Workstream 1: a clinical outcomes study

The impact of place of birth and early care on mortality and morbidity.

Sections of this chapter have been reproduced with permission from Pillay et al.61 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Introduction

We address the primary research question of whether place of birth, by designation of unit as NICU or LNU, matters for preterm babies born at 27–31 weeks gestation. We study this for outcomes of mortality and key morbidities published in the literature for preterm births and added a measure of clinical care process62 to assess difference in the culture of care provided between units.

Aim

To evaluate differences in mortality and morbidity outcomes by type of unit of birth and early care in the first 72 hours of life of very preterm babies (born at 27–31 weeks gestation) in England.

Study design

Retrospective large-scale data analysis exploring and comparing outcomes between babies born and cared for initially in maternity services colocated with NICU compared with LNU, using routinely collected national data.

Objective

To assess, for preterm babies born at 27–31 weeks gestation and admitted to neonatal units in England (population), whether birth into maternity services colocated with NICU (intervention) compared to LNU (comparator) and early care in the first 72 hours in the respective neonatal units offered a gestation-specific:

1. survival advantage while in neonatal care and up to 1 year of age (outcome), and
2. reduction in major neonatal morbidity (outcome).

Methods

Data source

Data were obtained from the NNRD. All NHS neonatal units in England used their electronic patient record (EPR) supplier systems to routinely submit detailed information to the NNRD on the outcomes of and clinical care provided to babies. Submitted data are quality-assured and curated to a research standard. NNRD data comprise demographics, diagnoses, health outcomes, daily interventions (including the level of care, namely intensive, high-dependency, special, or normal care, provided each day) and treatments administered during the inpatient episode of care. We used data collected for babies admitted to neonatal units in England. This baby-level data set was linked by NHS Digital Office for National Statistics63 for mortality data, for up to 1 year of age.

The OPTI-PREM cohort comprised preterm babies born at 27–31 weeks gestation born in maternity services colocated with a NICU or an LNU in England, and who were discharged from or died in neonatal care between 1 December 2014 and 31 December 2018. We selected date of discharge as opposed to date of birth, so that we could have a defined end date for data extraction from the NNRD.

Outcome variables

Primary outcomes were death in neonatal care and death during the first year of life (infant mortality). Secondary outcomes were important neonatal outcomes [ROP, severe/serious brain injury (SBI), NEC, BPD], and one measure of clinical care process [receipt of any breast milk feeds at discharge from neonatal care (BMF)]. We compared outcomes between babies born in maternity services colocated with NICU with those in services colocated with LNU.

Definitions

Severe retinopathy of prematurity: a baby with Stage 3+ ROP64 in one or both eyes. All babies who had laser surgery were also included on the assumption that they would have Stage 4 ROP to warrant surgery.

Severe/serious brain injury: a baby having one or more of the following at any time while in neonatal care: periventricular leukomalacia, unilateral or bilateral grade 3 or 4 intraventricular haemorrhage, porencephalic cysts or hydrocephalus.

Necrotising enterocolitis: a baby undergoing surgical treatment/laparotomy for NEC while in neonatal care, with information extracted from the abdominal X-ray table in the NNRD data and supplemented with information reviewed in the entries on daily data of NEC. Cases where suspected NEC was recorded were excluded.

Bronchopulmonary dysplasia: oxygen dependence at 36 weeks post menstrual age, and a requirement for oxygen for at least 80% of the days before reaching 36 weeks post menstrual age. Babies who were discharged before 36 weeks post conceptional age were regarded as not oxygen dependent. The data were extracted using the added oxygen variable within the daily data. It was clinically accepted that babies who were oxygen dependent would not be discharged before 36 weeks gestation in England.

Receipt of any breast milk feeds at time of discharge from neonatal unit (BMF): was extracted from the final day records on the NNRD.

Place of care for babies: a baby’s care was assigned to NICU or LNU according to whether the maternity service in which they were born was colocated with NICU or LNU.

Composite outcomes: outcomes for ROP, SBI, NEC, BPD and death were combined to describe the composite adverse outcomes for babies.

