Despite continuing advances in neonatal care and decades of improving outcomes, it has recently been reported that survival rates have leveled-off in premature infants during the 1990s 123. In the state of Delaware, infant mortality rates improved during the early 1990s similar to national trends, only to see a more recent increase which has been attributed to increasing mortality in very low birth weight infants 4. In the United States, infant mortality increased in 2002 for the first time in nearly 5 decades 56. In 2002 the United States fetal mortality rate declined while infant mortality increased 6. A shift from fetal to neonatal deaths is one possible explanation for this trend. The potential impact of increasing fetal survival on neonatal intensive care has not been explored.

Severe intraventricular hemorrhage (IVH) is a significant morbidity in premature infants and is one of the major determinants of long term outcome in very low birth weight infants 7. Severe IVH has recently shown to have declined in occurrence prior to the 1990's and to have remained unchanged thereafter 8. Other common morbidities in premature infants including chronic lung disease have also been recently described as increasing in survivors 9. The reasons for increasing morbidities and recent lack of improvement in survival rates of preterm infants have not been conclusively explained. It has been postulated that the technology for caring for premature infants has reached it limits. One factor that has yet to be investigated is the longitudinal effect of illness severity on morbidity and mortality in premature infants. We hypothesized that illness severity in very low birth weight (VLBW) infants was increasing over time. The objectives of this study were to investigate whether illness was increasing over time and to determine the effect of illness severity on the outcomes of death and/or severe IVH in a population of VLBW infants. As the fetal mortality rate has been declining in the United States 6, and may be leading to a shift from fetal death to physiologically compromised preterm infants, we also explored the fetal death rate at our institution during the study time period.

This investigation consisted of a cohort study of VLBW infants, <1500 grams, cared for at a single level III Neonatal Intensive Care Unit (NICU) during a 9-year period, July 1993–July 2002, n = 1414. The NICU at Christiana Hospital is a level III NICU serving the state of Delaware and cares for both inborn (90%) and outborn infants. Neonatal care within the State of Delaware is regionalized. One other level III NICU in the same network serves as a referral center for VLBW infants only if they require surgical care. No other nursery in the State offers intensive care for preterm infants with birth weights <1250 grams. A small number of infants with birthweight between 1250–1500 grams are cared for at two other level II sites within the same state neonatal network. There were no changes in the number of hospitals providing neonatal intensive care during the study time period. The NICU at Christiana Hospital is university affiliated and staffed by residents in pediatrics, neonatal nurse practitioners, and fellows in neonatal/perinatal medicine in addition to attending neonatologists. Decisions concerning resuscitating babies at the limits of viability are made by the attending neonatologist after family consultation. In general, infants 23 to 24 weeks gestation are offered a trial of intensive-care if desired by the family following consultation. This policy did not change during the study period. Throughout the study period Christiana Hospital was the only hospital in the state offering perinatology services and high risk obstetrical care. The Institutional Review Board at Christiana Care Health Services approved this research project. Data were obtained from a computerized database and review of the medical record. Informed written consent was not obtained.

Illness severity was quantified using two objective methods: the Score for Neonatal Physiology (SNAP) 10 and total thyroxine (T₄). SNAP was routinely calculated after 1996 and therefore was available on infants born after 1996. SNAP was calculated on physiologic data from the 1^st 24 hours of life. T₄ was obtained from the State of Delaware Newborn Screening Program and used as a proxy for illness severity. We, and others, have previously shown that total T₄ is correlated with illness severity in VLBW infants 111213. T₄ was obtained on the 5^th day of life as part of routine newborn screening. Therefore, infants who died before the 5^th day of life did not have a measurement of T₄. The mean ± SD of total T₄ in infants in the State of Delaware during the study period was 13.1 ± 4.2 μg/dl. In order to analyze outcome over time, the study sample was subdivided into consecutive three-year cohorts. Cohort 1 included infants born 7/93–7/96, Cohort 2, 7/96–7/99, and Cohort 3, 7/99–7/02.

Cranial ultrasounds were routinely obtained on the 4^th day of life and then monthly until discharge. Cranial ultrasounds were obtained more frequently if clinically indicated. Ultrasounds were done using a 7.5 mHZ transducer and studies were interpreted by a pediatric radiologist. IVH was graded using the classification system of Papile 14. Severe IVH was considered grade III-IV. Gestational age was based on the best obstetrical estimate. Modified Ballard exam was used only if an obstetrical estimate was not available 15.

