Risk Factors and Estimation Tool for Death Among Extremely Premature Infants: A National Study
OBJECTIVES: The goals were to assess risk factors and mortality rate changes over time and to develop simple estimates of mortality rates for specific groups of infants at 23 to 26 weeks of gestation.
METHODS: Data from the Israel national very low birth weight infant database on 3768 infants born in 1995–2006 with gestational ages (GAs) of 23 to 26 weeks were evaluated, and we developed a tool for estimating infants' mortality rates.
RESULTS: Major factors associated with death were GA, gender-specific birth weight percentile, prenatal steroid therapy, and multiple births. There was a steady decrease in mortality rates for all GAs during the study period. In 2004–2006, mortality rates before discharge were 89%, 67%, 46%, and 26% for infants with GAs of 23, 24, 25, and 26 weeks, respectively. Estimated mortality rates were calculated as the sum of the percentages determined for each of 4 parameters, as follows: GA of 26, 25, 24, or 23 weeks, 0%, 17%, 34%, and 51%, respectively (P < .001); birth weight percentile of >75th, 25th to 75th, or <25th, 0%, 16%, and 32%, respectively (P < .001); no prenatal steroid treatment, +22% (P < .001); multiple birth, +7% (P = .1). Estimated mortality rates for the 48 subgroups of infants ranged from 0% to 100% and correlated well with observed rates (intraclass correlation coefficient: 0.89).
CONCLUSION: Mortality rates for infants born at 23 to 26 weeks of gestation could be estimated simply on the basis of GA, gender-specific birth weight quartiles, prenatal corticosteroid therapy, and multiple births.
WHAT'S KNOWN ON THIS SUBJECT:
The decision to implement intensive care for extremely premature infants is controversial, and there is a lack of recent evidence to support such decisions.
WHAT THIS STUDY ADDS:
The mortality rates for infants born at GAs of 23 to 26 weeks could be estimated on the basis of GA, gender-specific birth weight quartiles, prenatal corticosteroid therapy, and multiple birth.
To treat or not to treat extremely premature infants born at the limit of viability is a difficult decision, because the birth of such infants has a major impact on the parents, medical teams, and society. Ethical and economic issues are involved in considerations beyond the medical abilities. Although the decision to implement intensive care for extremely premature infants is controversial,1,–,5 there is a lack of recent evidence to support such decisions.3,4
Single-unit–based data on survival rates may be biased easily by local variations in patient characteristics, medical care, and attitudes. Conversely, there are trade-offs between sample size and generalizability. Large population samples, although increasing the precision of estimates of mortality rates, may reflect data from numerous perinatal units and thus more-heterogeneous medical care and attitudes. Few recent large studies have addressed survival rates for extremely premature infants according to gestational age (GA) in geographically defined populations.6,–,9 Some of the problems inherent in the interpretation and generalization of mortality data for infants of borderline viability were reviewed by Watts and Saigal.10 Variations may be attributable, for example, to the use of various denominators, rounding of gestational week, or timing of the deaths reported. Because birth weight and GA are regarded as strong predictors of outcome, management policies often are based on those parameters.4,5,10 Other demographic characteristics or treatments also may modify outcomes, and Tyson et al11 recently looked for additional factors that affect prognoses for extremely premature infants. Their cohort was based on the National Institute of Child Health and Human Development Neonatal Research Network in the United States. Those authors concluded that 4 factors, in addition to GA, could predict favorable outcomes with intensive care, including female gender, prenatal corticosteroid treatment, single versus multiple birth, and higher birth weight.
We assessed a population-based cohort of infants born at 23 to 26 weeks of gestation in the years 1995 to 2006, derived from the Israel national very low birth weight (VLBW) infant database. Our aims were to identify perinatal risk factors associated with death, to assess the changes in mortality rates during the study period, and to develop a simple, practical tool for estimating mortality rates for defined subgroups of infants.
This study was based on analyses of data collected by the Israel Neonatal Network on VLBW infants (≤1500 g) born in Israel between January 1995 and December 2006. All 28 neonatal departments in Israel, representing the Israel national VLBW infant database, were included in data collection. Included were all infants born alive at 23 to 26 completed weeks of gestation.12
Data were collected prospectively on a prestructured form and included information on the parents, the mother's pregnancy history and prenatal care, details of the delivery, and the infant's status at delivery, diagnoses, procedures and complications during the hospital stay, and outcome at discharge. Stillbirths and miscarriages are not reported to the database. All infants born alive in Israel receive a unique identification number at birth. Patient information received by the database coordinator is checked for missing items and logical errors, corrected, completed, and then entered into the computerized database. Patient information is cross-checked with the Israel national birth registry, and data on any missing infants are requested from the birth hospital. Data are collected for all infants until discharge or death. The birth hospital and patient identification remain confidential, through consensus agreement of all participating centers. This study was approved by the Helsinki Committee of the Sheba Medical Center (Tel Hashomer, Israel).