Statistical methods

Sample size

We estimated a sample size of approximately 26,250 based on 6000 births per year and a 48% NICU and 41% LNU split between place of birth. We assumed an in-unit mortality of approximately 5%, giving over 1000 deaths in the sample. Assuming 5.0% of babies in the NICU died before discharge, this anticipated sample size of 26,250 admissions would give a minimum detectable difference of 2.6% in mortality for the LNU with at least 80% power at the two-sided 5% significance level in a simple comparison. The inclusion of over 1000 deaths in the study was expected to provide sufficient statistical power to investigate the potential predictors of mortality and morbidity.1

Measured and unmeasured confounders

In the absence of a randomised allocation of babies to each ‘neonatal unit type’, we considered that appropriate statistical methods were required to address the biases that would likely arise from a non-randomised comparison of outcomes of babies receiving care in each unit type. To control for measured confounders, we used propensity score matching to create a sample of babies managed in each setting, who were balanced in terms of neonatal and maternal characteristics. The following observed neonatal and maternal characteristics were used in the matching exercise and were selected after discussions at multiple OPTI-PREM collaborator and study steering committee meetings: gestational age; sex; birthweight z-score; multiplicity; maternal ethnicity; maternal age; index of multiple deprivation (IMD) score and mode of delivery.65,66 The result of this matching exercise (see Report Supplementary Material 1) suggested that we had limited information about measured confounders. For instance, preterm babies born into maternity services colocated with NICUs could be more complex due to high-risk maternal and fetal illnesses compared to those delivered in services colocated with LNU. The NNRD contains limited maternal clinical data to conduct an adjusted analysis of all confounders of importance. In addition, the potential existence of unobserved characteristics that varied by unit type could result in biased estimates of outcomes, if left unaddressed. To account for the available measured confounders and additional unmeasured confounding, an instrumental variable (IV) approach was used.67,68

The instrumental variable approach

With this approach, the aim is to identify a variable that supports women giving birth to their preterm baby in a particular setting and that naturally controls for observable and unobservable confounders. This variable is referred to as an instrument, and in line with previous work we considered that the difference in the distances between the mother’s residential postcode and her nearest NICU postcode, and the mother’s residential postcode and her nearest LNU postcode (known henceforth as the excess travel time to a NICU) would likely be an appropriate IV.9 Details of how this variable was generated for babies in the study are provided in Report Supplementary Material 2. To be considered an appropriate IV, a variable should be highly correlated with the independent variable of interest (in this case the ‘unit type’) but have no direct effect on the outcomes of interest (mortality and complications). Crucially, an IV should not be related to the unobservable confounders biasing the estimate between the independent variable of interest and the outcome. In the context of each of the requirements for an IV outlined above, data have previously shown that high-risk women will likely deliver in a NICU if they are living closer to a hospital with a NICU, and thus we hypothesised that excess travel time to a NICU would be highly correlated with ‘unit type’.9 We also considered that excess travel time to a NICU would not have a direct impact upon infant costs and outcomes, except through the route of whether the baby was delivered in a NICU or an LNU. Finally, it is highly unlikely that excess travel time to a NICU would be related to any unobserved confounders as women generally do not choose to live close to a hospital with a NICU in case they ever have a preterm birth. Excess distance is thus likely to be randomly distributed across the cohort. Travel time to a facility is a common candidate of instrument in epidemiological studies with observational data.69

To assess the strength of our chosen instrument, we evaluated (1) whether there was a strong correlation between ‘unit type’ and excess travel time, and (2) whether the instrument distributed measured confounders equally using the median of the excess travel time distribution across the whole cohort as a cut-off value. We then estimated our IV models for primary and secondary outcomes using bivariate probit regression. The choice of the bivariate probit instead of the standard probit model was the binary nature of ‘unit type’ (NICU or LNU).70 The models also included the measured confounders used in the matching exercise and described above.

To account for the multiple comparisons used within our analyses, we set the CIs for statistical significance for all analyses at 99% (p ≤ 0.01).

