European Journal of Epidemiology

, Volume 22, Issue 11, pp 791–798

Obstetrical volume and early neonatal mortality in preterm infants

Authors

    • Department of Epidemiology, Public Medicine and Healthcare Systems ResearchHannover Medical School
  • Paul Wenzlaff
    • Center for Quality and Management in Health Care
  • Christian F. Poets
    • Department of NeonatologyUniversity of Tübingen
Perinatal Epidemiology

DOI: 10.1007/s10654-007-9182-6

Cite this article as:
Bartels, D.B., Wenzlaff, P. & Poets, C.F. Eur J Epidemiol (2007) 22: 791. doi:10.1007/s10654-007-9182-6

Abstract

Objective Regionalised perinatal care with antenatal transfer of high risk pregnancies to Level III centres is beneficial. However, levels of care are usually not linked to caseload requirements, which remain a point for discussion. We aimed to investigate the impact of annual delivery volume on early neonatal mortality among very preterm births. Methods All neonates with gestational age 24–30 weeks, born 1991–1999 in Lower Saxony were included into this population-based cohort study (n = 5,083). Large units were defined as caring for more than 1,000 deliveries/year, large NICUs as those with at least 36 annual very low birthweight (<1,500 g, VLBW) admissions. Main outcome criterion was mortality until day 7. Adjusted Odds Ratios (adj. OR) and 95% confidence intervals (CI) were calculated based on generalised estimating equation models, accounting for correlation of individuals within units. Results Within the first week of life, 20.6% of all neonates deceased; 10.2% were stillbirths, 3.7% died in the delivery unit, and 6.7% in the NICU. The crude OR for early neonatal mortality after having been delivered in a small delivery unit (excluding stillbirths) was 1.36 (95%CI 1.04–1.78; adj. OR 1.16 (0.82–1.63)). It increased to 1.96 (1.54–2.48; adj. OR 1.21 (0.86–1.70)) after the inclusion of stillbirths. Conclusion This study has shown a slight, but non-significant association between obstetrical volume and early neonatal mortality. In future studies the impact of caseload on outcome may become more evident when referring to high-risk patients instead of to the overall number of deliveries.

Keywords

Early neonatal mortalityHospital volumePerinatal carePreterm birthQuality of care

Abbreviations

Adj.

Adjusted

AGA

Appropriate weight for gestational age

CI

Confidence interval

GA

Gestational age

GEE

General estimating equation

LGA

Large for gestational age

MV

Missing value

NICU

Neonatal intensive care unit

OR

Odds ratio

RDS

Respiratory distress syndrome

SGA

Small for gestational age

VLBW

Very low birthweight (<1,500 g)

Introduction

Pregnancy may take a hazardous course for either mother or child. Thanks to advancements in prenatal care and ultrasound technology, most risks are now detectable early, and timely referral to a perinatal centre is often possible, with beneficial effects on perinatal outcome [1]. The structural characteristics of the tertiary perinatal centre, however, may also influence outcome and prognosis of mother and child. In this regard, it has been shown that hospital volume can serve as a surrogate quality criterion for e.g. experience, staff, multidisciplinarity, and 24 h physician availability, which may all affect patient outcome [2]. Most countries have defined levels of perinatal care based on staff, equipment and risk profiles of mother and foetus, but not on hospital volume [3]. Since the 1950s, national professional organisations and regulatory authorities have considered recommended minimum numbers of annual hospital deliveries. Discussed cut-offs range from 1,500 to 2,000 annual deliveries [4], but have not yet been implemented consistently. Moreover, delivery unit and NICU are usually combined within perinatal centres, thus the individual impact of the delivery vs. neonatal unit volume cannot be distinguished [57].

In Germany, delivery units and NICUs are still detached in most hospitals, and only last year the government introduced definitions of level of care. Recently, we have shown an 80% increased mortality risk in very premature infants mainly cared for in small NICUs, and a stronger effect of NICU than delivery unit volume. However, the latter result may have been biased due to exclusion of newborns not admitted to NICU [8]. Therefore, we further analysed the effect of delivery unit size, now including all deliveries and focusing on early neonatal mortality. The question under study was whether size of the delivery unit has an impact on early neonatal mortality of very premature neonates. Showing volume effects for both, delivery unit and NICU, would raise the question of including minimum caseload requirements into definitions of levels of perinatal care in the future.