Definitions used were concordant with those in the manual of operations for the Vermont-Oxford Network database13 and were reported in detail previously.14 Death was considered as death before discharge from the hospital. The best estimate of GA (in completed weeks) was determined through the hierarchy of obstetric measures (last menstrual period, obstetric history and examination findings, and first-trimester ultrasonography) and a neonatologist's estimate on the basis of early postnatal physical and neurologic examination findings. Gender-specific birth weight z scores and percentiles were determined according to the intrauterine growth charts reported by Kramer et al.15 Prenatal steroid therapy was considered as either no treatment or any treatment, which included partial and complete courses of therapy. Delivery room resuscitation included endotracheal intubation, cardiac massage, and epinephrine administration.
Birth weight and birth weight z scores for male and female infants were compared by using the Wilcoxon rank-sum test. Mortality rates for subgroups of infants were compared by using the χ2 test. All tests were 2-tailed, and P < .05 was considered statistically significant. Stepwise multivariate regression analyses were used to determine factors at birth that were significantly associated with death for each gestational week group (GA of 23, 24, 25, or 26 weeks). Variables included in the analyses were birth weight or birth weight z scores,15 gender, ethnicity (Jewish/non-Jewish), infertility treatment, prenatal steroid therapy, plurality (multiple/singleton), maternal hypertensive disorders, and amnionitis. Results are presented as odds ratios (ORs) with the appropriate 95% confidence intervals. Subsequent to these analyses, a linear regression model was used to determine estimates for mortality rates, on the basis of patient data for the most recent period (2004–2006). Parameter estimates were rounded to provide a simple, practical method for estimating mortality rates for specific groups of infants. To evaluate whether the observed and estimated values for mortality rates were correlated and their means were not significantly different, the 1-way, random-effects, intraclass correlation coefficient was calculated.16 Statistical analyses were performed by using SAS 9.1 (SAS Institute, Cary, NC).
Study Group and Mortality Rates
During the years 1995–2006, 17 904 VLBW infants (≤1500 g) were registered in the Israel VLBW infant database, accounting for >99% of all VLBW infants born alive in Israel. The study population included all 3768 infants with GAs of 23 to 26 weeks. Infants born at <23 weeks of gestation (n = 282) were not included in the analysis. The clinical characteristics of the infants for each gestational week are shown in Table 1. Cesarean delivery rates increased from 17% at 23 weeks of gestation to 63% at 26 weeks of gestation. Resuscitation in the delivery room was undertaken for 61% of infants born at 23 weeks of gestation. Female infants were of significantly lower median birth weight (ranging from 21 g less at 23 weeks of gestation to 52 g less at 26 weeks of gestation); however, birth weight z scores were similar for the male and female infants. Mortality rates decreased from 93% at 23 weeks of gestation to 75%, 54%, and 32% at 24, 25, and 26 weeks of gestation, respectively. Mortality rates decreased significantly over the 12-year period and, during the latest period (2004–2006), the mortality rates during the initial hospitalization decreased from 89% for infants with GAs of 23 weeks to 68%, 46%, and 27% for those with GAs of 24, 25, and 26 weeks, respectively, which reflects an average decrease of ∼20% per week of gestation.
Factors Associated With Death
Stepwise logistic regression analyses identified factors significantly associated with death for each of the 4 gestational week groups. Birth weight (ORs ranging from 1.5 to 3.1 for each 100-g decrease), male gender (ORs ranging from 1.3 to 2.4), and no prenatal steroid therapy (ORs ranging from 2.8 to 5.8) were significant factors for all 4 gestational week groups and multiple births for the 24-week (OR: 1.4) and 25-week (OR: 1.9) groups. Because female infants were always significantly lighter than male infants in each gestational week, an additional analysis was undertaken by including gender-specific birth weight z scores or birth weight percentiles instead of birth weight.15 In all of these models, male gender was not significantly associated with death in any gestational week. Each 1-unit decrease in z scores was associated with ORs for death of 1.9 to 3.3. To account for the effect of infants' size at birth, in subsequent analyses gender-specific birth weight percentiles were considered in 3 quartile groups (<25th percentile, 25th to 75th percentiles, and >75th percentile), as determined from the charts reported by Kramer et al15 (Table 2).