Morbidity and care process outcomes

We extracted key morbidity outcomes and the BMF care process outcome from the diagnoses, daily data on the NNRD, discharge summary and other relevant data entries. We assumed that when no morbidity was entered in the daily data, diagnoses, discharge summary or recorded elsewhere it was assumed that there was no event rather than missing data, except for ROP. All babies < 32 weeks gestation are nationally mandated to have a ROP screening examination at 4–6 weeks of age. Therefore, where this was not recorded on the data extract, we presumed it to be missing data. These babies were removed from the ROP analyses. For the analyses on ROP, BPD and BMF we also excluded babies who had died while in neonatal care. For morbidities of SBI and NEC, we included babies who had died while in neonatal care.

Sensitivity analyses

We undertook sensitivity analyses to explore data from prespecified subgroups of babies. These were defined in our statistical analysis plan and undertaken for mortality and morbidity outcomes.

1. Early transfers: to ensure that our results were not influenced disproportionately by babies who were born in a LNU but who were critically unwell and required clinical transfer of care to a NICU, we repeated the analyses after excluding (1) babies who were transferred in the first 24 hours, and (2) those transferred in the first 72 hours of life. We chose a 72-hour cut-off based on the NHS specifications for neonatal services, which recommend that critically ill babies in an LNU should be transferred to a NICU, or that their condition should be discussed with the neonatal team in a NICU for ongoing care after 48 hours of stabilisation.7 We added a further 24 hours to accommodate the logistics of the actual transfer process. We defined ‘early transfers’ as babies who were transferred out and were receiving care in a location other than that into which the baby was born, in the first 24 hours, and up to the first 72 hours of life. We also excluded babies who had died in the first 24 and up to the first 72 hours from analyses. We considered the babies who were not transferred and had not died within this 72-hour time frame to be representative of those receiving early care in their respective units.
2. Singleton compared with multiple pregnancies: to account for maternity services colocated with a NICU being more likely to deal with complex multiple pregnancies, which carry a higher risk of neonatal mortality, we repeated our analyses excluding all babies born to mothers with multiple pregnancies.

Analysis independent of neonatal intensive care unit–local neonatal unit unit designation

To avoid overlooking higher-performing LNUs that were working as NICUs and vice versa, we undertook a further analysis comparing mortality outcomes for babies born and cared for in high-volume compared with low-volume neonatal units.11 We defined high-volume units as above the uppermost quartile for number of intensive care bed-days offered to babies born at < 32 weeks gestation during the OPTI-PREM 5-year study period. Units below the uppermost quartile were defined as low-volume units.

Results

Summary statistics

From the NNRD data extraction of 29,842 preterm babies born at 27–31 weeks gestation and discharged from neonatal care between 1 January 2014 and 31 December 2018, a total of 18,847 babies were studied (Figure 2). Births outside of NICU and LNU, babies with known congenital anomalies (see Appendix 1, Table 31) and missing data were excluded (see Report Supplementary Material 3), resulting in 10,379 babies born in maternity services linked to NICU and 8468 to LNU. A comparison of this cohort to the group with missing data (n = 7438) is outlined in Appendix 2, Table 32. They were similar regarding gestational age, place of birth and mortality outcome. Our data compared favourably when we compared the birth statistics for the latter 3 years of our cohort with data reported to the Office for National Statistics of all very preterm babies born in England between 2016 and 2018 (see Appendix 3, Table 33).

FIGURE 2 OPTI-PREM data study flow chart. (Figure not reproduced here; see the original article.)

Twice as many preterm babies born at 27 weeks gestation were born in maternity services colocated with NICU and admitted to NICU compared to LNU [Table 1: 1507 (14.5%) compared with 777 (9.2%)]. This difference decreased as gestational age in weeks increased, reaching equivalent numbers by preterm birth at 29 weeks, with higher numbers of births at 31 weeks gestation in LNU compared with NICU (34% compared with 28%). Preterm babies born and admitted into NICU were smaller in weight, had similar sex distribution, median Apgar scores at 5 minutes, receipt of antenatal steroids and admission temperatures compared to those preterm babies born and admitted to LNU. There was a higher proportion of mothers of mixed or ethnic minority groups, higher levels of social deprivation scores and higher caesarean section rates in mothers who delivered their preterm babies in maternity services colocated with NICU (Table 1). However, when defining the cohort against the median differential time of excess travel] [median (range) 3.91 (–130.36 to 74.11) minutes], the groups appeared to be balanced (standardised differences < 0.1) (Table 2). This was noted in all categories except IMD scores, which is an acceptable imbalance linked to urbanicity when using distance instruments.72

