Do Clinical Markers of Barotrauma and Oxygen Toxicity Explain Interhospital Variation in Rates of Chronic Lung Disease?
Objective. To explore the hypothesis that variation in respiratory management among newborn intensive care units (NICUs) explains differences in chronic lung disease (CLD) rates.
Design. Case–cohort study.
Setting. NICUs at 1 medical center in New York (Babies' and Children's Hospital [Babies']) and 2 in Boston (Beth Israel Hospital and Brigham and Women's Hospital [Boston]).
Study Population. Four hundred fifty-two infants born at 500 to 1500 g birth weight between January 1991 and December 1993, who were enrolled in an epidemiologic study of neonatal intracranial white matter disorders.
Case Definition. Supplemental oxygen required at 36 weeks' postmenstrual age.
Results. The prevalence rates of CLD differed substantially between the centers: 4% at Babies' and 22% at the 2 Boston hospitals, despite similar mortality rates. Initial respiratory management at Boston was more likely than at Babies' to include mechanical ventilation (75% vs 29%) and surfactant treatment (45% vs 10%). Case and control infants at Babies' were more likely than were those at Boston to have higher partial pressure of carbon dioxide and lower pH values on arterial blood gases. However, measures of oxygenation and ventilator settings among case and control infants were similar at the 2 medical centers in time-oriented logistic regression analyses. In multivariate logistic regression analyses, the initiation of mechanical ventilation was associated with increased risk of CLD: after adjusting for other potential confounding factors, the odds ratios for mechanical ventilation were 13.4 on day of birth, 9.6 on days 1 to 3, and 6.3 on days 4 to 7. Among ventilated infants, CLD risk was elevated for maximum peak inspiratory pressure >25 and maximum fraction of inspired oxygen = 1.0 on the day of birth, lowest peak inspiratory pressure >20 and maximum partial pressure of carbon dioxide >50 on days 1 to 3, and lowest white blood count <8 K on days 4 to 7. Even after adjusting for white blood count <8 K and the 4 respiratory care variables, infants in Boston continued to be at increased risk of CLD, compared with premature infants at Babies' Hospital.
Conclusion. In multivariate analyses, a number of specific measures of respiratory care practice during the first postnatal week were associated with the risk of a very low birth weight infant developing CLD. However, after adjusting for baseline risk, most of the increased risk of CLD among very low birth weight infants hospitalized at 2 Boston NICUs, compared with those at Babies' Hospital, was explained simply by the initiation of mechanical ventilation.
- newborn infant
- chronic lung disease
- bronchopulmonary dysplasia
- respiratory diseases
- newborn infant
- mechanical ventilation
Rates of chronic lung disease (CLD) among surviving premature infants vary substantially among newborn intensive care units (NICUs) even after adjustment for mortality and population-related risk factors for CLD.1–3 Recent advances in newborn intensive care, including antenatal glucocorticoid treatment and intratracheal surfactant administration, have neither reduced rates of oxygen dependence at 36 weeks' postmenstrual age (PMA),4 nor minimized disparities in CLD rates among NICUs.1,,2 In a search for NICU-specific care practices that influence the occurrence of CLD, antenatal glucocorticoid treatment was associated with reduced risk of CLD,5 while increased risk was associated with excessive fluid administration,6,,7 high inspired oxygen,6,,8 high peak inspiratory pressure (PIP),6,,8 and hypocarbia.9,,10 To evaluate the contribution of NICU care practices to the occurrence of CLD, we compared the management of neonatal respiratory failure among a large population of very low birth weight infants at 2 medical centers with differences in both styles of respiratory management and rates of CLD.
This case–cohort study of antecedents of CLD was nested in an epidemiologic study of 1607 infants born at 500 to 1500 g birth weight between September 1991 and August 1993 at 5 medical centers in Massachusetts, New York, and New Jersey.11,,12 All infants had data collected for the first 24 postnatal hours and at NICU discharge. A 30% subsample (the subcohort) had additional data collected for the remainder of the first postnatal week, as did all infants who were treated with supplemental oxygen at 28 postnatal days. The case group for this study consisted of infants either in the subcohort or in the remainder of the population who required supplemental oxygen at both 28 days' and 36 weeks' PMA. One hundred thirty-eight infants receiving supplemental oxygen at 28 days but not at 36 weeks were omitted from analyses and considered to be neither cases nor controls because of uncertainty as to how they should be classified. The final analyses compared 100 infants born at the low CLD incidence institution (Babies' and Children's Hospital, New York [Babies']) with 341 infants born at 2 Boston hospitals with higher CLD rates (Beth Israel Hospital and Brigham and Women's Hospital [Boston]). For purposes of analysis, the 2 neighboring Boston hospitals were viewed as 1 institution.
