Association between perinatal mortality and morbidity and customised and non-customised birthweight centiles in Denmark, Finland, Norway, Wales, and England: comparative, population based, record linkage study

Data

We used record linkage data from five countries: national data from Denmark (2004-10), Finland (2004-14), Norway (2012-16), Wales (1986-2016), and regional data from the city of Bradford in England (2010-19). Full details of each dataset are available in the online supplemental material. Information from the separate countries was harmonised, and the analysis sample was restricted to singleton births that occurred between 37 and 43 gestational weeks, as in most previous studies, to distinguish fetal growth problems from preterm birth effects. We excluded infants with known major congenital anomalies as causes of death or identified within the first year of life to exclude cases where congenital anomalies might be the cause of fetal growth restriction, and influenced by some of the customised variables (eg, body mass index). We excluded observations with missing data on sex, birthweight or gestational age at delivery, and with birthweight outside the 0.15-8.00 kg range. We also excluded observations with customisation variables outside the ranges previously suggested for customised charts (maternal height 100-213 cm, maternal pre-pregnancy weight or early pregnancy weight 30-300 kg, parity 0-15).15 Online supplemental table S1 describes the sample selection (overall number of participants excluded: Bradford 3919 (8%), Denmark 49 437 (11%), Finland 73 715 (11%), Norway 17 628 (6%), Wales 94 327 (10%)).

Predictors

Outcomes

We examined the following outcomes in all countries when available: stillbirth (birth of an infant without signs of life), neonatal death (death within first 28 days of life for liveborn infants), infant death (death within first year of life for liveborn infants), low Apgar score at 5 minutes for livebirths (score <7, which includes 4-6=moderately abnormal and 0-3=low, to increase power and precision), and admission to the neonatal intensive care unit (NICU) within the first week of life.

Analysis strategy

Patient and public involvement

The current research uses secondary data and was not informed by patient and public involvement. No patients were involved in setting the research question, developing the study design or analysis plan, or setting the outcomes measures. The results will be disseminated to stakeholders and the broader public as relevant. There are no plans to disseminate the results of the research to individual study participants because all participants are deidentified.

Summary of main findings

In this comparative record linkage study, we did not find evidence that SGA or LGA births identified with customised birthweight centiles differed from non-customised centiles in their association with mortality (stillbirth, neonatal and infant mortality) or morbidity (low Apgar score and admission to NICU). However, the low sensitivity for both implies that a high proportion of infants at risk of these outcomes would not be identified by SGA or LGA with either chart.

Study strengths and limitations

We used data from four nationally representative datasets of births (Denmark, Finland, Norway, and Wales) and one regional dataset (Bradford, England), enhancing statistical precision. The record linkage datasets were large and covered the whole population, so we were able to compare associations between non-customised and customised for rare outcomes such as perinatal mortality with greater precision than previous studies. The nature of our outcomes (mortality and admission to NICU) and the data from high income countries means very little, if any, outcome misclassification is likely. To our knowledge, this study is the largest so far of these associations; with over two million births, it is double the size of a previous systematic review10 and two previous record linkage studies of about one million births.2 13 However, we acknowledge that we still had limited precision for some estimates, and associations of LGA births with neonatal and perinatal deaths were particularly imprecise with wide confidence intervals. NICU admission data were also only available for Bradford data and customisation variables were missing for 33% of that data.

We restricted our analyses to term births to better distinguish fetal growth problems from preterm birth effects, and to facilitate comparisons with previous studies. However, we acknowledge that this restriction could introduce one form of selection bias (ie, healthier sample than those born prematurely) over another (we do not have weight for those continuing in utero, and centiles derived from babies born preterm do not reflect healthy fetal growth).18 The extent of missing data varied across different datasets; and for some dataset this was considerable (eg, data from Wales had missing weight for 54%). Nevertheless, results were similar in sensitivity analyses where we used either the restricted the data or partial customisation, and the customisation equation coefficients were similar across different datasets with different amounts of missing data (online supplemental table S2).

Here, we focused on the pooled analyses across all data sources but acknowledge some heterogeneity in estimates across locations. This heterogeneity could be due to differences in ethnic group (online supplemental tables S6-S10 detail the categories used in the different datasets) or other maternal characteristics, or could reflect differences between countries in fetal growth monitoring, how adverse perinatal outcomes are screened for, and their risk managed. The data come from different years within each country spanning from 1986 to 2019, during which time perinatal mortality has decreased globally.19 A change over time in the outcome is unlikely to affect associations of exposures with that outcome and neither would it affect the comparisons within countries of non-customised and customised SGA/LGA births with the outcomes we have explored here. However, differences within countries could exist in how strongly customised/non-customised centiles associate with the outcomes that are influenced by the different time periods that are covered in each study, for example, owing to differences and changes in the ethnic composition of the country. These differences within countries could bias our (pooled) analyses, but seems unlikely because we do not see differences between non-customised and customised SGA/LGA birth in their associations with outcomes within countries. In the future, if countries become, for example, more ethnically diverse, differences between countries in the strength of the association of customised, centile derived, SGA/LGA births could emerge and that might result in customised SGA/LGA births having stronger or weaker associations with adverse perinatal outcomes in different countries.

