 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ARTICLE |
a Departments of Health Policy
b Obstetrics, Gynecology, and Reproductive Science
c Pediatrics, Mount Sinai School of Medicine, New York, New York
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONREFERENCES |
|---|
METHODS. We performed a population-based cohort study using New York City vital statistics records on all live births and deaths of infants weighing 500 to 1499 g who were born in 45 hospitals between January 1, 1996, and December 31, 2001 (N = 11 781). We measured very low birth weight risk-adjusted neonatal mortality rates for each New York City hospital and assessed differences in the distributions of non-Hispanic black and non-Hispanic white very low birth weight births among these hospitals.
RESULTS. Risk-adjusted neonatal mortality rates for very low birth weight infants in New York City hospitals ranged from 9.6 to 27.2 deaths per 1000 births. White very low birth weight infants were more likely to be born in the lowest mortality tertile of hospitals (49%), compared with black very low birth weight infants (29%). We estimated that, if black women delivered in the same hospitals as white women, then black very low birth weight mortality rates would be reduced by 6.7 deaths per 1000 very low birth weight births, removing 34.5% of the black/white disparity in very low birth weight neonatal mortality rates in New York City. Volume of very low birth weight deliveries was modestly associated with very low birth weight mortality rates but explained little of the racial disparity.
CONCLUSION. Black very low birth weight infants more likely to be born in New York City hospitals with higher risk-adjusted neonatal mortality rates than were very low birth weight infants, contributing substantially to black-white disparities.
Key Words: infant mortality • racial disparities • quality of care • very low birth weight
Abbreviations: VLBW—very low birth weight
Black infants in the United States are more than twice as likely as white infants to die in the first year of life.1 In New York City, infant mortality rates were 3 times higher for black infants than for white infants in 2001.2 Neonatal deaths, that is, deaths that occur within 28 days after delivery, account for nearly two thirds of all infant deaths. Similar to the racial disparities in infant mortality rates, black neonates are more than twice as likely to die, compared with white neonates.3
Racial disparities in infant and neonatal mortality rates have been attributed largely to higher rates of preterm births and the resulting greater prevalence of low birth weight (<2500 g) and especially very low birth weight (VLBW) (<1500 g) infants in the black community.4 VLBW is 2.6 times more common among black infants than among white infants. VLBW births account for the large majority (70%) of neonatal deaths.3
Research from the Vermont Oxford Network and others showed that structural hospital characteristics such as NICU level (sophistication of nursery), NICU volume, and VLBW birth volume are associated with neonatal outcomes.5–7 However, these indicators explain little of the variation in mortality rates among hospitals.5,6 Findings from the Vermont Oxford Network also suggested that hospitals that serve large proportions of minority patients have mortality rates that are higher than expected.8 One shortcoming of many of these analyses is that they used voluntary registry data and therefore were subject to selection bias. For example, the Vermont Oxford Network includes less than one half of the hospitals in New York City that deliver VLBW infants.9
We sought to complement and to extend the findings of the Vermont Oxford Network by using population-based data to investigate the association of hospital neonatal mortality rates and race. Specifically, we asked whether black VLBW infants are born in hospitals with the same mortality rates for VLBW infants as are white VLBW infants. The objectives of our study were (1) to measure risk-adjusted neonatal mortality rates according to hospital for VLBW infants born in New York City, (2) to assess the extent to which black and white mothers use different hospitals to deliver their infants and to determine whether differences in where black and white infants are born contribute to black/white disparities in VLBW neonatal mortality rates, (3) to explore the association of hospital characteristics with VLBW neonatal mortality rates, and (4) to explore the impact of several potential strategies to reduce disparities.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
We included all live VLBW births (defined as birth weights of <1500 g and >499 g) that occurred in hospitals in the 5 boroughs of New York City. We excluded records for which birth weight was missing and births at 5 hospitals with <15 VLBW births each over this 6-year period. We attributed neonatal outcomes to the hospital of birth (whether or not the infant was later transferred), because the hospital of birth has been shown to be a more-important influence on infant survival rates than the hospital to which an infant is transferred.5,6,10
Statistical Models
We used 2 different approaches to our modeling. The first approach calculated risk-adjusted VLBW neonatal mortality rates without hospital-level variables. The second approach assessed specifically the association of hospital characteristics with risk-adjusted neonatal mortality rates.