The study was designed to evaluate the relationship between NICU care practices and the occurrence of CLD among survivors. The generalized null hypothesis was that variation in rates of CLD among NICUs was not explained by differences in NICU care practices. Detailed clinical information was collected concurrently during each infant's first week of NICU hospitalization, and prenatal, perinatal, and neonatal summary data were abstracted from medical records. Because oxygen toxicity and barotrauma were considered possible explanations for inter-NICU variability in CLD rates, we collected data about the method of respiratory support, ventilator settings, fraction of inspired oxygen (Fio2), blood gas data, and oximetry values.
To investigate whether mechanical ventilation techniques varied with the medical center and case status, we evaluated ventilator settings, arterial blood gases, and oxygen saturation data on the day of birth, days 1 to 3, and days 4 to 7. Ventilator setting data were obtained only from those infants receiving mechanical ventilation during the specified periods. We were interested in the associations of CLD with oxygen toxicity (assessed by Fio2, partial pressure of oxygen [Pao2], and oxygen saturation) and barotrauma (assessed by PIP, mean airway pressure [MAwP], and partial pressure of carbon dioxide [Paco2]).
The Wilcoxon rank sum test was used for univariate comparisons of median values between groups. Categorical data were analyzed with the Fisher's exact test. For stratified analyses, the Mantel–Haenszel test was used.
To approximate the sequence of biological events, we opted for a time-oriented approach to logistic regression analysis, grouping variables by temporal epoch.13 We created multivariate models that permitted the earlier covariates of CLD to enter and not be displaced by covariates occurring in later epochs. Indicator variables were used to assess the importance of missing respiratory data points. Because we were most interested in care practices during the first postnatal week, we grouped variables according to 4 epochs: pre/perinatal, day of birth, days 1 to 3, and days 4 to 7. Candidate variables for each epoch (Table 1) were selected based on presumed biological importance and/or distribution in univariate analyses of potential confounding variables. A step-down approach was used at each epoch, preserving significant variables from previous epochs. Also retained was the variable antenatal glucocorticoid treatment, because it was considered important from a biological perspective and previous studies suggested that it modifies CLD risk.14,,15
The unadjusted rates of CLD, assessed by requirement for supplemental oxygen at 36 weeks' PMA, differed substantially between the 2 centers: 4% among the Babies' Hospital infants and 22% among those at the Boston hospitals. Furthermore, the prevalence rate of CLD was higher among the Boston population than at Babies' for each stratum of birth weight, gestational age, and ethnicity (Fig 1). The markedly different CLD rates among infants of similar birth weights at the 2 medical centers prompted us to compare their population characteristics and NICU-specific care practices.
Center-Specific Characteristics of the Study Population
Three maternal or infant characteristics were identified as potential confounding variables because they were associated with both medical center and case status: gestational age, Medicaid insurance, and ethnicity (Table 1). Gestational age categorization showed a greater proportion of infants in the >28-week gestational age category at Babies' compared with Boston (55% vs 39%; P = .02), and among control subjects compared with case subjects (55% vs 19%; P < .001). The Babies' and Boston study populations differed in the proportion of white (28% vs 64%), black (14% vs 22%), and Latino (48% vs 7%) subjects (P < .0001). Case status also differed by ethnicity. Comparing ethnic proportions among cases and controls, white subjects composed 64% and 53%, black subjects 21% and 20%, and Latinos 7% and 21%, respectively (P = .006). The 3 confounding factors were in a direction that tended to favor a lower CLD rate at Babies' compared with Boston.
We evaluated 9 additional variables describing maternal and infant characteristics. All of these were similar between the 2 medical centers, including mortality before 36 weeks' PMA, which was 9% among Babies' and 10% among Boston subjects (P = .45). There were no significant differences between the 2 study centers in rates of other morbidities, including intraventricular hemorrhage, periventricular leukomalacia, necrotizing enterocolitis, and retinopathy of prematurity.