Mortality outcomes

There were 574 deaths (3.0%) while in NICU and LNU care, and a further 121 deaths (3.7% total mortality) in the first year of life after discharge from neonatal care. Babies admitted to NICU had a higher unadjusted mortality while receiving neonatal care (3.8% for NICU admission compared with 2.2% for LNU admission; p < 0.001) and higher unadjusted infant mortality (4.5% for admissions to NICU and 2.7% for LNU admissions; p < 0.001; Table 1). When comparing neonatal characteristics by median differential travel time to a NICU, the covariates were balanced indicating similar covariate distribution of measured and potentially unmeasured confounders (see Table 2).

After adjusting for measured and unmeasured confounders, the mean difference for mortality between babies born in maternity services colocated with NICU compared with LNU while receiving specialised neonatal care was −0.001 (99% CI −0.011 to 0.010; p = 0.842), and for infant mortality −0.002 (99% CI −0.014 to 0.009; p = 0.579). Comparisons for both the overall group of babies born at 27–31 weeks gestation, and for each week of gestation, did not reveal significant differences (Table 3).

Sensitivity analyses

Early transfers

A total of 12,831 (68.1% of 18,847) babies received all their neonatal care in one unit (wholly owned). Of the 6016 babies that were transferred to another unit, 720 (12.0%) babies were transferred in the first 24 hours of life, and 1545 (25.6%) in the first 72 hours of life. One out of every two babies (49.7%) born at 27 weeks gestation were transferred between units while in neonatal care, of which 13.6% occurred in the first 72 hours of life. This was much higher when compared to babies born at 31 weeks gestation, where 22.7% were transferred overall, with 6.4% in the first 72 hours of life (see Table 4). Overall, there was a high rate (31.9%) of movement between units for this cohort (Table 4).

We did not detect a significant difference in mortality during neonatal care and up to 1 year of age when sensitivity analyses were conducted excluding early transfers in the first 24 hours of life (mean difference 0.000; 99% CI −0.011 to 0.010; p = 0.915 while in neonatal care; mean difference −0.002; 99% CI −0.014 to 0.010; p = 0.659 for mortality up to the first year of life), and in the first 72 hours of life (mean difference 0.002; 99% CI −0.009 to 0.013; p = 0.634 for mortality while in neonatal care and mean difference 0.000; 99% CI –0.012 to 0.012; p = 0.961 for mortality up to 1 year of age). Similarly, when we considered the group who had either been transferred and/or had died in the first 24 hours, or in the first 72 hours of life, and excluded these from the analysis, we found no difference in mortality outcomes between NICU and LNU for our cohort.

Singleton compared with multiple pregnancies

There were 456 deaths while in neonatal care and a further 80 deaths after discharge up to the end of the first year of life (3.9%) in singleton babies (n = 13,572). There were 118 deaths while in neonatal care and a further 41 deaths (3.0%) in the first year of life of babies born in multiple pregnancies (n = 5275) in the OPTI-PREM cohort. No effect was found when only singleton births were analysed while in neonatal care and in the first year of life. We were unable to replicate the analyses for multiple pregnancies as the sample size did not fit the model.

Analysis independent of NICU-LNU designation

High-volume units for intensive care days (IC days) were units in the uppermost quartile and offered between 11,647 and 8068 IC days to babies born at 27–31 weeks gestation over the 5-year period. This averaged to > 1614 IC days per year. Low-volume units were therefore those that offered ≤ 1614 IC days per year.

When comparing mortality outcomes for the OPTI-PREM cohort, between high- and low-volume units based on the uppermost quartile of IC days versus lower three quartiles of ICU days provided, and independent of NICU and LNU designation, no significant differences were observed (Table 5).

Morbidity outcomes

Severe brain injury was documented in 735 (3.9%) of the 18,847 babies studied, ROP in 297 (1.65%) of 17,930 babies, BPD in 1819 (10.0%) of 18,273, NEC in 490 (2.6%) of 18,847. 10,220 (55.9%) of 18,273 babies received BMF at discharge from neonatal care.