Methods

We used linked databases from a quality assurance programme in the German state of Lower Saxony [9], one comprising obstetric data from 97% of all deliveries and one comprising neonatal data for all infants admitted to NICUs within 10 postnatal days. Details on data completeness and validity are described elsewhere [8, 10]. The following inclusion criteria were defined:
  • Born 1/1/1991–31/12/1999 in Lower Saxony (n = 7,745)

  • Valid gestational age (GA) data (n = 7,737)

  • Successful data linkage/no transferral to neighbouring states (n = 7,699)

  • Gestational age 24–30 weeks (n = 5,083)

German guidelines recommend life-support for all neonates with gestational age of at least 24 weeks. The outcome criterion was death within the first 7 days of life. Stillbirths usually occur before admission, and thus may not be a proper outcome criterion when investigating quality of care, in particular if antenatal care is usually provided outside the hospital—as it is in Germany. However, in retrospective, large population based studies such as ours, a precise documentation distinguishing ante- from intrapartum stillbirth is often questioned [11]. Hence we performed separate analyses, once including only livebirths (n = 4,566), once including also stillbirths (n = 5,083).

Variables

The volume cut-offs between ‘low’ and ‘high’ for delivery units were ≤/>1,000 births p.a., for NICUs </≥ 36 VLBW admissions p.a. Among the 107 delivery units, 79 were small, 28 were large. Among the 36 NICUs, between 26 and 29 were small and 7 to 10 were large. Ranges were due to three NICUs changing from small to large during the observation period. The small number of large hospitals within the state justifies these low cut-offs used for many years, otherwise no valid statistical modelling would have been possible. The following numbers elucidate this limitation in more detail: In Lower Saxony, 27 delivery units deliver less than 500 women annually, 34 units deliver 500 to 750 women, i.e. 57% of all units deliver less than 750 women, and only three units 1,500 or more. Nearly half of the NICUs (17 out of 36) have less than 20 VLBW-admissions/year, but they all have staff and equipment to provide comprehensive intensive care for even the most immature neonates. There is no 1:1 matching for delivery unit and NICU, because one NICU-team may collaborate with several delivery hospitals (but usually not vice versa). NICU size was used as a separate potential confounder variable, being an independent predictor for the outcome and associated with the exposure of interest, i.e. delivery size, to prevent biased results.

Having been ‘inborn’ means delivery and NICU admission within one perinatal centre, i.e. adjacency of delivery unit and NICU. Gestational age estimates (completed weeks, wk) were based on early ultrasound and coded as dummy variables, as the linearity assumption did not hold. The latter was also applicable for year of birth. Hospitalization days during pregnancy were categorised based on the median into <10 days vs. ≥10 days. Pregnancy at risk is a German standard definition for mothers e.g. >35 years of age, or mothers suffering from preeclampsia, diabetes, hypertension, or a fetus with IUGR, chromosomal abnormalities etc. Only smoking and non-smoking were distinguished, since data on number of cigarettes were not valid. Cervical-width at admission was classified into <4 cm vs. ≥4 cm. We considered as a potentially increased risk if the first ultrasound examination had not been performed within the first 10 weeks of pregnancy (median), if there were less than four such examinations, or if the number of appointments with an obstetrician during pregnancy had been lower than six. Time since last antenatal steroid administration was categorised on the clinically established cut-offs, as 24 h, 2–3 days, 4–7 days, 8–14 days, or >2 weeks. The five minute Apgar score was dichotomised based on the established threshold of <7 vs. ≥ 7. According to German population-based percentiles, distinguishing males/females and singletons/twins [12], small for gestational age (SGA) was categorised as birthweight below the 10th percentile, appropriate weight for gestational age (AGA) as between the 10th and 90th, and large for gestational age (LGA) as above the 90th percentile [13]. Twin percentiles were used for all multiples. Body temperature <36.0°C at NICU admission was labelled as hypothermia. We defined RDS as documentation of either this term or of exogenous surfactant administration, because early surfactant application can make a diagnosis of RDS impossible. Severe congenital malformations were classified according to the Weidtmann code [9] and implied life-threatening characteristics.

