Rate of bronchopulmonary dysplasia as a function of neonatal intensive care practices

FETAL AND NEONATAL MEDICINE

Rate of bronchopulmonary dysplasia a function of neonatal intensive care practices

as

Linda J. V a n Marter, MD, MPH, M a r c e l l o P a g a n o , PhD, Elizabeth N. AIIred, MS, Alan Leviton, MD, MS, a n d Karl C. K. Kuban, MD, MS From the Division of Newborn Medicine and Neuroepidemiology, Children's Hospital, and the Department of Biostatistics, Harvard School of Public Health, Boston, Massachusetts Some differences a m o n g neonatal intensive care units (NICUs) in i n c i d e n c e of b r o n c h o p u l m o n a r y dysplasia may reflect.variations in medical care practices. After adjusting for differences in the inherent risk of bronchopulmonary dysplasia a m o n g 223 infants of less than 1751 grn birth weight who were admitted to three Harvard-affiliated NICUs, we used multivariate analysis to explore the extent to which medical care practices during the first days of life varied with the rate of bronchopulmonary dysplasia. In our analyses, variables were g r o u p e d by three major hypotheses: o x y g e n toxicity, barotrauma, and fluid overload. The NICU d e s i g n a t e d I (the one with the highest rate of b r o n c h o p u l m o n a r y dysplasia) used much higher than e x p e c t e d c o l l o i d a l volumes during the first 4 days of life; in contrast, in the NICU d e s i g n a t e d 3 (the one with the lowest rate of b r o n c h o p u l m o n a r y dysplasia), infants consistently received lower than e x p e c t e d amounts of c o l l o i d a l solution. Signs of patent ductus arteriosus were also much more frequent than e x p e c t e d during this time at NICU I; rates were much lower than predicted at NICU 2 and were near predicted values at NICU 3. Maximum inspired o x y g e n fraction during the first 4 days varied significantly in a direction inconsistent with the o x y g e n toxicity hypothesis. Maximum arterial o x y g e n tension was significantly less than e x p e c t e d at the hospital with the lowest rate of bronchopulmonary dysplasia (NICU 3). None of six m e d i c a l care practices indicating potential for barotrauma varied with NICU e x c e p t for positive end-expiratory pressure, which varied in a direction suggesting a protective effect against bronchopulmonary dysplasia. These findings agree best with the hypothesis that differences in hydration during the first days of life a c c o u n t for some of the difference a m o n g NICUs in b r o n c h o p u l m o n a r y dysplasia occurrence. (J PEDIATR1992;120:938-46)

Supported by the Hearst Foundation; the National FoundationMarch of Dimes; Mead Johnson Nutritional Division; the Milton Fund of Harvard University; grants No. NS 20807 and No. NS 20658 from the National Institute of Neurological and Communicative Disorders and Stroke; grants No. HL 40454 and No. SCOR 2P50HL34616-03 from the National Heart, Lung, and Blood Institute; and Mental Retardation Center grant No. HD 06276 from the National Institute of Child Health and Human Development.

938

Dr. Van Marter also supported by a Charles A. King Trust fellowship from the Medical Foundation. Submitted for publication Nov. 26, 1990; accepted Dec. 30, 1991. Reprint requests: L. J. Van Marter, MD, MPH, Joint Program in Neonatology, Hunnewell 4, Children's Hospital, 300 Longwood Ave., Boston, MA 02115.

9/23/35938

Volume 120 Number 6

BPD F~o2 NICU Paco2 Pao2 PDA PIP RS

BPD and NICU practices

Bronchopulmonarydysplasia Fraction of inspired oxygen Neonatal intensive care unit Arterial carbon dioxide tension Arterial oxygen tension Patent ductus arteriosus Peak inspiratory pressure Risk score

