Abstract
Aim
This study quantified the impact of perinatal predictors and medical centre on the outcome of very low‐gestational‐age neonates (VLGANs) born at <32 completed weeks in Switzerland.
Methods
Using prospectively collected data from a 10‐year cohort of VLGANs, we developed logistic regression models for three different time points: delivery, NICU admission and seven days of age. The data predicted survival to discharge without severe neonatal morbidity, such as major brain injury, moderate or severe bronchopulmonary dysplasia, retinopathy of prematurity (≥stage three) or necrotising enterocolitis (≥stage three).
Results
From 2002 to 2011, 6892 VLGANs were identified: 5854 (85%) of the live‐born infants survived and 84% of the survivors did not have severe neonatal complications. Predictors for adverse outcome at delivery and on NICU admission were low gestational age, low birthweight, male sex, multiple birth, birth defects and lack of antenatal corticosteroids. Proven sepsis was an additional risk factor on day seven of life. The medical centre remained a statistically significant factor at all three time points after adjusting for perinatal predictors.
Conclusion
After adjusting for perinatal factors, the survival of Swiss VLGANs without severe neonatal morbidity was strongly influenced by the medical centre that treated them.
Keywords: Neonatal morbidity, Outcome variability, Perinatal predictors, Survival, Very low‐gestational‐age neonates
Abbreviations
- ANC
Antenatal corticosteroids
- BPD
Bronchopulmonary dysplasia
- CI
Confidence interval
- GA
Gestational age
- NICU
Neonatal intensive care unit
- NEC
Necrotising enterocolitis
- OR
Odds ratio
- PIVH
Periventricular/intraventricular haemorrhage
- PVL
Periventricular leucomalacia
- ROP
Retinopathy of prematurity
- VLGANs
Very low‐gestational‐age neonates
Key notes.
This study quantified the impact of perinatal predictors and medical centre on the outcome of very low‐gestational‐age neonates (VLGANs).
Using prospectively collected data, we identified 6892 VLGANs born in Switzerland at <32 weeks of gestation between 2002 and 2011.
Important predictors for adverse outcomes were patient‐level factors, obstetrical interventions, proven sepsis in the first week of life and the medical centre where the neonate was treated.
Introduction
Survival rates and survival without morbidity increase sharply with gestational age (GA) among very low‐GA neonates (VLGANs) born at <32 completed weeks, and GA has traditionally been used as the sole outcome predictor. More recently, models that have included perinatally known factors, such as estimated foetal weight, gender, single or multiple birth, antenatal corticosteroid (ANC) administration and delivery mode, have been shown to increase the accuracy of predicting survival 1, 2, 3.
Variations in the survival rates of preterm neonates have been observed between centres providing health care in several populations and networks 4, 5, 6, 7. Factors contributing to these differences in outcome have included the inherent risk of the patient population served, the local approach towards primary nonintervention at the border of viability and/or the effectiveness of the patient care delivered at a particular medical centre. Despite the availability of national guidelines on perinatal care at the limit of viability, between 22 and 26 completed weeks of gestation 8, survival rates vary widely across the nine tertiary level perinatal centres in Switzerland, even after adjusting for important patient‐related factors 9, 10.
The chance of survival changes rapidly throughout the first week of life in infants born at <32 weeks of gestation, due to the fact that mortality is highest during the first few days of life. Therefore, predictors for survival and short‐term morbidity may vary depending on postnatal age.
Existing models that aim to predict survival or morbidity rates of VLGANs do not address the potential centre effect on outcome and do not take into account the rapidly changing chances of survival during the first few days of life. Therefore, we sought to quantify the impact of different perinatal predictors and medical centre on survival and short‐term morbidity at three different time points in infants born at <32 weeks in Switzerland.
Materials and Methods
Data sources
The Swiss Neonatal Network and Follow‐Up Group maintains a database, the Minimal Neonatal Data Set (MNDS), to prospectively collect anonymised information about the demographics and outcome to the point of hospital discharge of all live‐born infants weighing between 400 and 1500 g at birth and with a GA of between 23 0/7 and 31 6/7 weeks. It covers infants born at, or transferred to, one of the nine level III neonatal intensive care units (NICUs) caring for VLGANs in Switzerland. Data collection and evaluation were approved by the institutional ethical review boards and by the Swiss Federal Commission for Privacy Protection in Medical Research. In this study, data from the Swiss MNDS were used to analyse information on all VLGANs of 23–31 completed weeks born alive between January 1, 2002, and December 32, 2011.