Comparison with other studies

Consistent with our findings, a systematic review and meta-analysis did not find conclusive evidence of differences between SGA and LGA defined with customised and non-customised charts in predicting several important perinatal outcomes,10 and three subsequent studies of perinatal morbidity or mortality within countries have also shown no differences.2 11 12 In the previous meta-analysis,10 infants defined as SGA according to non-customised and customised charts had increased pooled risks compared with non-SGA infants for perinatal death (odds ratio 4.0 (95% confidence interval 2.8 to 5.1) v 5.8 (3.8 to 7.8)), neonatal death (2.9 (1.2 to 4.5) v 3.5 (1.1 to 8.0)), and NICU admission (2.4 (1.7 to 3.2) v 3.6 (2.0 to 5.5)). The authors concluded that the overlap of confidence intervals suggested little evidence of difference between the two charts but acknowledged that differences could become apparent with larger studies. With double the sample size of that previous study, we had considerably more precision to estimate associations with SGA births, but found that the confidence intervals for non-customised and customised estimates still overlapped. One consistency across all studies so far, including ours, is that associations of SGA births (defined by any of the charts) with adverse outcomes are stronger and more consistent than those associations of LGA births.

Two studies examined associations of SGA or LGA births with perinatal outcomes have compared customised centile charts to the Intergrowth-21st standards.13 14 Researchers from both studies concluded that the customised charts identified births at risk of being defined as SGA better than the Intergrowth-21st charts, and that the differences between the two charts were likely related to physiological differences by ethnic group. The larger study included 1.25 million births from 10 different countries, including Bhutan, China, India, the US, and six European countries, with the vast majority (97%) from the US or European countries.13 The results in relation to SGA births were driven by results from India (n=6436) and Bhutan (n=2779), which reflected <1% of the whole cohort.

We cannot directly compare our results to those from these previous studies because our study compared general population, non-customised reference charts rather than Intergrowth-21st charts. However, there could be some value in comparing these results to those from our study in the Bradford population (the most ethnically diverse of our data sources), where we found that SGA births based on non-customised charts had consistently higher relative risks for all outcomes than those based on customised charts, although with overlapping confidence intervals suggesting no statistical difference in associations. Based on these data, birthweight customisation did not appear to better identify a more at-risk group in Bradford, which had the highest proportions of infants with non-white maternal ethnicity and the lowest median birth weight compared with the other countries.

The role of ethnic group in customising fetal growth charts remains unclear, and its use in algorithms guiding clinical practice could normalise non-physiological differences that might even result in increasing ethnic inequalities in health.20 The customisation literature is unclear over how ethnic group should be conceptualised (self-identified ethnic group v country of origin/birth), which might also be important in relation to the extent to which ethnic group is appropriate to customise on. However, the promotion of customised charts is widespread globally and assumed to be measuring the same predictor. In all Nordic countries, parental country of birth/origin is often used as a measure of ethnic group, which is different to the concept of self-reported ethnic group (or race, for example, in the US) used in other countries. We feel strongly that in any research and development of global tools, such as customised centile charts, it is important to be clear about the meaning and justification of the measure of ethnic group.

Implications of studies so far

The systematic review concluded that, given the limitations of observational studies, randomised controlled trials were needed to investigate the use of customised and non-customised charts in monitoring infant growth and in deciding when to intervene.10 However, given the low incidence of severe adverse outcomes such as stillbirth and neonatal mortality, alternative approaches including natural experiments21 or smaller adequately powered trials for surrogates22 are required to determine the efficacy of using customised versus non-customised charts in clinical settings. For example, a natural experiment compared change in stillbirth rates over time between Scotland, where very few units have adopted the growth assessment protocol (GAP, which uses customised charts), with the same in England; researchers concluded that little evidence existed to support GAP being more effective at reducing stillbirth than standard care.21

In the first large randomised controlled trial to compare GAP with standard care, including over 180 000 deliveries from 13 obstetric units in the UK, researchers found no difference between GAP and standard care for the primary outcome of antenatal detection of SGA births, the proposed route through which GAP is hypothesised to reduce stillbirth.23 Evidence of a lower stillbirth rate was seen—which could have been a chance finding, given that researchers saw no differences in the other 24 secondary outcomes and no effect on the primary outcome that would be the mechanism for reducing stillbirth.23 The natural experiment and randomised controlled trials did not test whether customised charts were better than non-customised charts at identifying pregnancies at risk, and results for both chart types will be influenced by the effects of standard care in the comparison group. Thus, large observational studies remain the most feasible design for answering questions about which centile chart best predicts adverse perinatal outcomes.

Our results and those of previous studies suggest that SGA and LGA births that are defined by the bottom and top deciles of birthweight distribution in general have poor general predictive ability for these rare outcomes, irrespective of using non-customised or customised charts. We found similar high specificity and low sensitivity for non-customised and customised charts. Taking the pooled point estimate of stillbirths as an example, our findings suggest that non-customised and customised SGA/LGA births could identify, on average, 34% and 38%, respectively, of those who went on to have a stillbirth, missing more than 60% of events. Recent literature on birthweight centiles shows that no cut-off point for SGA performs well for predicting neonatal morbidity and mortality.24–27 Our aim was not to develop models to predict outcomes on an individual basis, but to assess whether customised centiles of birth weight were more strongly associated with outcomes than non-customised centiles, at the population level. On an individual level, continuous birth weight combined with multivariable predictive models could have improved predictive accuracy. Our analyses also did not assess wider programmes of plotting fetal growth and the benefits of reacting to changes in expected growth rates, such as faltering fundal height measurements, which remains an important area of research. Nevertheless, the use of customised charts in clinical practice requires more resources, so their clinical benefit would need to be demonstrable and balanced with their cost.