Variables
Maternal and infant variables were obtained from New York City vital statistics records. In this data set, race is based on mothers' self-reports. Race and ethnicity are separate items on the birth certificate applications, and responses are coded in general conformance with the rules of the National Center for Health Statistics.2 Maternal race was categorized into 6 categories, namely, non-Hispanic white, non-Hispanic black, Asian/Pacific Islander, Puerto Rican, other Hispanic, and other. This study contrasts neonatal mortality rates between non-Hispanic white infants and non-Hispanic black infants.11 In this article, we refer to non-Hispanic black as "black" and non-Hispanic white as "white."
Similar to other published models and literature on neonatal mortality rates,6,12–14 we included maternal and infant characteristics associated with neonatal death in this cohort of very small infants. These factors were present before or immediately at birth and were not influenced by neonatal management.12 These variables included the mother's sociodemographic characteristics (age, race, education, insurance, and marital status), maternal behaviors (tobacco use during current pregnancy, alcohol use, and number prenatal care visits), clinical factors (congenital anomaly, maternal medical risk factors, multiple birth, and infant gender), delivery method (cesarean or vaginal), and initial birth outcomes (birth weight and 1-minute Apgar score). Prenatal visits and Apgar scores had nonlinear relationships with risk of neonatal death and were categorized on the basis of cutoff points suggested by graphical analyses. We assessed the possibility that the relationship between birth weights and mortality rates varied systematically according to race, and we did not find this to occur. We conducted these analyses by including interactions between birth weight and race. In the first model, birth weight was used as a continuous variable. In the second model, we replaced the continuous birth weight variable with indicators of birth weight in 100-g categories and included interaction terms between race and these birth weight categories. The interactions were not significant in either model, and we used birth weight as a continuous variable in the remaining analyses. We used similar techniques to assess the possibility that the relationship between birth weights and mortality rates varied systematically according to gender and multiple births, and we did not find this to occur. For the second set of models, we added hospital-level variables, including volume of VLBW deliveries from vital statistics files, NICU level, proportion of Medicaid admissions, ownership, and teaching status.15,16
Statistical Analyses
To estimate risk-adjusted mortality rates for each hospital, we developed a neonatal mortality model in which the dependent variable was an indicator for death within 28 days after delivery and independent variables included the maternal and infant variables described above. We estimated the model by using random-effects logistic regression, with a random intercept for each hospital.6 Model fit was assessed by using the area under the receiver operating characteristic curve statistics and the Hosmer-Lemeshow goodness-of-fit test.17 We achieved the best model fit by including both birth weight and the inverse of birth weight as explanatory variables. We used a split-half analysis to validate the model-building process and then applied the final model to the full data set.
We calculated the expected deaths for each hospital by summing the predicted probability of death of all infants at each hospital. The ratio of actual to expected deaths at each hospital represents a standardized mortality ratio, which we used to rank the hospitals from lowest to highest. These analyses did not include hospital-level variables, because doing so could distort the ranking of hospitals. For example, if we found, after adjustment for infant characteristics, that the numbers of deaths were higher at hospitals with a given characteristic, then building that characteristic into the model would increase the estimated numbers of expected deaths for those hospitals. This would result in lower ratios of observed/expected deaths and paradoxically higher rankings for hospitals that share the injurious characteristic.