Definition of neonatal variables
Gestational age was calculated based on ultrasound examinations during the first trimester of pregnancy and defined as the postmenstrual age in weeks and days. Birthweight standard deviation scores (z‐scores) for GA were calculated based on the growth curves published by Voigt et al. 11.
Antenatal corticosteroids administration was considered to have occurred if at least one dose was given prenatally. Major congenital malformation was defined as any type of malformation that had a severe impact on prognosis, such as complex congenital heart disease or malformation syndromes. Neonatal sepsis was assumed if positive blood cultures were present. We defined major brain injury as periventricular/intraventricular haemorrhage (PIVH) ≥ grade 3 12 and/or cystic periventricular leucomalacia (PVL) 13. Diagnosis of moderate or severe bronchopulmonary dysplasia (BPD) was based on the National Institute of Child Health and Human Development consensus definition as a requirement for supplemental oxygen and/or mechanical respiratory support at 36 weeks of postmenstrual age 14. Severe retinopathy of prematurity (ROP) was defined as ≥ stage 3 as suggested by the International Committee for the Classification of Retinopathy of Prematurity 15. Necrotising enterocolitis (NEC) was diagnosed in the presence of at least intestinal pneumatosis and/or portal venous gas (Bell's stage ≥2) 16.
Outcome definition
The main outcome assessed was survival to discharge without severe neonatal morbidity, in other words without major brain injury, moderate or severe BPD, ROP ≥ stage 3 or NEC ≥ stage 2.
Time points of risk assessment
Outcome predictors were evaluated at delivery, on NICU admission and on the seventh day of life. Infants who had already died at a previous assessment time point were excluded from the following assessments.
Statistical analysis
Missing data and multiple imputation
Missing data were more frequent in nonsurvivors than in survivors, raising concerns about overestimating survival and other bias. To prevent casewise deletion of infants from analyses, we imputed missing predictor values using the multivariate normal model and generated 10 completed data sets, resulting in imputed variables of binary variables taking on noninteger values, which were then carried forward without performing rounding 17, 18. Data were imputed for survivors and nonsurvivors. All nonmissing predictors and outcomes were included in the imputation model.
Predictors
Predefined predictors with high a priori plausibility were evaluated. The following prenatal predictors were tested at all time points: GA, z‐score for birthweight, gender, singleton/multiple birth, ANC, mode of delivery, major congenital malformation and the medical centre where the infants were cared for. In addition, sepsis within the first seven days of life was evaluated at the seventh day of life.
Model assessment and selection
The following candidate logistic regression models for the binary outcomes of survival to discharge without severe neonatal morbidity were evaluated for each time point:
GA as the only predictor
All considered predictors except medical centre
Applying backward elimination to model B
Model B including medical centre
Applying backward elimination to model D.
Backward elimination was performed by dropping nonsignificant predictors (p value >0.05) one at a time until all the predictors had a p value of <0.05.
The continuous predictors GA and z‐score for birthweight were tested for departure from linearity. Nonlinearity was addressed by including GA as a categorical variable, by weeks, and birthweight by categorising the z‐score. Following the usual convention, z‐score for birthweight was split into ten 0.5‐range categories from <−2 to more than +2.
We chose one primary model from A to E for each time point using the following procedure: the cross‐validated c‐statistic was calculated across the 10 multiple imputed data sets. This resulted in 10 c‐statistics for each model, which were combined using Rubin's rule to provide the final cross‐validated c‐statistic and confidence interval (CI) for each model 19. For each time point, the model with the highest cross‐validated c‐statistic was chosen as the final prediction model. Calibration of the models was assessed with plots.
The models were screened for statistically significant interactions, using the p value of the Wald test for the interaction term <0.05. This showed that including any of the interaction terms did not improve the cross‐validated c‐statistics. However, to examine centre‐to‐centre difference based on GA groups, we fitted the final models to include an interaction term for GA groups (<25 weeks, 25–27 weeks or >27 weeks) and medical centre.
Prediction
We calculated predicted survival probabilities and their 95% CIs from the combined estimates of the multiple imputed data sets by applying Rubin's rule 19. All statistical analyses were performed using Stata 13 (Stata Statistical Software: Release 13; StataCorp LP, College Station, TX, USA).
Results
Between January 1, 2002, and December 31, 2011, 6892 infants with a GA of between 23 and 31 completed weeks were cared for at the nine level III perinatal centres in Switzerland, and 5854 (85%) infants survived. Of all the deaths, 394 (38%) occurred in the delivery room and 644 (62%) occurred in the NICU: 405 (63%) of the NICU deaths occurred early, within seven days of admission, and 239 (37%) occurred after the seventh day of life. Of the 5854 survivors, 4905 (84%) survived the neonatal period without any severe neonatal complications. Table 1 summarises the characteristics of the patient population.
