 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
a Departments of Obstetrics, Pulmonology and Neonatology, and Epidemiology, Social Medicine and Health System Research, Hannover Medical School, Hannover, Germany
b Children’s Hospital and Harvard School of Public Health, Boston, Massachusetts
c Center for Quality Assurance and Management in Health Care, Hannover, Germany
d Departments of Obstetrics and Pediatrics, Perinatal Infectious Disease Epidemiology Unit, Hannover Medical School, Hannover, Germany
e Neuroepidemiology Unit, Departments of Neurology, Children’s Hospital and Harvard Medical School, Boston, Massachusetts
f Department of Neonatology, University of Tübingen, Tübingen, Germany
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
METHODS. We analyzed population-based data from a quality assurance program in Lower Saxony (Germany). Perinatal data for almost all very low birth weight infants born in 1991 to 1999 (n = 7745) were available. Analyses were restricted to infants born at 24 to 30 weeks (n = 4379). Data validation procedures, univariate data analyses, and logistic regression models based on general estimating equations were performed.
RESULTS. Neonatal mortality among infants admitted to NICUs was 12.2% in small NICUs and 10.2% in large NICUs. The mortality rate in small NICUs was increased significantly. Compared with infants from large delivery hospitals (>1000 births per year) and large NICUs, the adjusted odds ratio was 1.94 for neonates for whom both units were small, 1.75 for those from large delivery units but small neonatal units, and 1.16 for those for whom only the NICU was large. Stratification according to gestational age revealed the greatest impact on mortality for infants of <29 weeks.
CONCLUSIONS. Results suggest that creating larger perinatal centers may improve perinatal health care. The volume of the NICU was associated more strongly with 28-day mortality than was the volume of the delivery hospital, and it had the largest impact on survival for infants of <29 weeks.
Key Words: very low birth weight infant • neonatal mortality • hospital volume • regionalization
Abbreviations: VLBW—very low birth weight • SGA—small for gestational age • OR—odds ratio • CI—confidence interval
Very low birth weight (VLBW) infants (<1500 g) are at increased mortality risk.1–4 The volume of the delivery unit and the size or level of the NICU seem to influence mortality.5–8 Currently, there are no population-based studies of risk-adjusted VLBW mortality that control for both NICU volume and size of the delivery service.
Delivery hospitals and neonatal care units are often detached. This "decentralization" concept, with many small units, was reinforced (eg, in Germany) to minimize distance and duration of neonatal transport. Today, the obvious advantages of large perinatal centers and in utero transport, as well as economic benefits, have led to the quest for centralization. Unfortunately, supportive data are lacking.
In the absence of a specific classification system that quantifies the level of care, hospitals can be compared with respect to outcomes by using hospital caseload as a proxy. Most previous studies showed that high hospital volume was associated with better outcomes.9 Difficulties in those studies included differences in case mixtures and study populations, varying definitions of hospital categories, presence of bias,10 and focus only on specific procedures.11
Previous studies on neonatal outcomes and hospital volumes did not include both NICU volume and delivery hospital volume5–8 and often did not, or did not sufficiently, include risk adjustments.7,12,13 In particular, VLBW studies were often biased by the use of birth weight cohorts without gestational age limits.14,15
Each year, 
80000 infants are born in the German state of Lower Saxony. Approximately 800 of these have birth weights of <1500 g. We hypothesized that VLBW neonates from high-volume NICUs (
36 VLBW admissions per year) would have a lower 28-day mortality risk, compared with neonates from low-volume NICUs, with adjustment for delivery hospital volume (
1000 or >1000 deliveries per year). Moreover, we assumed a strong influence of gestational age. Our goal was also to offer population-based support for the design of NICU-level definitions and referral guidelines.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
1000 versus >1000 births per year for delivery hospitals and <36 versus 36 VLBW admissions per year for NICUs. The small number of large hospitals in Lower Saxony justifies this relatively low cutoff value. Each newborn was assigned to the NICU where the child was mostly cared for during his or her first 4 weeks of life, because some (6.1%) were transferred shortly after birth to a different NICU, of potentially different size. Among the 107 delivery units, 79 were small (27.2% of deliveries, n = 1190) and 28 were large (72.8%, n = 3189). Among the 36 NICUs, 26 to 29 were small (accounting for 37.7% of the neonates, n = 1652) and 7 to 10 were large (accounting for 62.3% of the neonates, n = 2727). Ranges are attributable to 3 NICUs changing from small to large during the observation period. All NICUs admitting VLBW infants in the state have staff and equipment to provide comprehensive intensive care (including mechanical ventilation) 24 hours per day for even the most immature neonates. Currently there is no formal grading system for level of care in Germany. The effects of different combinations of delivery hospital and NICU volume were analyzed by assigning 1 of the 4 possible combinations to each neonate.
