities can result in decisions to induce delivery before term (whereby the newborn would not be in this sample) or early term (newborn is in the sample, but risks of severe neonatal morbidity are mitigated). In an optimal, albeit unrealistic, scenario where all cases of suboptimal growth are detected and appropriately managed, birthweight percentiles could have no value for predicting severe neonatal morbidity at term, despite the high screening performance of growth charts. Stillbirths also create selection biases when only live births are studied, since high stillbirth rates may reduce risks of adverse neonatal outcomes in the liveborn population and conversely indicate the delivery of a fetus with severe growth abnormalities to avoid a stillbirth might increase neonatal morbidity. Despite the difficulties in interpreting the study's results about the predictive value of growth charts, the study by John et al. provides important knowledge about risks associated with birthweight across the birthweight-for-gestational age percentile distribution in the understudied population of term newborns. The authors employed innovative methods to resolve issues of ounce and digit preferences and to model the association between birthweight percentiles and neonatal morbidity at each week of gestation using nonlinear regression models. Their models make it possible to calculate, for each gestational age, the percentile at which morbidity is lowest and the percentiles at which morbidity odds increase by 10%, 50% or 100%. They find that severe neonatal morbidity is only substantially affected by very low and very high birthweight percentiles, with about 2% of infants at each extreme having higher rates of these adverse outcomes. Across other percentiles, risks are stable. These results illustrate that severe morbidity at term is not associated with birthweight for most infants born in the US with birthweights over the 3rd percentile and less than the 97th percentile. These results can orient preventive action, including evaluating growth screening programmes, and could serve as a benchmark for international comparisons with countries that have different screening practices and different levels of term mortality and morbidity. Another contribution of this study is to show that the percentages of SGA and LGA fetuses differ depending on the chart used, but that risk patterns remain the same. This finding of variation in SGA/LGA percentages between charts has led professional and scientific societies to urge that charts be validated to ensure that they ‘fit’ the population before local use.9, 10 Local validation studies were also recommended by the WHO group when they published their estimated fetal weight charts.2 In this more straightforward approach to evaluating performance (previously, the main one employed by researchers developing charts), the aim is to assess whether the chart accurately describes the birthweight (or estimated fetal weight) distribution of the population in which it will be applied. In other words, are 10% of fetuses/newborns below the 10th and above the 90th percentiles? These evaluations do not provide any information about health risks, but they make it possible to quantify the number of fetuses or infants who would be flagged for closer surveillance in screening programmes using different thresholds. Furthermore, charts that accurately describe the population provide valuable information about risks associated with percentile thresholds for clinical care and research. This knowledge can inform screening decisions and serve as a basis for developing more complex screening algorithms using other clinical or biological parameters or adapted to selected populations, for instance, high-risk versus low-risk populations. This approach turns the attention of evaluation studies away from generic measures of chart performance (selecting the ‘best’ chart) to a focus on the use of charts for improving obstetric and neonatal care and ultimately child health and development. While designing research to answer these questions is challenging, it provides a more realistic framework for selecting the outcomes and populations for studies to evaluate the clinical performance and health impact of growth charts. Alice Hocquette is a midwife and epidemiologist, and Assistant Professor in the Department of midwifery at Université Paris-Cité with a research affiliation in the Obstetrical, Perinatal and Pediatric Epidemiology Research Team, U1153, Paris. Her research focuses on the evaluation of growth anomalies and screening strategies, including the iatrogenic effects of screening on parental well-being and health service use. Jennifer Zeitlin is a tenured Research Director at INSERM (French National Institute of Health and Medical Research) in the Obstetrical, Perinatal and Pediatric Epidemiology Research Team, U1153, Paris and the Université Paris-Cité. Her research focuses on the impact of the organisation and quality of medical care on maternal and newborn health. She leads European projects on perinatal health indicators (Euro-Peristat) and very preterm birth cohorts. She also works on quality of care and disparities with a research group at the Department of Population Health Science and Policy, Icahn School of Medicine at Mount Sinai, New York, where she is an adjunct professor. She is deputy editor at Paediatric and Perinatal Epidemiology.