Severe/serious brain injury

There was a significant difference with a higher proportion of SBI in babies born and cared for in a LNU (mean difference −0.011; 99% CI −0.022 to −0.001; p = 0.007, Table 6). The highest mean difference in gestation-specific SBI was in the group of babies born at 27 weeks gestation (−0.040). 14.5% (45/310) of preterm babies born at 27 weeks gestation and transferred in the first 72 hours of life had SBI, compared to 3.5% (13/368) of those born at 31 weeks gestation and transferred in the first 72 hours of life.

Statistical significance was not maintained after exclusion of early postnatal transfers for the whole group (early transfers up to 72 hours n = 1545; mean difference −0.002, 99% CI −0.013 to 0.08; p = 0.554), and then separately, after exclusion of all babies born at 27 weeks gestation (mean difference −0.008; 99% CI −0.019 to 0.002; p = 0.037) (Table 7).

For babies born at 27 weeks gestation, birth in maternity services colocated with NICU (tertiary neonatal unit) reduced the risk of SBI from 11.9% to 7.7%, a reduction of 4.2%. This translated into a number needed to treat (NNT) of 25 (99% CI 10 to 59) indicating that 25 babies at 27 weeks gestation would need to be delivered in a NICU, rather than a LNU to prevent one SBI.

High-versus low-volume analysis

We analysed SBI by high-versus low-volume units (using the IV approach and adjusting for confounders), defined by the uppermost quartile for IC care days provided compared to the lower three quartiles (see Report Supplementary Material 4). For the SBI analysis, high-volume units were defined as those offering ≥ 8304 IC days/5 years, or approximately > 1614 IC days/year, and low-volume units < 8304 days/5 years or ≤ 1614 IC days/year. There was a significant difference with a lower proportion of SBI in babies born and cared for in high-volume units compared to low-volume units (mean difference −0.065; 99% CI −0.121 to −0.009; p = 0.003). For babies born at 27 weeks gestation, birth in a high-volume unit reduced the risk of SBI from 0.242 to 0.028, a mean difference of 0.289 (99% CI 0.035 to 0.542; p = 0.003). Here the NNT for 27 weeks was 4 (99% CI 2 to 29), indicating that four babies at 27 weeks gestation would need to be delivered in a high-volume unit, to prevent one SBI.

Statistical significance was lost after exclusion of early postnatal transfers up to 72 hours for the whole group (mean difference −0.0423; 99% CI −0.093 to 0.008; p = 0.029) and then separately, after exclusion of all babies born at 27 weeks gestation (mean difference −0.045; 99% CI −0.096 to 0.005; p = 0.019). Four thousand five hundred and ninety-seven babies were in units considered high volume by the above definition; these units were all NICU (see Report Supplementary Material 4).

We further analysed high- versus low-volume units comparing the upper three quartiles for IC days to the lowermost quartile for IC days provided by units per year. Against this definition, high-volume units were those offering ≥ 2197 IC days/5 years or approximately ≥ 440 IC days/year and low-volume units < 2197 days/5 years or < 440 IC days/year. There were no significant differences in SBI rates (mean difference −0.010; 99% CI −0.026 to 0.006; p = 0.115) (see Report Supplementary Material 4) between high- and low-volume units using this definition. Here 13,969 babies were in units considered high-volume units (comprising both NICU and LNU), and 4812 in low-volume units (comprising only LNU) (see Report Supplementary Material 4).

Bronchopulmonary dysplasia

The rate of BPD was higher in the overall group born into maternity services colocated with NICU (mean difference 0.018; 99% CI 0.001 to 0.035; p = 0.006 workstream 1, Table 6). This significance remained after excluding early transfers at 24 and up to 72 hours of life (mean difference for early transfers up to 72 hours was 0.029; 99% CI 0.011 to 0.047; p < 0.001). The highest mean difference in gestation-specific BPD was in the group of babies born at 27 weeks gestation (−0.080). When preterm babies born at 27 weeks gestation were excluded from the overall BPD analyses, the significant difference seen in overall BPD was lost (mean difference 0.011; 99% CI −0.004 to 0.027; p = 0.065).