Table 1.
Characteristics of study population
| Survivors 5854 | Missing % (n) | Nonsurvivors 1038 | Missing % (n) | |
|---|---|---|---|---|
| Predictors | ||||
| Gestational age in weeks and days, Mean ± SD | 29 2/7 ± 14 | 0 (0) | 26 0/7 ± 16 | 0 (0) |
| Birthweight in grams, Mean ± SD | 1240 ± 370 | 0 (0) | 840 ± 410 | 1 (10) |
| Female, % (n) | 46 (2707) | 0 (0) | 45 (464) | 0.1 (1) |
| ANC, % (n) | 82 (4823) | 4 (234) | 52 (537) | 13 (138) |
| Caesarean section, % (n) | 78 (4595) | 0.14 (8) | 54 (565) | 6 (62) |
| Major congenital malformation, % (n) | 3 (201) | 0 (0) | 16 (164) | 0 (0) |
| Multiples, % (n) | 34 (1966) | 0 (0) | 25 (263) | 0 (0) |
| Inborn, % (n) | 94 (5479) | 0 (0) | 95 (982) | 1 (12) |
| Sepsis within first seven days of lifea, % (n) | 3 (166) | 2 (117) | 6 (58) | 23 (236) |
| Any missing predictor, % (n) | 6 (353) | 29 (300) | ||
| Outcomes, % (n) | ||||
| Died in delivery room | NA | NA | 38 (394) | 0 (0) |
| Died after admission to NICU but before day of life seven | NA | NA | 39 (405) | 0 (0) |
| Died after day of life seven but before discharge | NA | NA | 23 (239) | 0 (0) |
| Moderate or severe BPD | 9 (548) | 0.2 (9) | NA | NA |
| PIVH ≥ III and/or cPVL | 6 (323) | 0.1 (7) | NA | NA |
| ROP ≥ stage 3 | 1.8 (103) | 0 (0) | NA | NA |
| NEC | 1.8 (108) | 0 (0) | NA | NA |
| Any severe neonatal morbidityb | 16 (949) | 0.3 (16) | NA | NA |
ANC = Antenatal corticosteroids; BPD = Bronchopulmonary dysplasia; cPVL = Cystic periventricular leucomalacia; NEC = Necrotising enterocolitis; NICU = Neonatal intensive care unit; PIVH = Periventricular/intraventricular haemorrhage; ROP = Retinopathy of prematurity.
For infants alive on day of life seven (n = 5705).
Any severe neonatal morbidity: moderate/severe BPD and/or PIVH ≥ grade III and/or cPVL and/or proven NEC and or ROP ≥ stage III.
The cross‐validated c‐statistics of the five models for each time point are listed in Table S1. The models were used for predictions after applying backward elimination for prenatal and postnatal predictors, including the medical centre. Figure S1 shows the calibration plots.
Table 2 lists the logistic regression models with the effects of the various predictors on survival without severe neonatal morbidity.
Table 2.