Rates of occurrence of bronchopulmonary dysplasia among neonatal intensive care unit survivors vary among tertiary care centers. 1 This observation is compatible with the hypothesis that medical care practices influence the likelihood of BPD. Recently, speculation regarding the origin of BPD has focused on three mechanisms through which intensive care practices may produce adverse pulmonary effects: toxic reaction to oxygen, barotrauma, and fluid overload. 1-3 Our finding of significantly different rates of BPD among three Harvard-affiliated NICUs 3 prompted us to attempt to identify care practices associated with the rate of BPD. METHODS This study was a part of a prospective, randomized clinical trial of phenobarbital prophylaxis against neonatal intracranial hemorrhage. 4 The trial prospectively enrolled 280 infants treated in the NICUs of three Harvard-affiliated hospitals (Brigham and Women's Hospital, The Children's Hospital, and Massachusetts General Hospital) between June 1981 and April 1984. Infants were recruited for the phenobarbital study if they required intubation within the first 12 hours of life, weighed less than 1751 gm at birth, and had no intracranial hemorrhage demonstrable by cranial ultrasonography before 12 hours of age. Previous studies of the determinants of BPD have employed either clinical1'2 or radiologic5'6 diagnostic criteria. The most widely accepted clinical criterion is requirement of supplemental oxygen for 28 days or longer. We chose as an additional criterion the presence of stage II, III, or IV BPD in the radiographic findings, as described by Northway et al. 5 and Edwards. 6 The 76 infants whose cases met both clinical and radiographic criteria were considered to have BPD. The referent group consisted of 147 infants who survived until the end of the first postnatal month and did not have either of the two BPD criteria. Thirty-one infants were excluded because they died before postnatal day 28 and were therefore ineligible under the BPD criteria. An additional 26 infants (15% of the surviving group .without BPD) were excluded from analysis because they had one but not both of the BPD criteria and thus represented an unclear diagnostic group.

939

Because postnatal predictors of BPD risk are evident within the first few postnatal days, 2, 3 we focused on this period. Our interest in NICU-specific care practices originated in part from the persistent significance of the NICU variable at each stage of our multivariate analyses that adjusted for other risk factors for BPD, including exposures and characteristics during the prenatal and intranatal periods. The generalized form of the null hypothesis is that a given medical care practice not associated with the risk of BPD. If we were to establish an association between any measure of medical care and BPD risk in this ecologic study, we needed to demonstrate that the variation in BPD incidence among hospitals varied with (in the same order or direction and with similar magnitude) the measure of the given medical care practice. Preliminary analyses were performed with the chi-square test, the Fisher Exact Test, the MantelHaenszel test for stratified data, and the Wilcoxon ranksum test. Finally, we created logistic regression models with variables grouped sequentially by prenatal, perinatal, and postnatal epochs. 7 Decisions about care reflect physicians' perceptions of the need for different levels of care. For example, the "sicker" the baby is, the more likely a physician is to treat that baby "aggressively." Medical care practice variables and measures of their primary effects can thus convey information about the indications for specific care practices. The bias of attributing to medical care the influence that should be attributed to the indications for that care has been labeled confounding by indication. 7 We were fortunate to have observations from three different NICUs with differing care practices. We used logistic re~ression to adjust for differences in population characteristics associated with "inherent" BPD risk and minimize the bias of confounding by indication.8 We therefore modeled BPD risk as a function of antenatal and perinatal characteristics evident before the initiation of any medical care. This model allowed us to calculate a risk score for each infant equivalent to the log odds of BPD (Pagano M, et at: unpublished data). In this study we used RS to measure the intrinsic risk of BPD for each infant.8 Specifically, RS is the sum of products of the respective coefficients in the final perinatal logistic regression model multiplied by the values of the covariates for each individual. The RS gives a relative risk of BPD; the higher the RS, the greater the risk of BPD. We asked the following questions: (1) How well does our model predict BPD? (2) Do care practices differ at the three hospitals during the first few days of life? (3) If care practices are different, do they vary with the observed rate of BPD (after adjustment for the RS)?

940

Van M a r t e r et al.