Variation in NICU-Specific Care Practices
NICU-specific respiratory care practices were of principal interest and a number of early cardiorespiratory measures differed by both medical center and CLD case status (Table 1). Indomethacin treatment was prescribed for only 2% of the subjects at Babies' Hospital compared with 28% of Boston subjects (P < .0001) and was used in 11% of control and 45% of case subjects (P < .0001). Surfactant was administered more often in Boston (45% vs 10%; P < .0001), and was received preferentially by cases (65% vs 23%; P < .0001). Continuous positive airway pressure (CPAP) was used primarily at Babies' (63% vs 11%), whereas mechanical ventilation was used primarily at Boston (75% vs 29%; P < .0001). Overall, cases were less likely than controls to receive CPAP (1% vs 34%) and more likely to receive mechanical ventilation (77% vs 42%;P < .0001). There was minimal difference in the use of postnatal steroids between centers (4% vs 3%; P = .77).
Oxygen Toxicity and Barotrauma
Infants at Babies' were not only less likely than infants at Boston to be treated with mechanical ventilation, but those who received mechanical support did so for shorter periods (median of 13 days at Babies' vs 27 days at Boston). During all 3 periods, infants who later developed CLD received greater median Fio2 than did infants who survived without CLD (Table 2). However, infants at the 2 medical centers were similar in median Fio2, Pao2, and oxygen saturation values for infants in each diagnostic group during all 3 periods. An exception to this generalization was the tendency, before the fourth postnatal day, of infants with CLD at Babies' to have lower median PaO2s than did infants with CLD at Boston.
MAwP was the ventilator setting most consistently associated with CLD. At both medical centers during each period, infants destined to develop CLD had higher median MAwPs. On the day of birth and days 1 to 3, infants in the CLD case group at Babies' had higher median MAwPs than did case infants at Boston. Among noncase infants, MAwPs were higher at Babies' on the day of birth and days 1 to 3 but similar between noncase infants at the 2 institutions on days 4 to 7. In all time intervals, case and control infants at both medical centers had similar values of PIP and positive end expiratory pressure (PEEP).
Ventilation differed between institutions but was similar between case and control infants at each institution. At each period, median Paco2s were higher among Babies' case and control subjects than among their peers at Boston. However, within each medical center, case and control subjects had similar median lowest and mode Paco2s. Median highest Paco2s were greater for infants who went on to develop CLD at both medical centers and during all 3 periods. As expected, the elevated Paco2s were associated with median lowest pHs that were lower on the day of birth and days 1 to 3 among infants who later developed CLD.
Models were generated independently for the total population and for infants who received mechanical ventilation using a list of candidate variables (Table 3). Multivariate models for the subsequent epochs were evaluated in 2 ways. We first used a dichotomous variable for the presence or absence of mechanical ventilation during each period: day of birth, days 1 to 3, and days 4 to 7. Next, we evaluated each model assessing the contributions of ventilator settings, measures of gas exchange, and selected additional clinical variables. The odds ratios (ORs) and their 95% confidence intervals (CIs) are shown for variables included at each step, from epoch 1 (day of birth) through epoch 3 (postnatal days 4–7). All values reflect adjustment for the significant variables in the epoch 0 model and for the other variables included in the epoch.
The most parsimonious model for the pre/perinatal epoch (epoch 0) and for the day of birth (epoch 1) were similar for both the total population and the group of infants who received mechanical ventilation. The epoch 0 model included 3 statistically significant variables: gestational age, magnesium sulfate infusion during labor, and hypothyroxinemia (blood thyroxine ≤5.3 μg/dL). Antenatal glucocorticoid therapy also was added to the model because of its presumed biological importance. Each of the other variables listed in the pre/perinatal epoch section of Table 2 was tested and did not contribute significantly to the model (ie, P > .05).
The models incorporating the dichotomous mechanical ventilation variable for each of the 3 epochs showed infants receiving mechanical ventilation to be at substantially increased risk of CLD (Table 4). After adjusting for the pre/perinatal risk factors, the ORs for mechanical ventilation were 13.4, 9.6, and 6.3 for day of birth, days 1 to 3, and days 4 to 7, respectively. When the epoch models were adjusted for mechanical ventilation, the contribution of the Boston variable was reduced in each of the 3 models to ORs between 1.8 and 2.3. Restricting the analyses to the population of infants who received mechanical ventilation and adjusting for specific measures of ventilation had a similar effect on the ORs for Boston: each was reduced to a point estimate between 1.5 and 2 and the 95% CIs crossed unity.