Logistic models of the multiple imputed data for different time points
| Outcome | Survival without severe neonatal morbiditya | ||
|---|---|---|---|
| Time point | At delivery | On NICU admission | On day of life seven |
| OR (95% CI) | OR (95% CI) | OR (95% CI) | |
| Gestational age | |||
| 23 weeks | Reference | Reference | Reference |
| 24 vs 23 weeks | 13.7 (3.3–57.4) | 2.2 (0.61–8.0) | 2.6 (0.64–10.3) |
| 25 vs 24 weeks | 3.2 (2.2–4.6) | 2.3 (1.54–3.3) | 1.93 (1.27–2.9) |
| 26 vs 25 weeks | 2.3 (1.71–2.97) | 2.2 (1.71–3.0) | 2.0 (1.48–2.7) |
| 27 vs 26 weeks | 1.69 (1.32–2.2) | 1.65 (1.30–2.1) | 1.55 (1.19–2.0) |
| 28 vs 27 weeks | 1.74 (1.36–2.2) | 1.70 (1.34–2.2) | 1.61 (1.25–2.1) |
| 29 vs 28 weeks | 1.29 (1.01–1.66) | 1.31 (1.03–1.67) | 1.24 (0.96–1.60) |
| 30 vs 29 weeks | 1.67 (1.3–2.14) | 1.63 (1.28–2.1) | 1.66 (1.28–2.16) |
| 31 vs 30 weeks | 1.79 (1.37–2.3) | 1.68 (1.30–2.2) | 1.67 (1.26–2.2) |
| BW (z‐score) | |||
| <−2 | 0.16 (0.10–0.24) | 0.17 (0.11–0.25) | 0.22 (0.14–0.35) |
| −2 to −1.5 | 0.37 (0.27–0.51) | 0.39 (0.28–0.53) | 0.46 (0.33–0.66) |
| −1.5 to 1 | 0.51 (0.39–0.65) | 0.52 (0.40–0.67) | 0.57 (0.43–0.76) |
| −1 to −0.5 | 0.69 (0.55–0.86) | 0.70 (0.56–0.86) | 0.76 (0.60–0.96) |
| −0.5 to 0 | 0.94 (0.77–1.15) | 0.94 (0.77–1.14) | 0.92 (0.74–1.15) |
| 0 to 0.5 | Reference | Reference | Reference |
| 0.5 to 1 | 1.11 (0.88–1.39) | 1.08 (0.86–1.34) | 1.04 (0.82–1.33) |
| 1 to 1.5 | 1.27 (0.91–1.77) | 1.29 (0.94–1.77) | 1.34 (0.93–1.93) |
| 1.5 to 2 | 0.84 (0.50–1.41) | 0.96 (0.57–1.61) | 1.06 (0.57–1.96) |
| >2 | 0.14 (0.07–2.63) | 0.22 (0.11–0.45) | 0.69 (0.22–2.2) |
| Prenatal steroids | 1.60 (1.32–1.94) | 1.4 (1.17–1.70) | – |
| Female | 1.39 (1.21–1.59) | 1.44 (1.26–1.65) | 1.47 (1.27–1.69) |
| Caesarean section | – | – | – |
| Multiples | 1.17 (1.01–1.36) | 1.22 (1.06–1.41) | 1.27 (1.08–1.49) |
| Birth defect | 0.19 (0.15–0.25) | 0.24 (0.18–0.31) | 0.43 (0.31–0.60) |
| Sepsis first seven days of life | NA | NA | 0.49 (0.35–0.70) |
| Centre | |||
| 1 | Reference | Reference | Reference |
| 2 | 1.68 (1.32–2.1) | 1.54 (1.22–1.93) | 1.34 (1.05–1.72) |
| 3 | 2.0 (1.50–2.7) | 1.89 (1.42–2.51) | 1.66 (1.21–2.3) |
| 4 | 1.26 (0.87–1.83) | 1.14 (0.80–1.63) | 1.02 (0.71–1.51) |
| 5 | 1.67 (1.33–2.1) | 1.38 (1.11–1.71) | 1.17 (0.92–1.48) |
| 6 | 2.1 (1.55–2.8) | 1.89 (1.43–2.5) | 1.45 (1.08–1.94) |
| 7 | 2.1 (1.68–2.55) | 1.89 (1.53–2.3) | 1.48 (1.12–1.86) |
| 8 | 2.8 (2.1–3.6) | 2.6 (2.0–3.5) | 2.1 (1.58–2.9) |
| 9 | 4.5 (3.4–6.0) | 3.9 (3.0–5.3) | 3.4 (2.5–4.6) |
ANC = Antenatal corticosteroids; BPD = Bronchopulmonary dysplasia; cPVL = Cystic periventricular leucomalacia; NEC = Necrotising enterocolitis; NICU = Neonatal intensive care unit; PIVH = Periventricular/intraventricular haemorrhage; ROP = Retinopathy of prematurity.
Severe neonatal morbidity: moderate/severe BPD and/or PIVH ≥ grade III/IV intracranial haemorrhage and/or cPVL and/or proven NEC and/or ROP ≥ stage III.
The medical centre where care was provided was statistically a highly significant predictor of outcome after adjustment for prenatal and postnatal predictors. For example, at delivery, an infant born at medical centre nine, which had the highest overall survival rate, had an odds ratio (OR) for survival without severe neonatal morbidity of 4.5 (95% CI 3.4–6.0), compared to reference centre one, which had the lowest overall survival rate. This exceeded the effect of being born at 25 weeks vs 24 weeks (OR 3.2, 95% CI 2.2–4.6) or even being born at 27 weeks vs 25 weeks (OR 3.8, 95% CI 2.9–5.0). This association with the centre where the infants were treated persisted on NICU admission (OR 3.9, 95% CI 3.0–5.3) and on the seventh day of life after an additional adjustment for proven sepsis during the first week of life (OR 3.4, 95% CI 2.5–4.6).