The Journal o f Pediatrics June 1992

Table. Variables in RS that describe infants enrolled in the controlled clinical trial of phenobarbital prophylaxis for intracranial hemorrhage conducted at three NICUs

Birth weight (gm)* Gestational age (wk)* Female (%) Corticosteroid treatment (%)]" Toxemia (%)]" Intubation in delivery room (%)]" BPD rate (%) Observed]" Predicted]"

NICU I

NICU 2

NICU 3

1118 • 312 29.8 + 2.3 47.5 31.1 4.9 52.5

1166 + 319 30.0 _+ 2.3 54.8 38.1 18.3 35.2

1291 -!-_303 30.9 _+ 2.4 50.0 13.9 8.3 31.4

49.2 42.0

31.7 ' 31.0

17.0 25.0

*Mean _+SD. tp <0.05

We assumed that the babies with the highest inherent risk of BPD would also be at highest risk for exposure to medical care practices that may influence the risk of BPD. T o balance the mix of patients in each NICU, we calculated RS-adjusted mean values for practice-related variables at each NICU and thus obtained predicted values based on the inherent risks of BPD at each hospital. Comparison of these predicted practice variables with actual observations then formed a ratio of observed value to predicted value. Because each NICU's observed/expected ratios for the first 4 days of life varied minimally, the four daily ratios were averaged, yielding a summary ratio for each NICU. One of the most compelling indicators of a causal relationship is evidence of a dose-response relationship. We therefore sought dose-response relationships between the hypothesis-linked variables at each NICU and the risk of BPD. We predicted that the ratios favoring higher than expected rates of adverse exposure would parallel the rates of BPD. For example, the N I C U with the highest ratio of observed to predicted change in medical care practice measures would also be expected to have the highest incidence of BPD; the NICU with the lowest ratio would have the lowest incidence of BPD. We grouped care practices according to data consistent with three hypotheses of greatest interest in the cause of BPD: oxygen toxicity, barotrauma, and fluid overload. If the oxygen toxicity hypothesis is correct, the incidence of BPD at each N I C U should parallel the propensity of the staff at that NICU toward administration of high inspired oxygen concentrations during the first days of the infant's life. Factors in the first 4 days of life that we considered in the oxygen toxicity hypothesis were as follows: highest F~02, highest arterial oxygen tension, lowest Pao2, and the ratio of highest FIO2 to highest Pa02. Excessive barotrauma was a second major area of interest. According to this hypothesis, NICUs in which staff ex-

posed infants to greater mean airway pressures or were more aggressive in supplying ventilatory support to their patients would have higher rates of BPD. In the absence of recorded mean airway pressure data we considered the following variables in assessment of potential for barotrauma: presence of pneumothorax or pulmonary interstitial emphysema; lowest arterial carbon dioxide tension; highest Paco2; mean Pat02; and mean maximum values for peak inspiratory pressure, positive end-expiratory pressure, continuous positive airway pressure, and ventilator rate. The final hypothesis evaluated was that excessive fluid therapy is a risk factor for BPD. In this set of analyses we included daily birth weight-adjusted intake volumes of crystalloid and colloidal solutions, lowest and highest serum sodium levels, furosemide and pancuronium treatment, and clinical signs of patent ductus arteriosus. According to this hypothesis, infants at the NICU with the highest BPD rate excess (beyond that predicted by RS) would also have the highest mean volumes of crystalloid and colloid fluids and as a consequence would have the highest frequency of PDA, highest frequency of furosemide therapy, and lowest serum sodium values. To determine to what extent variation in BPD incidence among NICUs can be explained by these presumed mechanisms, we adjusted risk ratios for BPD that compared NICUs 1 and 2 with NICU 3 for the variables assigned to each hypothesis. Separate analyses were carried out for each individual variable and for combinations of variables. RESULTS Step-wise logistic regression showed the following prenatal and perinatal factors to be significant in the final, most parsimonious, multivariate model: birth weight, gestational age, antenatal maternal glucocorticoid treatment, tracheal intubation in the delivery room, maternal toxemia, and ad-