To assess racial disparities18 in the use of hospitals with the lowest-mortality rates, we calculated the cumulative distributions of births among hospitals ranked from lowest to highest mortality ratio for black and white VLBW infants. We used the Kolmogorov-Smirnov test19 to assess whether the distributions of births among hospitals differed for white and black infants. To assess whether any observed difference in this distribution might be attributable to preferential referral of high-risk white mothers to the lowest-mortality hospitals, we constructed similar cumulative distributions for moderate low birth weight (1500–2499 g) and normal birth weight (
2500 g) black and white births.
To address the effects on black infant mortality rates of these differences in the hospital of birth, we reestimated the model discussed above, including dummy variables for each hospital. This model was estimated as a standard logistic model, without the hospital-level random effects. We then randomly allocated black mothers to hospitals in the same proportion that white mothers were allocated, and we recalculated the predicted probability of death for black VLBW infants. The mean of predicted probabilities before reassignment of hospitals exactly equals the observed black mortality rate,17 and the difference between the observed probability of death and the predicted probability after reassignment can be attributed to the reassignment. We used bootstrap methods20 to estimate a SE for this difference.
To investigate the association between hospital characteristics and VLBW neonatal mortality rates, we estimated random-effects logistic models that included maternal and infant characteristics, as well as the hospital characteristics described above. Because high volume of VLBW births at a hospital was associated with better survival rates, we estimated the effects of racial differences in the distribution of births at high-volume hospitals by randomly allocating black mothers to high-volume VLBW hospitals in the same proportion as white mothers were allocated, and we recalculated predicted mortality rates.
Our final step was to examine the impact on overall mortality rates and disparities in mortality rates of 2 potential improvement strategies. First, we estimated the effect of reducing the mortality rates of the highest mortality tertile of hospitals to the average mortality rate of the other hospitals. We did this by estimating a logistic model with mother/infant characteristics and a single dummy variable for whether the birth hospital was in the highest mortality tertile, setting this dummy variable equal to 0 for all infants, and calculating predicted deaths. Second, we estimated the impact of all black and white VLBW births taking place at high-volume VLBW hospitals, by using similar methods but using a dummy variable for low volume in place of the variable for high mortality tertile.
In preparation for these analyses, we attempted to link these data with statewide administrative data on hospital discharges by using a validated algorithm. We found that the cost of doing so would have been the loss of 
25% of our observations. We concluded that the opportunity to present true population data from vital statistics files outweighed the advantages of greater detail in diagnoses available with the administrative data.
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
|
Hospital of Birth and Neonatal Mortality Rates
The cumulative distribution of births among hospitals ranked from lowest to highest mortality ratios differed for black and white mothers (P = .006) (Fig 1). Forty-nine percent of all white VLBW births occurred in the lowest mortality tertile of hospitals, compared with 29% of all black VLBW births. Eleven percent of white VLBW births and 21% of black VLBW births occurred at hospitals in the highest mortality tertile.
|
|
2500 g) infants all had similar distributions among hospitals (P = .950 for VLBW versus normal birth weight and P = .995 for low birth weight versus normal birth weight), as did white infants of various birth weights (P = .995 for both pairwise comparisons). Regardless of the birth weights of their infants, white mothers were more likely to deliver at lower-mortality hospitals, compared with black mothers.
Volume of VLBW Deliveries and Mortality Rates
Hospitals that delivered more VLBW infants experienced lower mortality rates, compared with lower-VLBW volume hospitals (Table 2). This relationship was not consistent, in that low- and high-volume hospitals could each be found in the low and high mortality tertiles (Fig 1). Black VLBW births were somewhat less likely to occur at high-volume hospitals, compared with white VLBW births (77% and 86%, respectively). We estimated the impact of this association on black VLBW neonatal mortality rates by using a modeling procedure exactly analogous to the one described above. If black VLBW infants were born at high-volume hospitals at the same percentage as white VLBW infants, then the estimated mortality rate for black VLBW infants would be 137.1 deaths per 1000 VLBW births, which represents a 1.4% decrease in mortality rates for black VLBW infants, and the black/white mortality rate disparity would decrease by 10% (Table 3).