When we used the model that allowed interactions between GA groups and centres, the OR for survival without severe morbidity at delivery for an infant born <25 weeks in centre nine compared to centre one was 4.5 (CI 2.5–8.1). The OR for an infant born between 25 and 27 weeks was 5.3 (CI 3.0–9.3), and for an infant born >28 weeks, it was 3.3 (CI 2.2–4.9). The corresponding ORs for an infant on admission to the NICU were 3.4 (CI 1.89–6.5), 4.6 (CI 2.6–7.9) and 3.0 (CI 2.1–4.4), respectively.
Gestational age‐specific prediction estimates and CIs for all time points were calculated for singleton infants without severe malformations with favourable perinatal predictors, namely z‐score for birthweight of +0.5 to 1, female gender, ANC, and, for the model on day of life seven, no sepsis during first week of life and compared to singleton infants without severe malformations and unfavourable perinatal predictors, namely z‐score for birthweight of −1 to −0.5, male gender, no ANC, and, for the model on the seventh day of life, proven sepsis during first week of life.
Figure 1a–c and the corresponding Table S2a–c illustrate the above‐mentioned predictions for the entire country, using the model without the centre as a predictor, and the centres with the highest and lowest predicted rates of survival without severe neonatal morbidity – centres one and nine – for the three time points. Findings were similar if only survival to discharge was examined (Fig. 2a–c, Table S3a–c).
Figure 1.
Prediction of survival without severe neonatal morbidity at the time of birth (a), at the time of NICU admission (b) and on day of life seven (c) (favourable risk factors: z‐score for BW +0.5 to 1, female, antenatal corticosteroids (ANC), and, for the model on day of life seven, no proven sepsis during first week of life; unfavourable risk factors: z‐score for BW −0.5 to −1, male, no ANC, and, for the model on day of life seven, proven sepsis during first week of life).
Figure 2.
Prediction of survival to discharge at the time of birth (a), at the time of NICU admission (b) and on day of life seven (c) (favourable risk factors: z‐score for BW +0.5 to 1, female, antenatal corticosteroids (ANC), and, for the model on day of life seven, no proven sepsis during first week of life; unfavourable risk factors: z‐score for BW −0.5 to −1, male, no ANC, and, for the model on day of life seven, proven sepsis during first week of life).
As GA increased, the impact of perinatal risk factors, as well as medical centre on survival and survival without severe neonatal morbidity, became less pronounced, although differences persisted up to 31 weeks of gestation.
Discussion
Our study quantifies the impact of different prenatal and postnatal risk factors and medical centre on the probability of survival to discharge without any severe neonatal morbidity in a large national cohort of VLGANs in Switzerland.
Our three prediction models illustrate how prognosis changed rapidly throughout the first week of life. At the lowest GAs – 23 and 24 completed weeks – the chances of survival without severe neonatal morbidity increased substantially for those infants who were admitted to a NICU (Fig. 1a,b). This is not surprising, as delivery room deaths in borderline viable infants are usually the result of primary nonintervention rather than failed resuscitation. At the limit of viability, such an approach invariably leads to death 20, 21, 22. Despite the availability of national guidelines or legal definitions of human viability, considerable centre‐to‐centre differences in the initiation of life‐sustaining therapies at the limit of viability were documented in many countries and are likely to reflect a particular NICU's culture 23.
In the present study, chances for survival without severe neonatal morbidity were substantially reduced for all VLGANs who sustained an episode of sepsis during the first week of life (Table 2, Fig. 1c).
Finally, our study quantifies centre‐to‐centre outcome variability in this vulnerable population in Switzerland (Figs 1a–c and 2a–c, Tables S2a–c and S3a–c). The fact that centre‐to‐centre differences existed for both survival to discharge without severe neonatal morbidity and survival to discharge suggests that more proactive treatment approaches resulted in higher survival rates without increasing short‐term morbidity rates. Over the last decade, several publications have addressed centre‐to‐centre differences in multiple populations and networks 4, 5, 7, 24, 25. Three factors provide possible explanations for centre differences: first, the inherent risk of the patient population served by a centre; second, the approach of any given centre towards primary nonintervention in infants born at the border of viability; and third, the effectiveness of the patient care delivered at a particular centre. In our study, the association between medical centre and survival without severe neonatal morbidity remained substantial after adjusting for important patient‐level factors (Table 2). This is consistent with centre‐to‐centre outcome differences reported by other networks that could not be explained by differences in patient demographics 5, health insurance coverage 4 or hospital case load 26, 27. These findings suggest that differences in centre‐specific neonatal practices strongly influence outcome 24, 27, 28.