A fetal death was defined as any fetus ≥ 20 weeks of gestation that was not live-born. No policy changes in the definition for classifying fetal death were made throughout the study period. Fetal death rate is reported per 1000 births plus fetal deaths. Mothers were classified as receiving steroids if they received any doses of antenatal steroids. Betamethasone is the steroid used routinely at our institution. Clinical chorioamnionitis was diagnosed by the attending obstetrician, based on the presence of fever, uterine tenderness and/or foul smelling amniotic fluid. Any pregnancy with two or more fetuses was considered multiple-gestation. As infertility and in vitro fertilization have been increasing during the study period 16, infertility was investigated as an independent variable. For the purposes of this study, mothers were classified as having a history of infertility if their infants were conceived by in-vitro fertilization or if they received ovulation-enhancing medications such as clomiphene citrate. Mothers were classified as having preeclampsia if given the diagnosis by the attending obstetrician per ACOG guidelines 17. Oligohydramnios was diagnosed by the attending obstetrician, no criteria for a minimal amniotic fluid index was used as it was not regularly available in the medical record. Prolonged rupture of membranes was considered greater than 24 hours. Maternal age was based on age at time of birth. Infants with birth weight < 1000 grams were analyzed as a sub-group.

Statistical analyses included both univariate and multivariate analysis. Univariate analysis included chi-squared for categorical variables, and analysis of variance for continuous variables with normal distribution. Levene's test of homogeneity of variances was used to assess data distribution. Mann-Whitney U test was used for continuous variables that were not normally distributed. Kruskal-Wallis ANOVA was also used to compare ordinal data over time. Pearson correlation was used to compare two continuous variables. Multivariate analysis included logistic regression and linear regression. The contributions of increased to death and/or severe IVH were calculated by comparing observed to predicted death and or IVH during Cohort 3. Data from the logistic regression models were used to calculate a standarized rate 18 for the outcome of death and/or severe IVH. A p value < .05 was considered statistically significant. Statistica (V7.0, Tulsa, OK) was used for all statistical calculations

A total of 1414 VLBW infants were cared for in the neonatal intensive care unit during the 9-year time period and comprised the study sample. Cohort 1 (1993–1996) consisted of 393 infants; there were 485 infants in Cohort 2 (1996–1999), and 536 in Cohort 3(1999–2002). In the study sample, 610 infants (43%) were < 1000 grams birth weight. The proportion of infants with birthweights <1000 grams, 42%, 41%, and 46% (p = .26) in Cohorts 1, 2, and 3 respectively, did not change over time. The distribution of infants, in 250 gram increments did not differ between Cohorts (Table 1). Of the 1414 infants, 1207 (85%) survived until the 5^th day of life and had a measurement of T_4.

Birthweight and maternal age were not different between the first, second and third cohorts (Table 2). There was a small but significant decrease in gestational age over time. There was no difference in the percentage of inborn infants over time. There was an increased use of antenatal steroids and antibiotics during the 2^nd and 3^rd Cohorts compared to the 1^st Cohort. There was also an increased occurrence of preeclampsia and cesarean delivery over time. There were no differences in the occurrence of maternal HELLP syndrome, diabetes, illicit drug use, clinical chorioamnionitis, prolonged rupture of membranes, oligohydramnios, multiple gestation, or infertility over time. In the infants in the study sample, there was no change in the use of mechanical ventilation over time but the use of rescue high frequency ventilation and surfactant did increase over time. Because dopamine can inhibit pituitary hormone production 19, the use of dopamine and dexamethasone in infants was investigated. The use of dopamine in infants increased progressively over time while the use of postnatal dexamethasone increased between Cohort 1 and Cohort 2 then dropped during Cohort 3.

Postnatal illness severity increased over time, as measured by total T₄ and SNAP (Table 3). Linear regression models were used to assess the effect of birth cohort on illness severity. Forward stepwise linear regression was used. Variables entered into the model included those that were significant on univariate analysis and variable which are known confounders for illness severity such as birthweight. Birth cohort remained significant in the models when illness severity was measured using SNAP as the dependent variable (model r² = .41, p < .01) and T₄ as the dependent variable (model r² = .35, p < .01). Other than birth cohort, gestational age and birthweight were the only independent variables contributing more to the variability in both models of illness severity. In addition to gestational age and birthweight, the models controlled for maternal antibiotics, multiple gestation birth, race, maternal steroids, preeclampsia, cesarean delivery and inborn status.