Finally, we examined the effects on the overall and racial disparities in VLBW mortality rates of 2 different potential improvement strategies (Table 3). If mortality rates in the highest mortality tertile of hospitals were reduced to the average of the other 30 hospitals, then the black VLBW neonatal mortality rate would be reduced to 127.7 deaths per 1000 black VLBW births. The white VLBW neonatal mortality rate would be reduced to 113.3 deaths per 1000 white VLBW births. Such reductions would remove a little more than one fourth of the black/white disparity. If all white and black VLBW births took place at high-volume hospitals, then more-modest reductions would occur (black: from 139.0 to 134.0 deaths per 1000 VLBW births; white: from 119.6 to 116.8 deaths per 1000 VLBW births; reduction in disparity: 11%).
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
We examined the impact of 2 strategies involving the potential redistribution of black VLBW births and 2 other strategies aimed at improving outcomes at specific kinds of hospitals. Our data suggest that the strategy aimed at improving outcomes at the highest-mortality hospitals would produce the greatest benefit. Our findings must be taken in context. The impact of hospital of birth in our data likely underestimates the total impact of hospital of birth on neonatal mortality rates. Our results primarily relate to the influence of hospital of birth on the care of neonates. Research has demonstrated that two thirds of the decrease in VLBW neonatal mortality rates is attributable to improvements in neonatal management.21 The quality of labor and delivery care has a major impact on neonatal outcomes. We recognize that by following the emerging standard and including mode of delivery and 1-minute Apgar scores,6,12,22 our analysis will reflect only a portion of the impact of labor and delivery care on neonatal mortality.
Our findings are consistent with data from the Vermont Oxford Network, as well as studies from other areas in medicine that document that black patients and white patients are treated at different sites of care and black patients often are treated at higher-mortality hospitals.23,24 Why hospitals that treat greater proportions of black VLBW infants experience higher risk-adjusted neonatal mortality rates is not known. Studies of acute myocardial infarction treatment have shown that black patients tend to receive care in hospitals with higher mortality rates and lower rates of effective, evidence-based, medical treatments, compared with white patients.24 Whether black infants are more likely to receive care in New York City hospitals with lower rates of evidence-based medical treatments, compared with white infants, requires additional study.
The reasons why women deliver at specific hospitals likely involve a number of factors, including where a patient resides and the distance to the hospital (both of which may be influenced by patterns of racial segregation), physician referral, patient choice, access, insurance contracts, and the management of possible medical emergencies during pregnancy. We could not evaluate fully the extent to which each of these factors contributed to the site of delivery in this study. We did not find evidence that preferential physician referral of high-risk white mothers to the lower-mortality hospitals explained the difference in distribution among hospitals for VLBW white and black infants. All white mothers (not just those who delivered VLBW infants) were more likely to receive care at hospitals with lower VLBW neonatal mortality rates.
Consistent with the findings of others,25 the correlation between volume and mortality rates was highly variable in our data. Although 62% of low-volume hospitals were represented in the high-mortality tertile, 33% were in the low-mortality tertile (Fig 1). In addition, the difference in the proportions of black (23%) and white (14%) VLBW births that occurred at low-volume hospitals was modest. Our data suggest that use of VLBW volume as a proxy to guide improvement activities offers smaller potential benefit and carries the risk that care at some low-volume hospitals that have low mortality rates may be disrupted.
The wide variations in risk-adjusted mortality rates we found among hospitals suggest that differences in quality of care may exist. It is possible that patient-level risk factors not included or poorly measured in the vital statistics data we used contributed to these findings. However, we incorporated the vast majority of patient factors used by others, and the statistical properties of our model were robust. Variations in quality of health care are well documented, including variations in the use of effective treatments for prematurity.26,27 Furthermore, some studies have demonstrated that focused quality improvement programs can increase hospital-specific proportions of neonates receiving effective care, although such improvements have not, to date, been linked to improved hospital mortality rates.22 Our estimate that a substantial number of VLBW deaths might be averted by reducing the mortality rate in the highest-mortality hospitals just to the average of the rest suggests that such an improvement strategy is worth pursuing.