The magnitude of the centre‐to‐centre outcome differences in our study is comparable to the one described by Lee et al. 4 for Canadian NICUs and by Vohr et al. 5 for the National Institutes for Child Health and Development Neonatal Research Network (NICHD NRN) centres. Alleman et al. 24 confirmed these findings for a more recent cohort of extremely low‐birthweight infants cared for by the NICHD NRN and found that centre intervention rates were statistically significant, but only when it came to predicting mortality for infants <25 weeks GA. This contrasts with the results of our study, where centre‐to‐centre outcome variability extended beyond the most immature infants, and primary noninterventions in infants born at the border of viability did not entirely explain the observed centre‐to‐centre differences. A study by Smith et al. from the NICHD NRN found similar results to our study. They reported that that centres with a more aggressive approach to care for infants born at 22–24 weeks also had reduced rates of death, death or ROP, death or NEC, death or late‐onset sepsis, and death or neurodevelopmental impairment for more mature infants born at 25–27 weeks 29.
Limitations
Because data were more frequently missing in nonsurvivors, casewise deletion of infants with missing data from the prediction models would have overestimated survival probabilities. We used multiple imputation to reduce bias due to missing data, but our estimate of the associations could still be biased if missing data depended not only on the variables we used to impute missing values, but also on the unknown missing values themselves 30. A 10‐fold cross‐validation was used to internally validate the models and avoid overfitting, but the models were not externally validated. Furthermore, we cannot exclude the fact that some of the centre‐to‐centre differences were a chance finding due to multiple comparisons between the nine NICUs. Our study does not provide information on long‐term neurodevelopmental impairment. However, severe neonatal morbidities, such as major brain injuries, moderate/severe BPD, proven NEC and severe ROP, are known to correlate with long‐term neurodevelopmental outcomes 31, 32. Finally, the study design did not allow us to analyse trends over time.
Conclusion
There was strong evidence for wide centre‐to‐centre outcome variability in rates of survival without severe neonatal morbidity among VLGANs in Switzerland over a time period of 10 years that could not be explained by patient‐level factors. The observed outcome differences were not restricted to infants born at the limit of viability, but extended to more mature infants and persisted after the first week of life. Identifying and implementing potentially better practices in perinatal care may influence important outcomes.
Funding
No specific funding was received for this study.
Conflict of Interest
The authors declare no conflict of interest in relation to the manuscript.
Supporting information
Figure S1 Calibration plots for the final models for predicting survival without severe neonatal morbidity at different time points.
Table S1 10‐fold cross‐validated c‐statistics for different time points and models using the multiple imputed data set.
Table S2 (a–c) Predictors for survival without any severe neonatal morbidity* at the time of delivery (a), at the time of NICU admission (b) and on the seventh day of life (c) for inborn singleton infants without major birth defects.
Table S3 (a–c) Predictors for survival to discharge at the time of delivery (a), at the time of NICU admission (b) and on the seventh day of life (c) for inborn singleton infants without major birth defects.
Acknowledgements
We would like to thank the participating centres and collaborators: Aarau: Kantonsspital Aarau, Kinderklinik (Philipp Meyer‐Schiffer); Basel: Universitäts‐Kinderspital beider Basel, Abteilung für Neonatologie (Sven M. Schulzke); Bern: Inselspital, Abteilung für Neonatologie (Mathias Nelle), and Universitätskinderklinik, Abteilung für Intensivpflege (Bendicht Wagner); Chur: Kantonsspital Graubünden, Kinderklinik (Walter Bär); Genève: Hôpitaux Universitaires de Genève (HUG), Département de l'Enfant et de l'Adolescent, Service de Néonatologie et des Soins Intensifs de Pédiatrie (Riccardo E. Pfister); Lausanne: CHUV, Service de Néonatologie (Jean‐François Tolsa), and CHUV, SI Pédiatriques, Médico‐Chirurgicaux (Jacques Cotting); Luzern: Kinderspital Luzern, neonatologische und pädiatrische Intensivstation (Thomas M. Berger); St. Gallen: Kantonsspital St. Gallen, Klinik für Geburtshilfe und Gynäkologie (Andreas Malzacher), and Ostschweizer Kinderspital, Intensivpflege‐ und Frühgeburtenstation (John P. Micallef); and Zürich: USZ, Klinik für Neonatologie, Departement Frauenheilkunde (Hans‐Ulrich Bucher), and Universitätskinderklinik Zürich, Abteilung für Intensiv‐medizin und Neonatologie (Bernhard Frey).