There was no difference in the rate of mortality over time (Table 3). However, those infants who died during Cohort 3, died earlier compared to those infants who died during both the 1^st and 2^nd Cohorts. There was an observed increase in both severe IVH and the combined outcome of death and/or severe IVH over time. In the sub-group of infants less than 1000 grams, illness severity increased over time as measured by both SNAP and T₄ (Table 4). Similar to the general study sample, mortality did not increase over time, but there was an increase in severe IVH as well as the outcome of death and/or severe IVH in the infants <1000 g. Table 5 shows changes in T₄ and SNAP in the study cohorts based on 250 gram birthweight increments. Illness severity decreased with increasing birthweight in each cohort (as measured by increasing SNAP and decreasing T₄.) Illness severity increased over time in each birthweight subgroup as measured by increasing SNAP. Illness severity, as measured by decreasing T₄, increased over time in each birthweight subgroup, with the exception of infants 500–750 grams which approached statistical significance.

During the study period the fetal death rate at Christiana Hospital decreased over time. The fetal death rate was 7.8/1000 in Cohort 1, 6.7/1000 in Cohort 2, and 5.3/1000 in Cohort 3 (p = .01, Cohort 1 vs 3).

Our data show an increase in illness severity in VLBW infants over a 9-year period at a regional level III NICU in the State of Delaware. During this same time period, there was an associated decrease in the fetal death rate, and an increased occurrence in severe IVH. The contribution from increased illness accounted for a majority of the variability of the observed increase in the composite outcome of death and/or severe IVH during the final cohort.

To our knowledge this is the first report of increasing illness severity in a population of VLBW infants. Our report is consistent with data from Vermont-Oxford Network and others who have recently reported no improvement in survival of VLBW infants during the 1990's 123. Although the reasons behind these plateaus in survival rates were not clear, it has been hypothesized that neonatal care has reached its limits. In our population of VLBW infants, the increase in severe IVH was temporally associated with an increase in illness severity as quantified by both SNAP and T₄. Our data suggest that a majority of the increase in the combined outcome of death and severe IVH could potentially be accounted for by increasing illness. In addition, VLBW infants who died during the 3^rd Cohort (1999–2002), died at an earlier time compared to the first 2 cohorts. The increased illness severity, documented on the 1^st day of life by SNAP scores, and the earlier time of death, suggest an antenatal or perinatal etiology for the observed increase in illness.

We speculate that a decrease in fetal death rate has resulted in an increased number of compromised infants and an increase in neonatal illness. Fetal compromise has previously been shown to be associated with neonatal death 20. In our study sample, there was an increase in cesarean delivery, antibiotics, and steroids over time as has been observed in other recent studies 1921. Despite these obstetrical interventions, we observed no improvement in survival and increase in severe IVH, one of the major morbidities in VLBW infants over time. Changes in perinatal management of VLBW infants, in an attempt to aggressively care for pregnancies that may have previously been considered non-viable from a medical or gestational age stand-point, may be contributing to the decrease in fetal mortality and subsequent increase in neonatal illness. Alternatively, the observed increase in illness severity may be attributed to other factors, which we were unable to quantify in this study, such as increasing maternal stressors in the antenatal period. The fact that survival rates have remained consistent despite increasing illness suggests that neonatal care may be continuing to advance but not at a rate fast enough to overcome a steady increase in illness severity. Although the VLBW infants in Cohort 3 had a similar occurrence of death compared to the older cohorts, increasing illness may be contributing to an increase in neonatal morbidities such as severe IVH in survivors. Consistent with our study, other important neonatal morbidities including bronchopulmonary dysplasia 9, retinopathy of prematurity 22 and neurodevelopmental disabilities 21 have all been recently shown to be increasing in surviving preterm infants.