An important strength of our study was that it was population-based and not subject to selection bias. Unlike data from voluntary registries, vital statistics records capture data on all VLBW infants. New York City vital statistics data also provide a rich source of self-reported data on race and ethnicity. The predictive power of our VLBW neonatal mortality model was high and resulted in a model similar to those developed by the Vermont Oxford Network.6 Like others, we found wide variations in neonatal hospital mortality rates for VLBW infants.6 We adjusted for a number of infant characteristics and found mortality risks consistent with those in previous studies.28 Higher Apgar scores, female gender, more prenatal care, and not having a major congenital malformation were associated with lower mortality rates in our cohort. As others have found, after controlling for birth weight, black race was also associated with lower mortality rates.3,28,29 As a group, however, black VLBW neonates in New York City were still more likely to die than were white VLBW neonates. Our data suggest that 2 factors may explain this finding. Black VLBW infants were more likely to be born at lower birth weights within the VLBW category, and black VLBW neonates were more likely to be born at hospitals with higher neonatal mortality rates. Interestingly, maternal smoking was associated with lower mortality rates in our study. Infants born to smokers are more likely to be small for gestational age, and it is possible that, in our cohort of VLBW infants, infants born to smokers were more mature at any given birth weight. Although our finding that maternal medical risks factors were associated with lower mortality rates is counterintuitive, this finding may be explained by the fact that patients with medical risk factors are more likely to be closely monitored during their pregnancies, may be more likely to have their deliveries managed aggressively, and may be attended by more-experienced obstetricians. Also interesting was the finding that "education not recorded" had a high odds ratio. It is plausible that favorable response bias or issues of literacy might concentrate lower-educated or lower-literate mothers in this category.
The use of birth certificate data allowed us to conduct a population cohort study of mortality rates for all VLBW births in New York City. These data have some limitations. Studies have assessed the reliability of birth certificate data and have found that reliability is high for demographic data, birth weight, Apgar scores, and delivery method; mixed for prenatal care, maternal comorbid illnesses, and maternal behaviors; and low for neonatal congenital anomalies.30,31 However, a detailed study from New York State on birth certificate data found good reliability for specific maternal medical conditions and for maternal behaviors.31 Another limitation of birth certificate data is that severity-of-illness factors not measured in the data set may influence neonatal mortality rates. However, like others, our results suggested that birth weight was by far the strongest predictor of death, and research has found that birth weight is reliably coded in birth certificate data.14
The finding that, in New York City, black infants who are born too small systematically receive care in institutions with worse outcomes, compared with those where white infants receive care, demands immediate attention. Our data suggest that improving outcomes at the lowest-performing hospitals may produce the greatest benefit. Because effective treatments for prematurity exist, ensuring that such treatments are used consistently at all hospitals at which VLBW infants receive care is a vital first step toward this improvement goal. Our findings define an imperative to improve care in New York City and to study other urban areas to identify and to ameliorate such trends. The excess deaths suffered by these tiny infants and their contributions to black/white disparities are unacceptable.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Address correspondence to Elizabeth Howell, MD, MPP, Department of Health Policy, Mount Sinai School of Medicine, One Gustave L. Levy Place, Box 1077, New York, NY 10029-6574. E-mail: elizabeth.howell{at}msnyuhealth.org
The opinions, view, and conclusions expressed in this article are those of the authors and not necessarily those of the Agency for Healthcare Research and Quality, the Commonwealth Fund, or the National Center for Minority Health and Health Disparities.
The authors have indicated they have no financial relationships relevant to this article to disclose.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2008 American Academy of Pediatrics. All rights reserved. |
||
|
||