The primary outcome was death before 28 postnatal days, because most deaths among premature infants occur during this period.4,15,18,19 Gestational age was determined on the basis of early ultrasound scans or the last menstrual period and was recorded as completed weeks. We defined growth status, according to German population-based percentiles, separately for male and female infants and for singletons and twins.20 Twin percentiles were used for all multiple births. Birth weight below the 10th percentile was defined as small for gestational age (SGA), between the 10th and 90th percentiles as appropriate weight for gestational age, and above the 90th percentile as large for gestational age. Respiratory distress syndrome was defined as documentation of either this condition or exogenous surfactant administration, because early surfactant application can make the diagnosis of respiratory distress syndrome impossible. Severe congenital malformations were defined according to the Weidtmann code16 and implied life-threatening characteristics.
Statistical Analyses
We first calculated Kaplan-Meier survival estimates for infants from small versus large NICUs and compared them by using the log-rank test. Next, we selected, from the 750 variables available as potentially relevant known confounders, those with clinical relevance, those of great interest, and those with potential relevance for accounting for case mixture adjustment (Table 2).
|
7. The numbers of hospitalization days during pregnancy were categorized, on the basis of the median, as <10 versus 10 days, because a linearity assumption did not hold and previous corresponding thresholds were not known. Time since last prenatal steroid treatment was categorized, on the basis of clinically established cutoff points, as 24 hours, 2 to 3 days, 4 to 7 days, 8 to 14 days, or >2 weeks.
Univariate analyses used 
2, t, and Wilcoxon tests. Variables with P values of <.20 in univariate analyses (testing the association with the outcome and with the exposure) were selected for multivariable modeling. Because gestational age affects outcomes strongly, it was included in all analyses, as were gender and multiplicity. Delivery hospital size was included initially in multivariable analyses independent of univariate analysis results.
Multivariable logistic regression analysis was based on general estimating equations. This allows adjustment for clustering, hypothesizing correlation among individuals within units; infants from the same hospital cannot be regarded as independent individuals because of hospital-specific treatment policies.21,22 Different models were calculated, reducing the significance level from .25 to .10 in stages of .05.23 Models were considered valid on the basis of the c-statistic and analysis of more-complex models without significant changes.23 All analyses were performed with SAS 8.02 software (SAS Institute, Cary, NC). In Germany, anonymized secondary data research does not require human research committee review.
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
Overall, the neonatal mortality was 10.9%, which was slightly lower than reported by others.4,10,26,27 This is probably partly attributable to the exclusion of infants who died in the delivery hospital. Moreover, one bias that cannot be controlled for is the lack of information on whether parents had requested a do-not-resuscitate order, which might affect neonatal mortality rates considerably.