There was no effect of place of birth on ROP, NEC or BMF. For NEC, ROP, BMF and composite outcomes, we did not detect a statistical difference overall or on analysis at each week of gestational age between babies born into maternity services colocated with NICU compared with LNU, even after exclusion of babies transferred within 24 and 72 hours after birth. Similarly, there was no significant difference when analysing only singleton births.

Discussion

Using robust IV analysis, we found for preterm babies born at 27–31 weeks gestation in England, that there was no survival advantage for birth in NICU (tertiary) compared with LNU (non-tertiary) settings, either during neonatal care or in the first year of life. However, there was a higher proportion of SBI in those babies born in maternity services colocated with LNU. We identified that these observations were driven by babies born at 27 weeks and those transferred out of LNU in the first 72 hours of life. To prevent one case of SBI, 25 babies would need to be cared for in NICU instead of LNU at this gestation.

When we investigated the volume of work undertaken by units, we found that units (which turned out to be all NICU) providing > 1614 IC days per year for babies born < 32 weeks gestation were associated with significantly lower SBI. Here, to prevent one case of SBI, four babies would need to be cared for in a high-volume NICU instead of other units at 27 weeks gestation.

Based on current practices, care for babies born at 28–31 weeks gestation in England in LNU appears to be equitable to that of NICU, even when transfers are excluded. Our findings suggest therefore that the safe threshold for care in either LNU or NICU for very preterm births in England is 28 weeks gestation. This finding has important implications for policy-makers considering optimising health service delivery for preterm babies in England. It provides scientific evidence to support implementation of the 2023 NHS 3-year delivery plan for maternity and neonates.73

Findings in context

While we identified a higher proportion of SBI in babies born at LNU and postnatal transfer, we cannot assign causality of the SBI entirely to the transfer process itself. Determining the contributions to SBI from 1) whether these babies were predisposed to this for reasons associated with their degree of illness, 2) the care received in LNU, or 3) the postnatal transfer process is not possible from our data, nor from other studies that have identified a higher likelihood of SBI in those very preterm babies transferred postnatally. Regardless, these observations are concerning, and together suggest that care for anticipated preterm births at 27 weeks gestation should be redirected to maternity services colocated with NICU in the antenatal period.

Severe brain injury in preterm babies can be associated with neurological findings ranging from mild developmental delay to severe cerebral palsy. The short- and long-term impact for preterm babies and their families,74 and in terms of neurological, psychological, educational and economic cost75–84 to society is substantial. If we can prevent one preterm baby born at 27 weeks gestation having SBI, the cost savings to the NHS, to babies, families and society would be substantial, considering the intensity of healthcare support required and high healthcare costs for cerebral palsy,85,86 a known outcome following SBI in preterm births.

The observation of a higher proportion of SBI in postnatal transfers in our study adds to growing evidence of the risks for very preterm babies being born in the wrong place, and postnatally transferred to tertiary centres. In a study of 17,577 preterm births in England, Helenius et al.87 identified an increased risk of SBI in upward transfers in the first 48 hours of life. In a study of very preterm babies born at < 32 weeks in Western Australia, Davis et al. noted a higher rate of combined brain injury (10.7% compared with 6.0%) in outborn compared with inborn babies. In OPTI-PREM, we were unable to delineate the reasons for transfer, that is whether the early transfers in our LNU cohort represented (a) upward transfers only, from LNU to NICU for a higher intensity of care, or (b) as dictated by local protocol or included (c) transfers for capacity reasons.87,88

During the period of the study, a few neonatal operational delivery networks encouraged antenatal transfer of mothers expected to deliver at 27–28 weeks gestation to maternity services colocated with NICU, or postnatal transfer to NICU. National recommendations for anticipated multiple births at < 28 weeks were to triage mothers to maternity services colocated with NICU,7,14 and for babies requiring > 48 hours of intensive care support in a LNU to be discussed with teams at NICU for potential postnatal transfer and care in NICU.7 These factors partly explain the observation of almost twice as many babies born at 27–28 weeks being born in maternity services colocated with NICU compared to LNU and may have masked any potential negative effect that birth in maternity services colocated with LNU may have held. We were unable to identify in utero transfers within the OPTI-PREM cohort data set, to adjust for this observation.