Statistics

Sex, GA, and multiplicity were used as forced-in variables in the complete modelling process for face validity reasons. In addition univariable analysis based on chisquare-, Fisher’s exact, Wilcoxon-tests were performed to identify potential confounders, i.e. factors being associated with both, delivery unit volume and early neonatal mortality, using a P-value of 0.20 [14]. Variables with more than 20% missing values (MV) were excluded. In total the following variables were analysed: Sex, GA, multiplicity, year of birth, non-German origin, single mother, employed during pregnancy, previous abortion/stillbirth, primiparity, maternal age, maternal body mass index, number of appointments with an obstetrician during pregnancy, ultrasound examinations, hospitalization days during pregnancy, pregnancy at risk, hypertension, diabetes, anaemia, smoking, cervical width at admission, placental insufficiency, (suspected) amniotic infection syndrome, maternal fever, delivery planned in this unit, fetal position, antenatal steroids, inborn, paediatrician present at birth, midwife present at birth, NICU volume, growth status (SGA, LGA), Apgar score, severe congenital malformations, hypothermia, RDS.

The multivariable modelling process was based on generalised estimating equation (GEE) models to estimate adjusted odds ratios (adj. OR) and corresponding 95% confidence intervals (CI) for mortality risk, accounting for correlation among individuals within units and for multiple births [15]. Initially starting with the forced-in variables and those selected from univariable analysis, the significance level was reduced in stages of 0.05 from 0.25 to 0.10 [14]. Model validity was assessed based on the c-statistic and on analyses of more complex models without resulting in significant changes [14]. The statistical software package SAS 9.1 (SAS Institute Inc., Cary, NC, USA) was used for all data analyses.

In Germany, anonymised secondary data research does not require IRB review.

Results

Of n = 5,083 infants with GA 24–30 weeks, n = 1,045 (20.6%) deceased before the end of the first week: n = 517 (10.2%) were stillbirths, n = 187 (3.7%) died in the delivery unit, and n = 341 (6.7%) in the NICU. Table 1 displays the population characteristics.
Table 1

Description of the study population, neonates born in Lower Saxony (Germany) 1991–1999 with GA 24–30 weeks, n = 5,083

Variables

 

N

%

Year of birth

1991

442

8.7

1992

470

9.3

1993

490

9.6

1994

532

10.5

1995

577

11.4

1996

639

12.6

1997

663

13

1998

626

12.3

1999

644

12.7

German

Yes

4266

83.9

No

333

6.6

MVa

484

9.5

Single mother

Yes

406

8

No

4,603

90.6

MV

74

1.5

Maternal age

<18

59

1.2

18–35

4,390

86.4

>36

634

12.5

Primiparity

Yes

2,875

56.6

No

2,207

43.4

MV

1

0.02

Mode of delivery

c-section

3,924

77.2

Vaginal

1,159

22.8

Maternal death

Yes

6

0.1

No

5,077

99.9

Gestational age (weeks)

24

388

7.6

25

515

10.1

26

601

11.8

27

708

13.9

28

927

18.2

29

1000

19.7

30

944

18.6

Gender

Male

2,632

51.8

Female

2,451

48.2

Multiple birth

Yes

1,370

27

No

3,713

73

Growth statusb

SGA

740

14.6

AGA

4,078

80.2

LGA

265

5.2

Death

No

3,842

75.6

Ante partum

517

10.2

Sub partu

46

0.9

≤7 days

482

9.5

8–28 days

137

2.7

>28 days

59

1.2

aMV indicates Missing value

bSGA—Small for gestational age; AGA—Appropriate weight for gestational age; LGA—Large for gestational age

In the high volume delivery units, significantly (P < 0.05) more woman were primiparities, singletons, smokers, and were older; they suffered more often from placental insufficiency, had their first ultrasound later and delivered more often SGA-neonates.

Analysis without stillbirths

Excluding all stillbirths gave a study population of n = 4,566, of whom n = 528 (11.6%) deceased within their first week of life. Deliveries occurred in n = 3,311 (72.5%) cases in large and in n = 1,255 (27.5%) cases in small hospitals. Within the large units, n = 362 (10.9%) infants died, within the small units n = 166 (13.2%); the crude OR (95% CI) was 1.36 (1.04–1.78). Besides the forced in variables, we selected from univariable analysis the variables year of birth, pregnancy at risk, primiparity, time of first ultrasound, number of ultrasound examinations, smoking, cervical width at admission, maternal hypertension, analgesia, antenatal steroids, inborn, severe congenital malformations, growth status, hypothermia, 5 min Apgar score, and NICU size. The modelling process finally gave an adj. OR of 1.16 (0.82–1.63) for the delivery unit size (Table 2). However, although considering only early neonatal mortality, NICU size still had a significant influence (adj. OR 1.61, 95% CI 1.05–2.46), whereas having been inborn did not remain significant during the modelling process.
Table 2

Final GEE model, excluding stillbirths, n = 4,566

  

Adj. OR

95% CI

Small delivery unit

1.16

0.82–1.63

Male

1.44

1.16–1.78

Gestational age

24 weeks

21.75

12.10–39.10

25 weeks

9.57

5.35–17.12

26 weeks

4.81

2.70–8.57

27 weeks

4.75

2.59–8.72

28 weeks

2.03

1.11–3.72

29 weeks

1.52

0.71–3.24

30 weeks

Ref.