The present investigation used T₄ as a proxy for illness as we and others have previously shown the association of T₄ and illness severity 111213. We feel that T₄ is a valid marker of illness severity as confirmed by the inverse correlation between SNAP and T₄ in our study sample. Previous research has shown equivalent ROC curves for T₄ and SNAP in relationship to the outcomes of severe IVH and death 11. In support of the fact that T₄ is a marker for illness severity, rather than an important part of the causal pathway, studies of thyroxine supplementation in preterm infants have failed to improve neonatal outcomes 11232425. Because T₄ was measured on the 5^th day of life, we can not rule out the possibility that early management of VLBW infants may have influenced illness severity and T₄. This possibility, however, remains unlikely as management strategies did not change dramatically at our center during the study period. The use of dopamine, which inhibits thyroid stimulating hormone production 19, did increase over time. However, increasing dopamine use would be expected with an increase in illness and would not explain an increase in SNAP. It also remains a possibility that low T₄ is an important part of the causal pathway in illness severity and that some primary change in thyroid function over time in our population contributed to the observed increase in illness severity. Even if a primary decrease in T₄ was responsible for the observed increase in illness, the importance of a progressive decrease in T₄ is highlighted by the association of hypothyroxinemia with death 26, IVH 27, cystic periventricular leukomalacia 28, and cerebral palsy 29 in VLBW infants.

Our study has a number of other important limitations. First, since our study sample was from a single region our findings may not apply to other populations. Replication of these analyses from other regions would be needed to determine whether the observed increase in neonatal illness is occurring elsewhere. Our study is also limited by the unavailability of SNAP on infants born during the 1^st Cohort and the unavailability of T₄ on those infants who died prior to the 5^th day of life. However, those infants who died prior to thyroid screening would have been expected to have an increase in illness severity compared to infant who survived. Thus our data may have potentially underestimated the trend of increasing illness. Because a small number of infants with birth weight between 1250 and 1500 grams are cared for at other centers in the state of Delaware they were not included in this analysis. There were no changes in the number of hospital providing neonatal intensive care or maternal-fetal medicine services during the study duration. Furthermore, the increased illness in our region was present in infants <1000 grams, all of which were captured in this study sample. Because all infants <1000 grams born in the state of Delaware during the study period were captured in our cohorts it is unlikely that any changes in referral patterns influenced the results. Although the number of infants <1500 grams increased in our neonatal intensive care unit over time, the proportion of infants <1000 grams did not change over time. Multivariate modeling also adjusted for changes in gestation over time. We can not conclusively rule out the possibility that head ultrasounds were interpreted differently over time. Our data however are from a single center with uniform ordering of cranial sonograms and definitions of severe IVH. Although we were able to look at many variables influencing illness severity including clinical chorioamnionitis, there may have been other variables such as histologic chorioamnionitis 30 or time of birth 31 which we were unable to explore. Finally, although also unlikely, we can not rule out a change in the quality of obstetrical or neonatal care as a cause for increasing death on/or severe IVH. The similarities in the distribution of Apgar scores over time do not suggest that any changes in early delivery room management are responsible for the observed outcomes. The drop in the standardized rate observed over time suggests an improvement in neonatal outcomes, given the magnitude of increased illness severity.

Recent national data have shown that mortality in VLBW infants is no longer declining despite advances in neonatal care 123. Our data are important in showing an increase in illness severity over a 9-year period in VLBW infants. We were also able to document an associated decrease in the fetal death rate during this same period. Infant mortality in the United States has been reported to have increased for the 1^st time since 1958 5. Similar to our regional findings, the recent increase in infant mortality in the United States was associated with a decrease in the fetal death rate 5. Based on our data we can not determine the cause of the observed increase in illness in VLBW infants, but speculate that there maybe a shift of fetal deaths to live, but severely physiologically compromised newborns. Future investigations and other regional data analysis will be necessary to confirm these findings.

IVH Intraventricular hemorrhage

NICU Neonatal Intensive Care Unit

SNAP Score for neonatal acute physiology

T₄ Total thyroxine

VLBW Very low birth weight

The author(s) declare that they have no competing interests.

DP was responsible for conceiving the design of the study, acquiring the data and maintaining the database, data analysis, drafting and revising the manuscript. KL was responsible for acquiring data, maintaining the database, and manuscript revision. RL participated in data analysis, design of the study, and manuscript revision. LB was responsible for acquiring the data on newborn thyroid function, interpreting data and manuscript revision. JW participated in data analysis and manuscript revision. JS participated in data analysis, design of the study, and manuscript revision. All authors read and approved the final manuscript.