Strengths and limitations

A major strength of this study is its use of large-scale, national, operational data, analysing over a million daily patient records, utilising a robust IV to address unmeasured confounders, while findings are supported by sensitivity analyses. The alternative of a prospective randomised controlled study in this cohort would be unethical, costly and not feasible. We were unable to study four major maternal medical confounders (pregnancy induced hypertension, diabetes, chorioamnionitis and multiple pregnancies compromised by twin-to-twin transfusion) because of incomplete recording of maternal details (see Chapter 6) on the neonatal database. Our adjustments relied on key measured confounders including indices of multiple deprivation (as most NICU are historically located in areas of higher social deprivation in England), sensitivity analyses for transfers and multiple births. We used a robust IV approach to account for the unmeasured confounders of fetal and maternal illness that could have resulted in a higher proportion of ill preterm babies being born in NICU compared with LNU by virtue of being redirected there antenatally or in labour. It is reassuring that on utility of the median excess travel time to describe the covariates, the groups appeared balanced, indicating similar covariate distribution of measured and (potentially) unmeasured confounders. We excluded a high number of babies due to incomplete data, but sensitivity analyses were reassuring, in identifying that the cohort excluded due to missing data was similar demographically and in clinical condition at birth (Apgar at 5 minutes, and admission temperature) to the cohort that was finally studied. We also excluded babies transferred in the first 24 hours and those that had died in the first 24 hours to account for the inadvertent baby born too ill, which may have biased our assessments. A further strength of the study is capturing mortality outcomes to 1 year through linkage with NHS Digital Hospital Episode Statistics for mortality. We did not use Apgar scores or neonatal clinical risk index for babies scores in the adjustments as we considered these an effect of care related to place of birth.

Deaths in the delivery suite could not be studied because of incomplete entries on the neonatal electronic records; linking NNRD neonatal unit admission data with our 20 existing maternity databases was not feasible during the study period. There is also the potential for data quality to be reduced as it relied on routinely collected EPR data. Further limitations are described in Risk of bias of estimated treatment effect in primary and secondary outcomes.

We identified a significant association with BPD in those born in maternity services colocated with NICU. This appeared to be intrinsic to NICU, not due to early postnatal transfers, and contributed to by those born at 27 weeks gestation. In comparison to SBI, we considered this a less impactful observation, and one that could be potentially influenced by variations in care processes between units over the period of the study, as well as by the indirect impact of social deprivation. Further clinical evaluation and qualitative work is required to understand the differences in quality of neonatal care provided around weaning from oxygen, and BPD.

Risk of bias of estimated treatment effect in primary and secondary outcomes

Due to the unpredictability of preterm birth, a randomised clinical trial format could not be utilised to answer the research questions that were being explored by OPTI-PREM. As a non-randomised study, the project is subject to the following biases:

Confounding bias

Limited knowledge on known maternal confounders that could affect treatment effect

We had limited data on some maternal comorbidities which may have influenced presentation and care in maternity care in centres colocated with NICU, and therefore influenced the degree of illness in the baby at birth, and subsequent outcomes. For example, there were 10,736 (38.1%) missing entries for maternal diabetes, 10,460 (37.1%) missing entries for maternal hypertension, and 10,595 (37.6%) for maternal infection. As a result, our matching exercise failed and the use of an IV approach to account for unmeasured and available measured confounders was employed primarily. IV methodology is subject to limitations, but we have mitigated against these by using a strong instrument, and also placing the results of our study in the current context and previous evidence available in preterm and critical care research, and not simply in isolation.72,89

Addressing maternal confounders through sensitivity analyses and adjusting for measured confounders

We performed sensitivity analyses accounting for multiple pregnancies (and twin-to-twin transfusions) which could have been redirected to NICU for antenatal care, and in whom outcomes may be more adverse between NICU and LNU. We adjusted for severe intra-uterine growth restriction, which is a manifestation of degree of maternal and fetal illness, and therefore may be different between babies born and cared for in NICU and LNU by using a birthweight z-score. We did not account for differences in degree of illness of baby at birth by using the Apgar score in our adjustments as we considered this to be a measure of the care provided by teams at resuscitation in LNU and NICU, which is what the project was intending to explore.