Volume 120 Number 6

BPD and N I C U practices

941

1.50

1.25

R a t i o

1.00

,).75

0.50

r

,

I

NXCU #I

NICU #2

NICU #3

A

Maximum

FI02

1.50

1.50

1.25

1.25

R

1.00

1.00 i o

0.75

0.75

0.50

B

T N I C U #I

0.50

NIcu #2 Maximum

NIcu #3

Pa02

NIcu ~I C

NXCU #z

~ICU #3

M a x FI02 / M a x Pa02

Fig. I. Observed/expected ratios summarizing mean values for first 4 days of life for the three significant variables of four variables assigned to oxygen toxicity hypothesis and differing significantly among NICUs: mean maximum FIO2, maximum Pao2, and the ratio of the two variables. NICU 1 has highest rate of BPD; NICU 3 has lowest rate of BPD. Although distributions for FIo2 and ratio of maximum FIo2 to maximum Pao2 are differing significantly at the p <0.05 level, the direction of these ratios is opposite to that expected if oxygen toxicity accounts for NICU differences in BPD incidence.

mitting N I C U . Each of these variables contributed to the model at p <0.05. All significant variables in the final multivariate model except for admitting N I C U were included in the calculation of each infant's RS, a measure of expected risk of BPD that reflected characteristics evident before the initiation of neonatal intensive care. On the basis of the RS

for each infant, the predicted rates of BPD a t N I C U s 1, 2, and 3 were 42%, 34%, and 25%, respectively. These rates were compared with observed rates of 49%, 32%, and 17%, respectively. W e concluded from these data that our model provided a reasonably good predictor of BPD risk. Despite the wide range in BPD rates (17% to 49%)i the

942

Van Marter et al.

The Journal of Pediatrics June 1992

1.50

1.25

1.00

0.75

0.50

A

NICU #I

t

r

.NICU #2

NICU #3

Lowest PaC02

t.s0

i.6o

t .2~

t .25

R

R

1.00

1.00 I o

i o

0.75

0.70

0.50

B

I

I

i

Nxcv #1

~ c u #z

Ntc. #s

Mean PaC02

0.50

C

i

i

I

NXCU#~

.Icv #2

NICU ~3

Maximum PEEP

Fig. 2. Observed/expected ratios summarizing mean values for first 4 days of life for the three significant variables of six variables assigned to barotrauma hypothesis. Although differing significantlyamong NICUs, none of these three variables-mean lowest Paco2, mean Paco2, and maximum positive end-expiratory pressure (PEEP)--varied in a direction consistent with rates of BPD.

NICUs did not differ significantly in the distributions of birth weight, gestational age, or gender (Table). The racial distribution also did not differ significantlyamong the three NICUs. Mothers of infants at N I C U 1 (with the highest BPD prevalence) were least likely to have had toxemia during pregnancy; those at N I C U 2 (with intermediate BPD prevalence) were most likely to have had toxemia during pregnancy (p = 0.026). The rate of antenatal glucocorti-

coid treatment was lowest at N I C U 3 (with the lowest BPD prevalence) and highest at N I C U 2 (with the intermediate rate of BPD). Infants at N I C U 3 were least likely to have required intubation in the delivery room; those at N I C U 1 were most likely to have required intubation in the delivery room (p = 0.045). Oxygen toxicity. Mean maximum FIO2 and Pao2 and the ratio of these two variables (Fig. 1) differed significantly

Volume 120 Number 6

BPD and NICU practices

943

t .50

1.25

1.00

0.75

0.50

A

i

i

i

NIcu #1 Nxcu #z Nxcv #3 Crystalloid Intake

1.50 2.0

1.25 1.5

III

R

1.oo~ i o

i o

t,0 0.75

0,50

B

t

I

I

mctr #i

m c v #2

N~cu #3

ColloidIntake

0.5

.=~

[

t

NICV#i

I

f

NICU #~

m c u ~3

Signs of PDA

3. Observed/expected ratios summarizing mean .values for first 4 days of life for the three significant variables of six variables assigned to excessive-hydration hypothesis. Distribution of observed-expected ratios for colloidal fluid administration is exactly that anticipated if over zealous hydration contributes to NICU differences in BPD indicidence. Ratios for PDA and crystalloid fluids also vary in the anticipated d~rections for two of three hospitals and are near null value for the third, Fig.

among the three N I C U s during the first 4 days of life. For mean maximum FI02 and the ratio of maximum F1o2 to Pa02, however, the hospital with the consistently highest observed/expected ratio had the lowest rather than the highest BPD rate. Only the ratio of observed to expected mean maximum Pao2 varied significantly in approximately the anticipated direction among the three N1CUs; both N I C U s 1 and 2 had near-expected values and N I C U 3 had a lower than expected value on average.