References
- 1. Tyson JE, Parikh NA, Langer J, Green C, Higgins RD. National Institute of Child Health and Human Development Neonatal Research Network. Intensive care for extreme prematurity ‐moving beyond gestational age. N Engl J Med 2008; 358: 1672–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Cole TJ, Hey E, Richmond S. The PREM score: a graphical tool for predicting survival in very preterm births. Arch Dis Child Fetal Neonatal Ed 2009; 95: F14–9. [DOI] [PubMed] [Google Scholar]
- 3. Medlock S, Ravelli ACJ, Tamminga P, Mol BWM, Abu‐Hanna A. Prediction of mortality in very premature infants: a systematic review of prediction models. PLoS One 2011; 6: e23441. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Lee SK, McMillan DD, Ohlsson A, Pendray M, Synnes A, Whyte R, et al. Variations in practice and outcomes in the Canadian NICU network: 1996–1997. Pediatrics 2000; 106: 1070–9. [DOI] [PubMed] [Google Scholar]
- 5. Vohr BR, Wright LL, Dusick AM, Perrit R, Poole WK, Tyson JE, et al. Centre differences and outcomes of extremely low birth weight infants. Pediatrics 2004; 113: 781–9. [DOI] [PubMed] [Google Scholar]
- 6. Simpson JM, Evans N, Gibberd RW, Heuchan AM, Henderson‐Smart DJ, Australian and New Zealand Neonatal Network . Analysing differences in clinical outcomes between hospitals. Qual Saf Health Care 2003; 12: 257–62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Serenius F, Sjörs G, Blennow M, Fellman V, Holmström G, Maršál K, et al. EXPRESS study shows significant regional differences in one‐year outcome of extremely preterm infants in Sweden. Acta Paediatr 2014; 103: 27–37. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Berger TM, Bernet V, El Alama S, Fauchère JC, Hösli I, Irion O, et al. Perinatal care at the limit of viability between 22 and 26 completed weeks of gestation in Switzerland. 2011 revision of the Swiss recommendations. Swiss Med Wkly 2011; 141: w13280. [DOI] [PubMed] [Google Scholar]
- 9. Berger TM, Steurer MA, Woerner A, Meyer‐Schiffer P, Adams M; Swiss Neonatal Network . Trends and centre‐to‐centre variability in survival rates of very preterm infants (<32 weeks) over a 10‐year‐period in Switzerland. Arch Dis Child Fetal Neonatal Ed 2012; 97: F323–8. [DOI] [PubMed] [Google Scholar]
- 10. Fischer N, Steurer MA, Adams M, Berger TM, Swiss Neonatal Network . Survival rates of extremely preterm infants (gestational age <26 weeks) in Switzerland: impact of the Swiss guidelines for the care of infants born at the limit of viability. Arch Dis Child Fetal Neonatal Ed 2009; 94: F407–13. [DOI] [PubMed] [Google Scholar]
- 11. Voigt M, Fusch C, Olbertz D, Hartmann K, Rochow N, Renken C, et al. Analyse des Neugeborenenkollektivs der Bundesrepublik Deutschland. Geburtsh Frauenheilk 2006; 66: 956–70. [Google Scholar]
- 12. Papile LA, Munsick‐Bruno G, Schaefer A. Relationship of cerebral intraventricular hemorrhage and early childhood neurologic handicaps. J Pediatr 1983; 103: 273–7. [DOI] [PubMed] [Google Scholar]
- 13. de Vries LS, Eken P, Dubowitz LM. The spectrum of leukomalacia using cranial ultrasound. Behav Brain Res 1992; 49: 1–6. [DOI] [PubMed] [Google Scholar]
- 14. Bancalari E, Jobe AH. Bronchopulmonary dysplasia. Am J Respir Crit Care Med 2001; 163: 1723–9. [DOI] [PubMed] [Google Scholar]
- 15. International Committee for the Classification of Retinopathy of Prematurity . The international classification of retinopathy of prematurity revisited. Arch Ophthalmol 2005; 123: 991–9. [DOI] [PubMed] [Google Scholar]
- 16. Bell MJ, Ternberg JL, Feigin RD, Keating JP, Marshall R, Barton L, et al. Neonatal necrotizing enterocolitis. Therapeutic decisions based upon clinical staging. Ann Surg 1978; 187: 1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Horton NJ, Lipsitz SR, Parzen M. A potential for bias when rounding in multiple imputation. Am Stat 2003; 57: 229–32. [Google Scholar]
- 18. Little RJA, Rubin BR. Statistical analysis with missing data, 2nd ed Hoboken, NJ: Wiley, ISBN: 978‐0‐471‐18386‐0. September 2002. [Google Scholar]