Previous studies compared the effects of hospital levels, mainly the delivery units, on different outcome criteria. Levels of care are usually well defined, on the basis of staffing and technical equipment but also neonatal criteria, such as gestational age, multiplicity, or requirement for ventilatory support.8,19,28 Most studies revealed significantly increased neonatal mortality for VLBW infants not delivered in level III hospitals, with adjusted ORs ranging from 1.13 to 2.28 for VLBW infants.6,28–30 Similar results come from comparisons of subspecialty versus nonsubspecialty perinatal centers31 and general versus university hospitals.32
Studies based on NICU caseloads are sparse, although they are common for surgical procedures (with caseload being used as an indirect quality measure).9 Although volume might serve merely as a proxy for other causal factors, satisfactory outcomes are more likely to be achieved in hospitals with substantial caseloads.33 The discussion about the relationship between delivery service volume and perinatal mortality, however, is already >25 years old.34 Moreover, using daily NICU census as a volume criterion5,8 might yield different results than using annual VLBW admissions.35
In nonfederal Californian hospitals, neonatal mortality was higher in low-volume versus high-volume level III NICUs (adjusted OR: 1.55; 95% CI: 1.22–1.96),5 which is in keeping with our results. International network data indicate that higher volume seems to be associated with reduced mortality up to a threshold of 50 VLBW admissions per year (95% CI: 41–61 VLBW admissions per year). Below this point, an 11% (95% CI: 5–16%; P < .045) reduction in mortality was calculated for every 10 additional infants admitted.35 In one study, investigators calculated that 20% of deaths could have been prevented if 90% of all VLBW infants had been born at the recommended level.19
Cifuentes et al8 investigated neonatal mortality for neonates of <2000 g and found an increased risk in small NICUs, intermediate NICUs, and hospitals without NICUs, in comparison with regional NICUs. They also reported that restriction to VLBW infants would have shown even larger effects, but they considered their results on volume effects as not conclusive.
The United Kingdom Neonatal Staffing Study Group did not find an association between hospital volume and hospital mortality among neonates.36 However, those analyses were not restricted to VLBW infants, and their study was underpowered to investigate the effect of caseload on mortality risk.
No association between patient volume and mortality has been reported by the Vermont Oxford Network,37 but the results were criticized for excluding nonparticipating small community hospitals.31 Moreover, gestational age was not included in the multivariable models.
We found stepwise increasing ORs (Fig 2) for infants from small delivery units and large NICUs (adjusted OR: 1.16), large delivery units but small NICUs (adjusted OR: 1.75), and hospitals where both units were small (adjusted OR: 1.94). Apparently, for infants admitted to a NICU, NICU volume had a larger impact on survival than did the volume of the delivery hospital. If we had included deaths that occurred in the delivery unit, then the effect of delivery unit volume might have been stronger.
The increased mortality risk with decreasing gestational age is well known15,35 but not linear, as often assumed.31 Our study is the first to offer separate multivariable models, stratified according to week of gestational age, for the association of hospital volume and neonatal mortality. This is an elaborate approach yielding relevant confounder insights, for example, gender, severe congenital malformations, and SGA were relevant across all gestational age weeks, whereas other variables were important only for certain gestational age groups. This finding could be of relevance for future gestational age-specific risk studies. A reasonable cutoff value for referral to large perinatal centers seems to be 28 weeks, which was also recommended recently by the American Academy of Pediatrics25 and German medical authorities.24
Other groups reported different influences of hospital volume across birth weight categories.12,28,31,38 Cifuentes et al8 reported that, the lower the birth weight, the stronger was the hospital effect. The findings of 2 study groups31,38 that, within their birth weight cohorts, mortality for neonates with birth weights of 1000 to 1499 g were affected most strongly by low-volume hospitals are intriguing. Overrepresentation of SGA infants with high mortality risk15 is 1 possible explanation.
Although the increased risk of male infants, infants of multiple births, and infants with severe malformations is in agreement with previous work,10,39 we could not confirm that infants with severe malformations are preferentially admitted to larger centers (crude OR for small hospitals: 0.80; 95% CI: 0.60–1.06).40 However, the greater incidence of high-risk pregnancies/infants in large hospitals12,36 was also obvious in our data. Large NICUs accounted for 68% of infants with gestational ages of 27 weeks, for 75% of all SGA infants, and for 65% of neonates with respiratory distress syndrome. Inborn infants are at decreased mortality risk. One reason why we did not see this effect in our analyses might be that hospital volume and inborn status were correlated strongly. In small NICUs, only 18% of infants were not inborn; in large NICUs, this proportion was 64%. As a result, the adverse effect of having been born in a small hospital would be underestimated.
Several methodologic issues should be considered. We used population-based data. Previous studies were often biased because of the use of non-population-based birth weight cohorts.35,36 Data from Lower Saxony seem to be representative of other German states,15 although we acknowledge that a larger sample would have been desirable.