Barotrauma. According to the barotrauma hypothesis, higher ventilator settings, excessive ventilatory support, or both should parallel the rate of BPD. Mean lower Pac02, mean Pac02, and maximum positive end-expiratory pressure differed significantly among N I C U s (Fig. 2). However, the observed/expected ratios for these and the other three variables evaluated as measures of b a r o t r a u m a - - m a x i m u m ventilator PIP, maximum ventilator rate, and highest P a f o 2 - - d i d not vary among our N I C U s in a direction con-

944

Van M a r t e r et al.

sistent with the rates of BPD. Similarly, the cumulative occurrences during the first 4 days of life of pneumothorax, pulmonary interstitial emphysema, or both were not significantly associated with the risk of BPD. Fluids. If the overhydration hypothesis is correct, babies at the NICUs with the highest BPD prevalence would have received the highest mean volumes of crystalloid and colloidal fluids and as a consequence would have had the highest percentage weight gain, highest frequency of signs of PDA, highest rate of furosemide therapy, and lowest serum sodium values. Of the variables significantly linked with the fluid-overload hypothesis (Fig. 3), only colloidal fluid administration exhibited an observed/expected ratio that precisely paralleled that of BPD incidence. Clinical signs of PDA differed significantly in the anticipated direction in two of the three NICUs; the third NICU's values were near the expected values. Crystalloid fluid administration was also significantly associated in a pattern similar to that for PDA but with less-marked differences among NICUs. Rate of furosemide treatment and lowest and highest serum sodium values did not differ significantly among NICUs. DISCUSSION Our multivariate (RS-adjusted) model predicted the rate of BPD at each NICU with a high degree of accuracy. However, adding a variable to the model to identify the NICU provided a significant increase in information about BPD risk. This phenomenon led us to speculate that some of the NICU-associated BPD risk was attributable to specific care practices. Variation in certain care practices among our study's NICUs enabled us to evaluate this speculation. Of the three etiologic hypotheses that we considered (oxygen toxicity, barotrauma, and excessive fluid therapy), excessive fluid therapy appeared to be the most consistent with the hypothesis that some of the NICU-BPD association reflects differences in medical care practices among NICUs. Not all of the hydration variables, including some linked with BPD in previous analyses,3varied significantlywith the NICU-specific rate of BPD. During the first 4 days of life, however, N I C U trends in birth weight-adjusted colloidal intake (and to a lesser extent crystalloid intake) and signs of PDA varied predominantly in a direction consistent with NICU rates of BPD. These data are consistent with the hypothesis that "excessive" administration of fluids during the first days of life is related to the pathogenesis of BPD. Smith et al. 9 and Tooley I~ were among the first to suggest that fluid overload may be important in the pathogenesis of BPD. Many lines of evidence have linked conditions