 

Twin

1.83

1.44–2.33

≥triplet

2.07

0.87–4.91

Small for gestational age

3.58

2.22–4.29

Large for gestational age

0.93

0.62–1.39

Hypothermia

1.64

1.24–2.15

5 min Apgar <7

3.07

2.36–3.99

Small NICU

1.61

1.05–2.46

Antenatal steroids

0.74

0.55–0.98

Severe congenital malformations

6.22

3.57–10.85

c-statistics P = 0.835

Analysis under inclusion of stillbirths

Among the complete (n = 5,083) study population, n = 3,562 (70.1%) were delivered in large and n = 1,521 (29.9%) in small units. In large hospitals, n = 613 (17.2%) infants deceased, in small hospitals there were n = 432 (28.4%), crude OR 1.96 (1.54–2.48). In addition to the forced in variables, we started the modelling process with the variables year of birth, pregnancy at risk, primiparity, time of first ultrasound, number of ultrasound examinations, smoking, cervical width at admission, maternal hypertension, antenatal steroids, inborn, severe congenital malformations, growth status, hypothermia, 5 min Apgar score, and NICU size. The final GEE model resulted in an adj. OR of 1.21 (0.86–1.70). Whereas NICU-size had been dropped during the modelling process, the inborn variable remained in this model (Table 3).
Table 3

Final GEE model, including stillbirths, n = 5,083

  

Adj. OR

95% CI

Small delivery unit

1.21

0.86–1.70

Male

1.41

1.14–1.73

Gestational age

24 weeks

19.58

11.27–34.02

25 weeks

8.66

4.96–15.12

26 weeks

4.62

2.70–7.89

27 weeks

4.69

2.65–8.31

28 weeks

1.95

1.10–3.46

29 weeks

1.49

0.72–3.06

30 weeks

Ref.

 

Twin

1.79

1.42–2.26

≥triplet

2.01

0.93–4.34

Small for gestational age

2.96

2.14–4.09

Large for gestational age

0.95

0.64–1.41

Hypothermia

1.62

1.24–2.14

5 min Apgar <7

3.00

2.24–4.02

Antenatal steroids

0.75

0.58–0.97

Severe congenital malformations

6.29

3.60–10.99

Inborn

0.65

0.45–0.94

c-statistics P = 0.834

Discussion

In this population-based study we could not find a statistically significant association between delivery unit volume and early neonatal mortality among very preterm infants. For consistency with previous studies, the study population was first analysed only including livebirths [7, 1622]. Additionally, all analytical steps were repeated also including stillbirths, since distinguishing ante- and intrapartum death may often be arguable [23]. However, this did not affect results significantly.