Observed Mortality Rates
The observed mortality rates for all infants according to GA and birth weight percentile groups are shown in Fig 1. A significant linear decrease in mortality rates with increasing percentile group was present for each gestational week (P < .001). Figure 2 shows the observed mortality rates according to gestational week, birth weight percentile groups, and gender, prenatal steroid therapy, and plurality. Mortality rates for male and female infants were similar in all groups (Fig 2A). No prenatal steroid therapy was associated with a higher mortality rate of ≥20% in almost all subgroups (Fig 2B). The slightly higher mortality rates for multiple births were not statistically significant in any group (Fig 2C).
Estimation of Mortality Rates
On the basis of these analyses, we developed a model for estimating mortality rates by using data for the most recent period (2004–2006). A total of 900 infants, of whom 463 (51.4%) died before discharge, were included. The model was based on 4 predictors of death, that is, GA (23, 24, 25, or 26 weeks), gender-specific birth weight percentile group (<25th percentile, 25th to 75th percentiles, or >75th percentile) (Table 2), prenatal steroid therapy (any or none), and plurality (singleton or multiple), which enabled estimation of mortality rates for 48 specific subgroups of infants (4 × 3 × 2 × 2) (Fig 3). The parameter estimate and rounded estimates determined through linear regression analysis with the model are shown in Table 3. Estimated mortality rates were calculated as the sum of the percentages determined for the 4 parameters, as follows: GA, +17% for each week below 26 weeks; birth weight percentile, +16% for each group of ≤75th percentile; no prenatal steroid treatment, +22%; multiple birth, +7%. The intraclass correlation coefficient between the observed and estimated mortality rates was 0.89 (95% confidence interval: 0.82–0.94), which indicates a high level of agreement between observed and estimated mortality rates. The estimated mortality rates for the 48 subgroups of infants, ranging from 0% to 100%, are shown in Fig 3. For example, a singleton male infant born with a birth weight of 685 g at 24 weeks of gestation who received prenatal steroid therapy would have a cumulative mortality risk of 34% for 24 weeks, 16% for birth weight in the 25th to 75th percentile group, 0% for prenatal steroid treatment, and 0% for singleton (Table 3); therefore, the estimated mortality risk would be 50%. This estimate can be determined easily by obtaining the birth weight percentile group from Table 2 and then the mortality rate from Fig 3. For comparison, the observed mortality rate in this group of infants was 56.4% (22 of 39 infants).
Guidelines for treatment of infants born at the border of viability vary.17,–,19 However, it is generally accepted that decisions regarding management are made jointly by parents and physicians. Parents should receive appropriate information about the potential for survival and risks of adverse outcomes. Parental choices regarding management should be respected within the limits of medical feasibility and appropriateness.17 The Nuffield Council recommendation for resuscitation suggests precedence for parents' wishes at 23 weeks, resuscitation as a standard unless parents and clinicians agree, in light of the infant's condition, that it is not in his or her best interest at 24 weeks, and intensive care at ≥25 weeks.18 The Canadian Paediatric Society and the Society of Obstetricians and Gynaecologists of Canada suggest complying with parental wishes at 23 to 24 weeks and attempting resuscitation at 25 to 26 weeks.19 All of these are generally GA-based guidelines that may not be based on satisfactory and up-to-date information for specific subgroups of infants.
Our population-based study found a steady decrease in mortality rates for each GA group of infants born at 23 to 26 weeks of gestation during the years 1995 to 2006, with the major factors associated with death being GA, birth weight percentile, prenatal steroid therapy, and multiple births. These factors were incorporated into a mortality estimation tool that was based on data from the most recent period, which enabled estimation of mortality rates for 48 subgroups of infants. The importance of counseling about the range of possible outcomes, which may be influenced by individual infant characteristics in addition to GA, has been emphasized,17 and this simple tool may enable treating physicians to provide parents with more-accurate assessments of the possible outcomes of these infants.
On the basis of the risk factors found over the 12-year study period and with the use of recent mortality data (2004–2006), our model for estimating mortality rates showed cumulative significant additional mortality rates of 17% for each gestational week below 26 weeks, 16% for each birth weight quartile group below the 75th percentile (<25th and 25th to 75th percentiles), and 22% for no exposure to prenatal corticosteroid therapy. Multiple births added 7% to the estimated mortality rate. The magnitude of the effect of increasing birth weight quartile and exposure to prenatal corticosteroid treatment on mortality rates was comparable to that of an increase of 1 gestational week.