The Journal o f Pediatrics June 1992

associated with fluid overload, including pulmonary edema 11, 12 and PDA, 13-17with adverse neonatal outcomes, including chronic lung disease. Other supporting evidence includes morphologic documentation that lung edema complicates respiratory distress syndrome. 12' 18,19 A recent multicenter, multivariate, risk-adjusted casecontrol study also found an association between higher fluid intake before 96 hours of age and oxygen dependence on day 30 of life.2~ Two other studies did not find significant relationships between hydration and BPD. The first, a clinical trial by Lorenz et al., 21 randomly assigned 88 infants weighing 750 to 1500 gm at birth to receive "low" or "high" fluid intakes. No significant differences were seen in a number of neonatal outcomes, including BPD. A recent casecontrol analysis by Kraybill et al. 2 of 147 surviving infants who received ventilatory support and weighed 751 to 1000 gm at birth found no association between BPD and mean weight loss, therapy with furosemide, or pancuronium administration. The differences in study results may relate to the study designs. Neither Lorenz et ai. 21 nor Kraybill et al. 2 conducted analyses incorporating a score to control for intrinsic BPD risk. In addition, the infants in both previous studies were maintained within a narrower range of hydration than were our subjects. Fluid overload and its correlates may simply be a proxy for other significant causes of BPD. The relative magnitude of the observed association with BPD was greater for colloidal than for crystalloid fluids. This difference prompts us to attribute some of the significant associations between the fluid hypothesis variables and BPD to severity of illness-speculation supported in part by our analyses of mean arterial blood pressure and lowest pH. Sicker babies may have hemodynamic instability or greater blood loss from frequent phlebotomy, factors that may result in higher requirements for colloidal fluids. Enhanced capillary permeability in infants who are sicker may also explain some associations, including inappropriate weight gain and low serum sodium concentration. It is possible that colloid administration is an indirect pathophysiologic result of leakage into the pulmonary interstitial fluid. Another possible confounder in the association between BPD and higher fluid intake is birth weight. Smaller babies have proportionately greater insensible water losses than do heavier, larger, more mature babies. Both the multivariate model and the fluid-intake variables were adjusted for birth weight in our analyses. The highest mean intake of colloidal fluids and the highest incidence of PDA among babies in the N i c u with the highest incidence of BPD are therefore probably not mere reflections of the proportionately greater fluid requirements of infants with the lowest birth weights.

Volume 120 Number 6

Both oxygen toxicity32 and barotrauma 3~ 31 have long been considered significant pathogenic factors for BPD. Indeed, in the seminal article describing BPD, Northway et al. 5 raised both oxygen toxicity and barotrauma as potentially responsible for the new syndrome. Features of mechanical ventilation thought to promote both barotrauma and BPD have included the use of positive-pressure intermittent mandatory ventilation,22 longer duration of intermittent mandatory ventilation,22 higher PIP, 23, 24 higher ventilator rate, 2 and lower positive end-expiratory pressure. 2 Avery et al., 1 in a multicenter descriptive study, they detected markedly different rates of BPD among 10 level III NICUs. Like us, they found the disparity to be unexplained by birth weight, race, and gender distributions among the nurseries. These investigators found the lowest rate of BPD in the N I C U that placed the greatest emphasis on avoiding mechanical ventilation and minimizing ventilator rates and pressures. Further credence was given to the barotrauma hypothesis by the case-control analysis of Kraybill et al. 2 of 147 infants receiving ventilatory support who survived for at least 30 days. They found that both highest Pac02 (<40 mm Hg) at 48 or 96 hours of life and a higher ventilator rate at 96 hours of life were associated with the development of BPD. In our study, however, the levels of maximum PIP, ventilator rate, and Pac02 did not differ significantly during the first 4 days of life among the three study NICUs. Our oxygen toxicity analyses were intriguing because the Fro2 results were entirely counterintuitive, suggesting that higher FI02 protects against BPD. No published clinical or basic scientific data are consistent with this hypothesis, Investigators continue to study the factors responsible for the observed species-specific 22'25 and age-dependent22'24'26 variation in the observed dose-response relationship. Pulmonary oxygen toxic effects have been well described in adult human beings and animals. Despite reports of decreased risk of oxygen toxic effects in immature animals,24 many investigators continue to favor oxygen toxicity as a primary causal factor in BPD among surviving premature infants--possiblyas a result of early gestational deficiencies in antioxidant enzymes.6' 22, 26, 27 Variations in NICU care practices cannot be correlated with the occurrence of BPD if the practices of interest do not vary significantly among NICUs. The results of our analyses suggest that oxygen toxicity and barotrauma are less likely than fluid overload to explain the differential rates of BPD at our three Harvard-affiliated tertiary care NICUs. However, mechanical ventilatory care practices in our three study NICUs may be sufficiently homogeneous to obscure significant associations between mechanical ventilation or arterial blood gas variables and BPD. Fluid management