Most previous studies used birth hospital as the primary independent variable.31,35 Because delivery units and NICUs are often adjacent, different effects are difficult to assess.7,8 We defined admittance to a NICU as an inclusion criterion because our primary focus was on NICU volume. Moreover, the NICU to which the pediatrician present at birth belonged could not be defined unambiguously. A 1:1 match between delivery hospital and related NICU was not always possible, because some delivery hospitals collaborated with 2 NICUs. This resulted in the exclusion of both fetal deaths and deaths that occurred in the delivery hospital, potentially introducing a (most likely conservative) bias. The effect of delivery unit volume, including stillbirths and early neonatal deaths, will be the subject of a separate investigation. Another potential source of bias was our decision to assign infants to the NICU in which they were cared for most during their first 1 month of life. Because the sickest infants usually are those transferred to larger centers, our decision likely put the larger hospitals at a disadvantage; however, we preferred this conservative bias to attributing automatically a potentially poor outcome to the initial NICU. In this cohort, only 6% of neonates were transferred. Reanalyzing the models with the initial NICU as the exposure variable changed the results only marginally. The adjusted OR for neonatal death in a small NICU was then 1.82 (95% CI: 1.29–2.59). One potential limitation of our exposure variable is its definition only by hospital volume. We could not differentiate levels of care because all NICUs were equipped to provide comprehensive care for even the most immature neonates, and we did not have detailed information on hospital structure and organization, which should be considered in future studies. These factors, however, are correlated strongly with hospital volume.5 Also, the volume cutoffs used for both NICUs and delivery hospitals may seem rather low by international standards10,35 but are indicative of the current situation in Germany. Use of established thresholds, such as 50 VLBW admissions per year,35 was not possible because, during that time period, only 3 to 5 NICUs fulfilled that criterion. The fact that we still found a volume effect supports the hypothesis of lower neonatal mortality in high-volume hospitals. Our results await confirmation with a more-recent data set and larger volume cutoff values, as well as analyses aimed at determining "optimal" classification cutoff values. We limited our primary outcome to death within 28 days after birth, which happened to comply with other work in this field4,5,18,19
We adjusted for case mixture by using prenatal, perinatal, and postnatal confounders.41,42 In contrast, other retrospective, population-based studies assessed birth and death certificates28 and thus were incapable of adjusting for the entire spectrum of perinatal confounders. Although physiologic factor-based case mixture adjustment is recommended, such as with Clinical Risk Index for Babies scores,43 these "baseline" scores might be prone to lead-time bias.41,44 Clinical Risk Index for Babies data were not complete in our database. As an alternative, we included 5-minute Apgar scores. However, we suggest that a 5-minute Apgar score of <7 should be considered a correlate and not a cause of increased mortality risk (adjusted OR: 2.95; 95% CI: 2.22–3.92). Case mixture adjustment was strengthened by including maternal characteristics. One interesting finding, although it did not achieve statistical significance after adjustment for confounders, was that infants of mothers with a cervix width of 4 cm at admission had increased neonatal mortality risk (crude OR: 1.79; 95% CI: 1.40–2.28; P < .0001). Obviously, what contributes to a short interval between admission and birth also contributes to increased mortality risk. This finding can be viewed as support for the need for in utero transfer to large perinatal centers.31
We adjusted for clustering by using general estimating equation modeling.22 Neonates from the same hospital are not statistically independent because of their being exposed to equal treatment policies. This issue is often neglected31,45 and might result in incorrect P values, incorrect CIs, and bias.21 To control for colinearity between the factors of previous stillbirths/abortions/interruptions, pregnancy at risk, and preterm labor, we analyzed models with different combinations of these variables, which did not affect the other estimates. Therefore, we decided to keep these potentially correlated variables for face validity reasons and to hold with our a priori fixed methodologic strategy.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Address correspondence to Dorothee B. Bartels, PhD, MSc, Departments of Obstetrics, Pulmonology and Neonatology, and Epidemiology, Social Medicine and Health System Research, Hannover Medical School, Carl-Neuberg-Strasse 1, OE 5410, 30625 Hannover, Germany. E-mail: bartels.dorothee{at}mh-hannover.de
The authors have indicated they have no financial relationships relevant to this article to disclose.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  E. C. Eichenwald and A. R. Stark Management and Outcomes of Very Low Birth Weight N. Engl. J. Med., April 17, 2008; 358(16): 1700 - 1711. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