Unknown confounders

The use of the IV method also addressed unknown unmeasured confounders that could have skewed babies born in NICU having worse outcome by virtue of other maternal and fetal comorbidities, and other factors related to NICU and LNU. We excluded babies with known congenital anomalies (see Appendix 1, Table 31) to further reduce a treatment effect bias that could have arisen from the redirection of expectant mothers with major fetal congenital anomalies to highly specialised fetal medicine units, which are colocated with NICU in England.

Differences in IMD distribution between groups were still observed across the IV instrument. It is common and is recognised that the use of distance measures (excess travel time) as instrument result in ethnicity and deprivation imbalances.90 In our study, we believe that this reflects the current geographic distribution of NICU (more socially deprived areas) and LNU.

Selection bias

This is low, as all babies born between the study period were included in the analysis, except for data from one unit, and 10 babies, whose parents declined consent. We could not determine exact number of babies that were excluded in the single unit that opted out of OPTI-PREM, which was classified as a LNU during the study period. On average 64 babies per year were cared for in LNU (see Chapter 2), which means that for the cohort ~1% of the sample was not selected (64 babies per year × 5 years).

However, we lost a large component of the sample to use our statistical methods (IV). When we compared maternal/baby characteristics of the final sample versus the original, they were broadly similar.

Our study was designed only to compare outcomes in those that survived to receive neonatal care and excluded deaths in the delivery suite or outside of hospital. This could have been influenced by the decision to provide sustaining life support or resuscitation support in LNU versus NICU. However, for babies born at 27–31 weeks gestation, this is clinically less of an issue as all babies are required to be offered care, unlike those born at 22–23 weeks gestation, where the decision on providing life sustaining support can be varied based on place of birth.91–95

Information bias

We consider this bias to be low, as the information on place of birth was always present, and the primary outcome of mortality was noted for all babies as they all required a discharge summary from the neonatal EPR, and we used national statistics for mortality up to a year.

For our secondary outcomes, we were limited by accuracy of data entry in the routinely collected EPRs. We mitigated against this by searching the database using information located in more than one field (e.g. for the definition of NEC), to derive our definitions.

Reporting bias

Our project report aligns with the published protocol,1 in which we meet the set objective. The risk for this is low.

Summary points from OPTI-PREM clinical outcomes study, workstream 1

What was already known prior to the study

• Care for extreme preterm births ≤ 26 weeks gestation offers a survival advantage and morbidity advantage if born in maternity services colocated with NICU and cared for in NICU.
• There is no evidence to support best place of birth and early care for the next most vulnerable group, that is very preterm babies born at 27–31 weeks gestation. We do not know if there is a threshold below which care offers survival and morbidity advantages in NICU or vice versa.

What the OPTI-PREM study adds

• There is no survival advantage for birth and early care in NICU compared with LNU for babies born at 27–31 weeks gestation in England.
• Babies born at 27 weeks gestation, and who were transferred out of LNU in the first 72 hours of life, were more likely to have SBI.
• Therefore, in England the threshold for safe birth and early care in neonatal units in either NICU or LNU (perhaps closer to home) is 28 weeks.
• We calculated that 25 babies at 27 weeks gestation would need to be delivered in a NICU, to prevent one SBI.
• Our secondary calculation was that four babies at 27 weeks gestation would need to be delivered in a high-volume unit (> 1614 ICU care days per year) to prevent one SBI.

How the OPTI-PREM study may potentially influence policy

(See Chapters 9 and 10).

• Review of resources to support antenatal triage of all anticipated preterm births at 27 weeks gestation in maternity services colocated with NICU.
• Clinical judgement for all inadvertent births at 27 weeks gestation in maternity services colocated with LNU, regarding transfers.
• These will need to be risk-assessed for clinical benefit against risk of SBI ± parental stress.
• Capacity transfers, or routine protocol-driven transfers, in well preterm babies born at 27 weeks out of neonatal units may not represent optimal practice.