To further evaluate the significance of treatment with mechanical ventilation, specific respiratory variables and selected potential confounding factors were evaluated using logistic regression models among the group of infants receiving mechanical ventilation (Table 5). All respiratory variables for the day of birth (epoch 1), including mode of respiratory support, ventilator settings, and blood gas data, were added to the retained epoch 0 variables. In epoch 1, only 2 variables were statistically significant: maximum PIP >25 and maximum Fio2 of 1.0 were associated with 2.2 and 1.8 times the risk of CLD, respectively, than were lower PIPs and Fio2s. Epoch 2 (days 1–3) was adjusted for the variables in epochs 0 and 1. Lowest PIP >20 (OR: 2.6) and Paco2 >50 (OR: 2.5) on days 1 to 3 were significantly associated with CLD. Adding epoch 3 (days 4–7) variables to those from the previous epochs, only white blood count <8000 (OR: 2.9) contributed significantly to the model.
Developmental immaturity is of principal importance in the pathogenesis of CLD. Structural immaturity, surfactant deficiency, and inadequate antioxidant defenses increase the likelihood of an immature infant requiring mechanical support and diminish his/her ability to cope with the potential hazards of the therapy. Pulmonary maturation is enhanced by antenatal maternal glucocorticoid treatment. To adjust for developmental immaturity, gestational age and antenatal glucocorticoid treatment were included in the epoch 0 model. Neonatal thyroxine level also was included among epoch 0 variables, although thyroxine generally is not tested until the infant is 1 to 7 days of age. We included thyroxine because hypothyroxinemia is closely associated with immaturity and severity of illness and its effects, if any, on the developing lung probably are initiated before postnatal testing. Among our study population, infants with hypothyroxinemia were twice as likely to develop CLD (adjusted OR: 2.0; 95% CI: 1.2,3.4). After adjusting for all epoch 0 risk factors, the decision to treat a premature infant with mechanical ventilation captured much of the increased CLD risk at the Boston NICUs.
In the seminal publication describing CLD among infants with hyaline membrane disease, Northway et al16 speculated that oxygen toxicity and barotrauma were responsible for the pathologic changes of bronchopulmonary dysplasia. Our data support this hypothesis and provide evidence that some of the interinstitutional variation in CLD rates among surviving very low birth weight infants can be explained by NICU-specific differences in respiratory management that might be associated with oxygen toxicity and barotrauma.
High inspired oxygen concentrations impair surfactant synthesis, exhaust the antioxidant defenses, and cause direct cellular injury to the immature lung. Several previous studies6,,8,17,18revealed an association between high inspired oxygen and CLD. In our analyses, inspired oxygen concentrations indeed were higher among case infants than controls and higher for similar infants at Boston than at Babies' Hospital. However, Pao2s and oxygen saturations were equivalent for case and control infants at each hospital, suggesting that higher Pao2is more likely to be a surrogate for another hospital-specific effect rather than a causal agent for CLD.
The role of barotrauma in CLD pathogenesis has been of interest because hypocarbia,9,,10 high ventilator pressures,6,,819–21 and pulmonary air leak8,,22 were reported to predict risk of CLD. Although a debate exists regarding the relative roles of barotrauma and volutrauma in lung injury, our use of the term barotrauma refers both to pressure and volume-induced injuries. The barotrauma hypothesis is biologically plausible because the preterm lung gas volumes per kilogram of body weight are small23 and the PIPs needed to inflate the surfactant-deficient lung often are fivefold greater than the physiologic inflation pressures of the normal lung. Barotrauma produces alveolar shear stress, disruption of alveolarization,24pulmonary air leak, and release of damaging cytokines and other biologically active substances.25 Furthermore, another known risk factor for CLD, patent ductus arteriosus (PDA),15,,26 increases the need for higher ventilator pressures because PDA causes pulmonary edema.