Our study supports the findings of Tyson et al,11 who challenged the use of a GA threshold alone in deciding whether to administer intensive care to extremely premature infants. In contrast to that report, however, we showed that inclusion of gender-specific birth weight z scores or percentiles, rather than birth weight, in the analysis resulted in the exclusion of gender as a significant factor for mortality rates in the GA groups studied. We speculate that this finding can be explained by the lower birth weight among female infants, compared with male infants of the same GA. Female infants of similar weights as male counterparts may be 3 to 4 days more mature and experience a 7% or 8% mortality rate advantage. As noted in our results, the birth weight z scores of male and female infants were similar and were highly significantly associated with death in all gestational weeks.
The use of prenatal steroid treatment was found to reduce mortality rates even for the most premature infants in our study, which supports the findings of others.11,20 In the EPICure study,20 steroid therapy also affected the rate of major abnormalities noted on cerebral ultrasound scans. The better outcomes for infants who received prenatal corticosteroid treatment may result at least in part from corticosteroids being used when obstetricians are committed to optimizing outcomes.21 Whether their use has benefit before GA of 23 weeks remains to be determined in randomized trials.22 Singleton birth was found to have a modest positive effect on mortality rates in our study. Several studies reported plurality to be a significant factor,11,20,23 whereas Draper et al24 found unexpectedly better outcomes for multiple births. A finding of no difference in survival rates between preterm singletons and twins with controlling for birth weight and GA was suggested by others.25,26
We noted steady improvement in mortality rates over time in all GA groups. Field et al6 reported improved overall survival rates for infants who were born at <26 weeks of gestation and were admitted to the NICU, to 47% in 2000–2005 from 36% in 1994–1999. No infant born at 22 weeks survived, and there was no improvement in survival rates for infants born at 23 weeks (18% to 19%); therefore, improvement was attributed to infants born at 24 and 25 weeks. Our study, similar to the EPICure study,20 reports on live births rather than NICU admissions, with the advantage of estimating possible mortality rates just before delivery but the disadvantage of not considering separately delivery room care (willingness to treat and infant immediate viability) and NICU care (medical improvements and limitations and short- and long-term outcome measures). Because our survival rates improved significantly over time for all GA groups, we used only the most recent period (2004–2006) for development of the mortality rate estimation tool, and we suggest that such a tool should be updated periodically for each population, according to its most recent data.
Our study is unique in being a national, population-based analysis with data on >99% of VLBW births in Israel and in including very recent data (1995–2006). The database includes data on pregnancy and perinatal variables collected by all NICUs using standard definitions that remained unchanged throughout the study period. Therefore, it represents a solid database for assessment of perinatal risk factors and mortality rates in this group of infants.
A number of limitations should be considered in assessment of the results and implications of this study. The database does not require reporting on how the best estimate of GA was assigned. In Israel, however, first-trimester ultrasound examinations for dating have become routine practice and are provided to all women, at no cost, in the framework of the national health insurance law. The study was observational in design, and variations in obstetric and neonatal attitudes might influence clinical practices and outcomes. The observed associations of perinatal factors with death might be biased by obstetric and neonatal practices that vary on the basis of GA or birth weight. For example, mortality rates at 23 weeks of gestation might be confounded by delivery room resuscitation practices. Similarly, part of the effect of GA might be attributable to practices that are predicated on expected outcomes as GAs decrease. Although active resuscitation was reportedly undertaken for >60% of infants born at 23 weeks of gestation, we have not attempted to assess the possible effect of a noninterventional approach to management on outcomes. Furthermore, because patient and birth hospital identifications remained confidential, the possibility of different practices or outcomes among NICUs could not be evaluated. The improving accuracy of sonographic birth weight estimations for small fetuses remote from term, using 3-dimensional and volumetric measurements,27 may enable the use of estimated fetal weight determined shortly before birth as a proxy for birth weight in such models. However, the use of estimated fetal weight in our model, for prenatal counseling, requires further validation. In developing this mortality rate estimation model, we used the complete national data set for the recent period; therefore, a separate population sample was not available for validation of the model. Despite this limitation, our analysis showed a high level of agreement between observed and estimated mortality rates. Finally, our study reflects recent Israeli data and may be limited in its applicability to populations in other countries, with different policies and ethical approaches and with different medical resources and capabilities. These results may facilitate the creation of mortality rate estimation tools for other populations, adopting the principles of our model and using their own mortality data.
Our national, population-based study found that mortality rates for infants born at 23 to 26 weeks of gestation decreased steadily during the most recent decade in each of the GA groups and could be estimated simply on the basis of 4 parameters available at birth, that is, GA, gender-specific birth weight quartile, prenatal corticosteroid therapy, and multiple birth. This information may be useful for counseling families regarding treatment options for these infants.