- 19. Marshall A, Altman DG, Holder RL, Royston P. Combining estimates of interest in prognostic modelling studies after multiple imputation: current practice and guidelines. BMC Med Res Methodol 2009; 9: 57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Verhagen AA, Janvier A, Leuthner SR, Andrews B, Lagatta J, Bos AF, et al. Categorizing neonatal deaths: a cross‐cultural study in the United States, Canada, and the Netherlands. J Pediatr 2010; 156: 33–7. [DOI] [PubMed] [Google Scholar]
- 21. Berger TM, Hofer A. Causes and circumstances of neonatal deaths in 108 consecutive cases over a 10‐year period at the Children's Hospital of Lucerne, Switzerland. Neonatology 2009; 95: 157–63. [DOI] [PubMed] [Google Scholar]
- 22. Arlettaz R, Mieth D, Bucher HU, Duc G, Fauchère J‐C. End‐of‐life decisions in delivery room and neonatal intensive care unit. Acta Paediatr 2005; 94: 1626–31. [DOI] [PubMed] [Google Scholar]
- 23. Berger TM. Decisions in the gray zone: evidence‐based or culture‐based? J Pediatr 2010; 156: 7–9. [DOI] [PubMed] [Google Scholar]
- 24. Alleman BW, Bell EF, Li L, Dagle JM, Smith PB, Ambalavanan N, et al. Individual and centre‐level factors affecting mortality among extremely low birth weight infants. Pediatrics 2013; 132: e175–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Horbar JD, Badger GJ, Lewit EM, Rogowski J, Shiono PH. Hospital and patient characteristics associated with variation in 28‐day mortality rates for very low birth weight infants. Pediatrics 1997; 99: 149–56. [DOI] [PubMed] [Google Scholar]
- 26. Rogowski JA, Horbar JD, Staiger DO, Kenny M, Carpenter J, Geppert J. Indirect vs direct hospital quality indicators for very low‐birth‐weight infants. JAMA 2004; 291: 202–9. [DOI] [PubMed] [Google Scholar]
- 27. Kusuda S, Fujimura M, Sakuma I, Aotani H, Kabe K, Itani Y, et al. Morbidity and mortality of infants with very low birth weight in Japan: centre variation. Pediatrics 2006; 118: e1130–8. [DOI] [PubMed] [Google Scholar]
- 28. Stoll BJ, Hansen NI, Bell EF, Shankaran S, Laptook AR, Walsh MC, et al. Neonatal outcomes of extremely preterm infants from the NICHD Neonatal Research Network. Pediatrics 2010; 126: 443–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Smith PB, Ambalavanan N, Li L, Cotten CM, Laughon M, Walsh MC, et al. Approach to infants born at 22 to 24 weeks' gestation: relationship to outcomes of more‐mature infants. Pediatrics 2012; 129: e1508–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Cummings P. Missing data and multiple imputation. JAMA Pediatr 2013; 167: 656–61. [DOI] [PubMed] [Google Scholar]
- 31. Bassler D, Stoll BJ, Schmidt B, Asztalos EV, Roberts RS, Robertson CM, et al. Using a count of neonatal morbidities to predict poor outcome in extremely low birth weight infants: added role of neonatal infection. Pediatrics 2009; 123: 313–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Schmidt B, Asztalos EV, Roberts RS, Robertson CM, Sauve RS, Whitfield MF, et al. Impact of bronchopulmonary dysplasia, brain injury, and severe retinopathy on the outcome of extremely low‐birth‐weight infants at 18 months: results from the trial of indomethacin prophylaxis in preterms. JAMA 2003; 289: 1124–9. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Figure S1 Calibration plots for the final models for predicting survival without severe neonatal morbidity at different time points.
Table S1 10‐fold cross‐validated c‐statistics for different time points and models using the multiple imputed data set.
Table S2 (a–c) Predictors for survival without any severe neonatal morbidity* at the time of delivery (a), at the time of NICU admission (b) and on the seventh day of life (c) for inborn singleton infants without major birth defects.
Table S3 (a–c) Predictors for survival to discharge at the time of delivery (a), at the time of NICU admission (b) and on the seventh day of life (c) for inborn singleton infants without major birth defects.