Although the multicenter comparison by Avery et al3 did not link a specific CLD prevalence with each of the medical centers, Babies' and Children's Hospital, Columbia University, was credited with the lowest CLD rate of the 8 study centers. The Babies' respiratory management strategy for very low birth weight infants emphasizes early and routine use of CPAP and more limited use of intubation and mechanical ventilation. Over the past 10 years, dissemination of information regarding this style of respiratory management led many NICUs, including the Boston centers in our study, to modify their approaches to respiratory support of very premature infants. However, inter-NICU differences in respiratory management not only persist but might have increased because of the recent introduction of newer technologies, such as high frequency jet and oscillatory, synchronized intermittent mandatory, assist-control, and proportional-assisted ventilation. Clinical trial data that might guide evidenced-based respiratory management are very limited. No randomized trial comparing CPAP and mechanical ventilation has been published, perhaps explaining why the Babies' and Children's Hospital style of respiratory management of premature infants has been neither completely understood nor universally adopted.
Using historical comparison groups, 2 previous observational studies showed diminished CLD rates associated with lower rates of intubation and mechanical ventilation.27,,28 Our study adds to these and other analyses of observational data the first opportunity to compare specific details of respiratory management at Babies' Hospital with practices and outcomes at another medical center with a higher rate of CLD among surviving premature infants. Among ventilated infants, higher PIP and Fio2 settings were associated with increased CLD risk. However, the dichotomous variables summarizing the presence or absence of treatment with mechanical ventilation during the 3 epochs captured the NICU-specific risk better than did the group of variables outlining the specific ventilator settings. One explanation for this apparent paradox is that, among the mechanically ventilated infants, the impact of the ventilation-related variables was reduced by restricting the population to the group of infants whose pulmonary illness was most severe.
Evaluating specific ventilator settings and blood gas data for ventilated infants, we found maximum PIP >25 on the day of birth and minimum PIP >20 on days 1 to 3 were associated with increased CLD risk. The Paco2 values associated with increased risk of CLD occurred on days 1 to 3 and 4 to 7, and these were indicators of hypercarbia rather than hypocarbia. Thus, our findings are inconsistent with those of previous investigators9,,10,29 who reported increased CLD risk among infants with lower Paco2 values. Together, these findings suggest that greater severity of respiratory illness contributed to CLD risk.
We considered alternative explanations for our study results. We did not find that improved survival accounted for differential CLD rates. However, the inherent need for greater respiratory support among infants with the most severe surfactant deficiency30potentially confounds the interpretation of the association between high inspired oxygen or ventilator PIP and the development of CLD. The lower rate of confirmed PDA at Babies' Hospital, reflected in diminished use of indomethacin, might have reduced the need for mechanical ventilatory support. Thus, confounding by indication cannot be excluded as a partial explanation for our study results.31 Other unmeasured confounding factors, such as ureaplasma urealyticum infection32,,33 or vitamin A intake34 could have influenced our findings.
At the initiation of our study, methods that adjust for severity of illness, including the Score for Neonatal Acute Physiology (SNAP)35 and the Clinical Risk Index for Babies (CRIB),36 were validated only for neonatal mortality. Therefore, the methods were not used to adjust for severity of illness among our study subjects. A recent abstract suggests that SNAP also predicts CLD risk.37 The impact of SNAP or CRIB on studies exploring the relationships between interinstitutional care practices and neonatal morbidities deserves further study.
Among our study subjects, NICU-specific risk of CLD was predominantly associated with the decision to use mechanical ventilation. Among those receiving mechanical ventilation, high inspired oxygen concentrations, high ventilator PIPs, and hypercarbia during the first postnatal week were associated with increased CLD risk. Although our analyses adjusted for a number of demographic, prenatal and perinatal variables, and additional therapies, the observational nature of the data makes it impossible to determine whether each respiratory factor is truly pathogenic or simply provides a surrogate for another factor in the causal pathway leading to CLD. Although excessive administration of oxygen and/or barotrauma might increase CLD risk, we found only partial evidence that either of these explains the observed interinstitutional variation in CLD rates. We found no support for the hypotheses that hypocarbia predisposes to or permissive hypercarbia protects from CLD. Judicious use of inspired oxygen and PIP might reduce CLD risk. However, our study results offer little evidence that adjusting ventilator settings with the specific goal of achieving hypercarbia has a beneficial pulmonary effect. Infants at the 2 Boston hospitals continued to be at increased risk of CLD compared with those at Babies' Hospital, even after adjustment for all significant factors in the multivariate models. Thus, among our study population, differences in respiratory management strategies incompletely explained interinstitutional variation in CLD rates. Ongoing efforts are needed to identify other risk factors and care practices that might constitute alternative preventive strategies for CLD among premature infants.