How can these results be interpreted? First, it is conceivable that caseload of the delivery unit does truly not affect early mortality. Yet this conclusion would be in contrast to other studies [1, 5, 11, 24], although comparability is often difficult because of the adjacency of delivery unit and NICU level in most studies. Second, a possible association may not have been detected due to our low cut-off (1,000 deliveries/year), which is a standard cut-off in Germany since larger units are sparse. Studies on low-risk births showing associations between hospital volume and mortality used slightly [25] or significantly higher [4, 11] thresholds. One rather old study concluded that obstetrical volume added minimal explanatory power to level of nursery care, but only infants with birthweight >1,000 g were included [26]. Thus, neonates with the potentially largest benefit from high volume units had been excluded. In a study from South Carolina, levels of care were defined by both, availability of medical speciality and subspeciality care as well as by volume of annual deliveries (Level III ≥ 1,500 annual deliveries), and lower neonatal mortality and shorter hospital stays among VLBW-neonates were reported for Level III centres [1]. Third, minimal caseload/volume criteria for NICUs are often linked to a specific subpopulation at risk, such as gestational age <32 weeks [27], or birthweight 500–1,499 g [1, 6, 21, 28], based on the concept that neonates being at highest risk will benefit most from centralised perinatal care. In contrast, previous studies on the impact of obstetrical volume and perinatal outcome referenced to all deliveries [4, 11, 25, 26]. Figure 1 shows the large overlap between small and large delivery units in regard to annual numbers of VLBW-births, exemplarily used as one case of high-risk birth. The figure demonstrates that using the cut-off based on all annual deliveries does not distinguish hospitals with small and large numbers of high risk infants, and thus may be a meaningless categorization. It stands to reason to assume that only delivery units caring for many high risk mothers and neonates provide better care for these patients than do facilities having fewer such patients. Unfortunately, we could not investigate this hypothesis since only two to four (due to annual variations) units cared for e.g. >50 VLBW-deliveries/year. Moreover, comparisons with previous studies would have been impossible had high risk birth been chosen as reference group. However, future studies should investigate a potential association of caseload and outcome using categorizations based on high risk deliveries instead of the overall delivery number. Fourth, hospital volume only serves as a surrogate parameter, while underlying factors, such as daily census, numbers of trained obstetricians/neonatologists during day–night and weekend shifts, or the presence of midwives and anaethesiologists may play a more decisive role than the annual number of deliveries, which could not be accounted for in our study.
https://static-content.springer.com/image/art%3A10.1007%2Fs10654-007-9182-6/MediaObjects/10654_2007_9182_Fig1_HTML.gif
Fig. 1

Annual numbers of very low birthweight (VLBW) infants in low (≤1,000 deliveries/year) and high volume delivery units. The box in the lower part of the graph indicates overlap in numbers of annual VLBW-deliveries between low and high volume units if defined by overall numbers of annual deliveries

Besides conflicting research results, several other reasons may be responsible for the fact that implementation of minimal caseload in perinatal care has been discussed for decades, but has not been consistently realised yet [4, 29]. In contrast to surgical procedures, where caseload constraints exist, perinatal care is much more complex and comprises various medical specialists, such as anaesthetists, obstetricians, neonatologists. Moreover, quality of care in terms of structural prerequisites may not be necessarily consistent with quality of care in terms of outcome; upgrading technical equipment and staff does not necessarily improve quality of care and outcomes. Although health, well-being and future of patients should govern medical and public health actions, the considerable political impact of any reorganisation of perinatal care, and the consequences concerning the competition of market share are further non-negligible aspects.

However, the recent establishment of midlevel NICUs, e.g. in the US, has shown a shift of high risk births (i.e. VLBW-neonates) from high-level to midlevel NICUS which may have negative consequences on their outcome [1, 30]. Such potentially negative shifts could be avoided by minimal caseload requirements.

Having been cared for in small NICUs (<36 VLBW admissions/year) remained a risk factor for early neonatal mortality in the study population without stillbirths, although only short- term survival was considered. One possible explanation is that obstetrical and NICU volume are often correlated and volume effects of both admix, i.e. larger delivery units will collaborate with larger NICUs. In our study, 66% of infants born in small delivery units were also cared for in small NICUs, and 58% of infants born in large delivery units were also cared for in large NICUs. In addition, the quality of neonatal care including early resuscitation in the delivery room may also have played a decisive role.

In contrast to other studies [31], inborn status did not show a protective effect in the first modelling process (n = 4,566). Again, correlation with hospital volume might be a possible explanation: In small obstetrical units 53% of infants were inborn, in large units it was only 46%. Including stillbirths (n = 5,083), these proportions changed to 49% vs. 38%. Such distribution imbalances may have biased results.

Male gender, growth retardation, and severe congenital malformations are known risk factors for increased mortality [10, 32]. That they turned out as risk factors also in this study underlines the data validity.

Hypothermia and low Apgar scores are not causal risk factors. If they were solely interims between (un)known causal factors and the outcome of interest, adjustment for these factors would be inappropriate. However, these variables might also be considered as surrogates for unmeasured additional confounders. This justifies their integration into the multivariable modelling process. Nevertheless, we repeated the multivariable analysis without these two variables, which did not change results significantly.

In conclusion, although we could not detect a statistically significant relationship between obstetrical volume and early neonatal mortality, further analysis, e.g. based on caseloads referencing to high-risk patients, should be performed. Potential problems arising from implementing minimal caseload requirements should fade into the background, if such rules enhanced quality of perinatal care. Perinatal regionalisation should aim at providing high level care to the appropriate proportion of the population who needs it and benefits from it, rather than to as large a proportion of the population as possible.

Copyright information

© Springer Science+Business Media B.V. 2007