The Israel national VLBW infant database is supported in part by the Israel Center for Disease Control and the Ministry of Health.
The Israel Neonatal Network, including centers participating in the Israel national VLBW infant database, is as follows: coordinating center: Women and Children's Health Research Unit, Gertner Institute for Epidemiology and Health Policy Research, Tel Hashomer; neonatal departments: Assaf Harofeh Medical Center, Rishon Le Zion; Barzilay Medical Center, Ashkelon; Bikur Holim Hospital, Jerusalem; Bnei Zion Medical Centre, Haifa; Carmel Medical Center, Haifa; English (Scottish) Hospital, Nazareth; French Hospital, Nazareth; Hadassah University Hospital Ein-Karem, Jerusalem; Hadassah University Hospital Har Hazofim, Jerusalem; Haemek Medical Center, Afula; Hillel Yafe Medical Center, Hadera; Italian Hospital, Nazareth; Kaplan Hospital, Rehovot; Laniado Hospital, Netanya; Maayanei Hayeshua Hospital, Bnei-Brak; Meir Medical Center, Kefar Saba; Misgav Ladach Hospital, Jerusalem; Naharia Hospital, Naharia; Poria Hospital, Tiberias; Rambam Medical Center, Haifa; Rivka Ziv Hospital, Zefat; Schneider Children's Medical Center of Israel and Rabin Medical Center (Beilinson Campus), Petach-Tikva; Shaare-Zedek Hospital, Jerusalem; Sheba Medical Center, Tel Hashomer; Soroka Medical Center, Beer-Sheva; Sourasky Medical Center, Tel Aviv; Wolfson Medical Center, Holon; Yoseftal Hospital, Eilat.
- Accepted November 25, 2009.
- Address correspondence to David Bader, MD, MHA and Amir Kugelman, MD, Bnai Zion Medical Center, Department of Neonatology, 47 Golomb St, Haifa, 31048, Israel. E-mail: ,
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- GA =
- gestational age •
- VLBW =
- very low birth weight •
- OR =
- odds ratio
- Higgins RD,
- Delivoria-Papadopoulos M,
- Raju TN
- Field DJ,
- Dorling JS,
- Manktelow BN,
- Draper ES
- Markestad T,
- Kaaresen PI,
- Rønnestad A,
- et al
- Moser K,
- Macfarlane A,
- Chow YH,
- Hilder L,
- Dattani N
- Watts JL,
- Saigal S
- Engle WA
- 13.↵Vermont-Oxford Network. Vermont-Oxford Network Database Project: Manual of Operations. 2nd ed. Burlington, VT: Vermont-Oxford Network; 1993
- Shinwell ES,
- Reichman B,
- Lerner-Geva L,
- Boyko V,
- Blickstein I
- Kramer MS,
- Platt RW,
- Wen SW,
- et al
- MacDonald H
- 19.↵Canadian Paediatric Society, Fetus and Newborn Committee; Society of Obstetricians and Gynaecologists of Canada, Maternal-Fetal Medicine Committee. Management of the woman with threatened birth of an infant of extremely low gestational age. CMAJ. 1994; 151(5): 547–553
- Costeloe K,
- Hennessy E,
- Gibson AT,
- Marlow N,
- Wilkinson AR
- Roberts D,
- Dalziel S
- Draper ES,
- Manktelow B,
- Field DJ,
- James D
- Vanhaesebrouck P,
- Allegaert K,
- Bottu J,
- et al
An Unfamiliar Reference Book Worth Knowing About: Have you heard about the Statistical Abstract of the United States? This book, published annually by the Census Bureau, contains real information on subjects that we usually only speculate about. According to an article in Newsweek by R. J. Samuelson (January 18, 2010), the book provides information on everything from consumer spending to internet usage. It even compares data obtained in the United States with that of other countries. For example, food is cheaper here than almost anywhere else in the world: perhaps it is the low food prices here that help contribute to overeating, and the obesity epidemic in this country. Internet usage according to the Statistical Abstract is now ubiquitous in this country, with 92% of Americans using the internet, up from 72% a decade ago. We spend on average 1613 hours per year watching television, the equivalent of 67 days! There is much information to be gleaned from this annual volume. Let us hope that the data provided will spur improvement based on those statistics that clearly show we need to do better. If you are interested in learning more about it, the entire book may be downloaded from: www.census.gov/compendia/statab/.
Noted by JFL, MD
- Copyright © 2010 by the American Academy of Pediatrics