This project was supported by Grant NS 27306 from the National Institutes of Neurological Disorders and Stroke and Grant HL 56398 from the National Institutes of Heart Lung and Blood.
We thank the women and babies who participated in this study. We thank our nursing, respiratory therapy, and neonatology colleagues at Brigham and Women's, Beth Israel-Deaconess, and Babies' and Children's Hospitals, without whose dedication and support this project would not have been possible. We also thank Kathleen Holahan and Julie Ristaino for assistance with manuscript preparation; Gloria Phillip, Evelyn Henry, and Yolene Semexant for medical record retrieval; and Alicia Harshfield, Mary Jo White, and Ruta Lew for subject recruitment and supplemental data collection.
- Received April 12, 1999.
- Accepted September 9, 1999.
- Address correspondence to Linda J. Van Marter, MD, MPH, Hunnewell 438, Children's Hospital, 300 Longwood Ave, Boston, MA 02115. E-mail:
↵§§ The members of the Neonatology Committee include Thomas Hegyi, MD, and Mark Hiatt, MD, from St Peter's Medical Center, New Brunswick, New Jersey and Robert Wood Johnson Medical School, Princeton, New Jersey; Ulana Sanocka, MD, from Columbia University and Babies' and Children's Hospital, New York, New York; Farrokh Shahrivar, MD, at Columbia University and St Luke's-Roosevelt Medical Center, New York, New York; and Linda J. Van Marter, MD, MPH.
- CLD =
- chronic lung disease •
- NICU =
- newborn intensive care unit •
- PMA =
- postmenstrual age •
- PIP =
- peak inspiratory pressure •
- PEEP =
- positive end expiratory pressure •
- Babies' =
- Babies' and Children's Hospital, New York, New York •
- Boston =
- Beth Israel Hospital and Brigham and Women's Hospital, Boston, Massachusetts •
- Fio2 =
- fraction of inspired oxygen •
- Pao2 =
- partial pressure of oxygen •
- MAwP =
- mean airway pressure •
- Paco2 =
- partial pressure of carbon dioxide •
- CPAP =
- continuous positive airway pressure •
- OR =
- odds ratio •
- CI =
- confidence interval •
- PDA =
- patent ductus arteriosus •
- SNAP =
- Score for Neonatal Acute Physiology •
- CRIB =
- Clinical Risk Index for Babies
- Hack M,
- Friedman H,
- Fanaroff AA
- Avery ME,
- Tooley WH,
- Keller JB,
- et al.
- Horbar JD,
- McAuliffe TL,
- Adler SM
- Van Marter LJ,
- Leviton A,
- Kuban KCK,
- Pagano M,
- Allred EN
- ↵Leviton A, Paneth N, Reuss ML, et al. Maternal receipt of magnesium sulfate does not seem to reduce the risk of neonatal white matter damage. Pediatrics 1997;99(4). URL: http://www.pediatrics.org/cgi/content/full/99/4/e2
- ↵Miettinen O. Theoretical Epidemiology. New York, NY: John Wiley and Sons; 1985:40
- Van Marter LJ,
- Leviton A,
- Kuban KCK,
- Pagano M,
- Allred E
- Ryan SW,
- Wild NJ,
- Arthur RJ,
- Shaw BNJ
- Shennan AT,
- Dunn MS,
- Ohlsson A,
- Lennox K,
- Hoskins EM
- Philip AGS
- Sinkin RA,
- Cox C,
- Phelps DL
- Van Lierde S,
- Smith J,
- Devlieger H,
- Eggermont E
- Schmidt B,
- Davis P,
- La Pointe H, et al
- Poets CF,
- Sens B
- Mariani G,
- Cifuentes J,
- Carlo WA
- ↵Miettinen O. Efficacy of therapeutic practice: will epidemiology provide the answers? In: Melmon KL, ed. Drug Therapeutics Concepts for Physicians. New York, NY: Elsevier; 1981:201–208
- Richardson DK,
- Gray JE,
- McCormick MC,
- Workman K,
- Goldmann DA
- Tarnow-Mordi W,
- Ogston S,
- Wilkinson AR,
- et al.
- Weisberger SA,
- Richardson DK,
- Allred EN,
- et al.
- Copyright © 2000 American Academy of Pediatrics