BPD and NICU practices

945

may simply vary more. Another possibility is that significant associations exist but become evident only after the first 4 days of life. Finally, our measures of clinical ventilatory practices may be too crude to detect significant differences. Our failure to eliminate NICU differences in BPD incidence by adjusting for variables assigned to the three casual hypotheses has at least two possible alternative explanations: ( 1) Our variables inadequately capture the essence of oxygen toxicity or barotrauma. (2) Some of the NICU associated differences in BPD incidence reflect other uninvestigated but biologically important differences in care. In summary, no trends were seen in the directions anticipated for either the oxygen toxicity or the barotrauma hypothesis. On the other hand, significant associations in a direction predicted by the hypothesis were observed for three correlates of fluid therapy: birth weight-adjusted intake of crystalloid and of colloidal fluids and signs of PDA. If confirmed, these results provide the basis for further studies of the pertinent pathophysiologic antecedents of BPD and of the potential benefits of modifying current NICU care practices. We thank Judy Sparrow for medical record retrieval, Susan Longobucco-Hynes and Annette Roberts for supplemental data collection, and Dr. Mary Ellen Avery for support and encouragement. We also thank Ms. Kathy Finn-Sullivan and Drs. Elizabeth Brown, Kenneth Huff, and Kalpathy Krishnamoorthy for their work on the original randomized trial. REFERENCES

1. AveryM, Tooley W, Keller J, Hurd S, Bryan MH, Cotton RB, et al. Is chronic lung disease in prematurely-born infants preventable? Pediatrics 1987;79:26-30. 2. Kraybill EN, Runyan DK, Bose CL, Khan JH. Risk factors for chronic lung disease in infants with birth weights of 751 to 1000 grams. J PEDIATR1989;115:155-20. 3. Van Marter LJ, Leviton A, Allred EN, Pagano M, Kuban KCK. Hydration during the first days of life and the risk of bronchopulmonarydysplasia in low birth weight infants. J PEDIATR1990;l 16:942-9. 4. Kuban K, Leviton A, Krishnamoorthy K, Brown E, Teele RL, Baglivo J. Neonatal intracranial hemorrhage and phenobarbital. Pediatrics 1986;77:443-50. 5. Northway W, Rosan R, Porter D, Pulmonary disease following respiratory therapy of hyaline membrane disease. N Engl J Med 1967;276:357-68. 6. Edwards D. Radiographic aspects of bronchopulmonary dysplasia. J PEDIATR1979;95:823-9. 7. Miettinen O. Theoretical epidemiology. New York: John Wiley, 1985:40. 8. Miettinen O. Stratification by a multivariate confounderscore. Am J Epidemiol 1976;104:609-20. 9. Smith CA, Yudkin S, Young W, MinkowskiA, Cushman M. Adjustment of electrolytes and water following premature birth (with special reference to edema). Pediatrics 1949;3:3448.

946

Van M a r t e r et al.

10. Tooley W. Epidemiology of bronchopulmonary dysplasia. J PEDIATR 1979;95:851-5. 11. Brown E, Stark A, Sosenko I, Lawson E, Avery ME. Bronehopulmonary dysplasia: possible relationship to pulmonary edema. J PEDIATR 1978;92:982-4. 12. Bland R. Edema formation in the lungs and its relationship to neonatal respiratory distress. Acta Paediatr Scand 1983; 103(Suppl):92-9. 13. Gay JH, Daily W J, Meyer BH, et al. Ligation of the patent ductus arteriosus in premature infants: report of 45 cases. J Pediatr Surg 1973;8:677-83, 14. Naulty C, Horn S, Conry J, Avery G. Improved lung compliance after ligation of patent ductus arteriosus in hyaline membrane disease. J PEDIATR 1978;93:682-4. 15. Bell EF, Warburton D, Stonestreet BS, Oh W. High-volume fluid intake predisposes premature infants to necrotising enterocolitis [Letter]. Lancet 1979;2:90. 16. Brown E. Increased risk of bronchopulmonary dysplasia in infants with patent ductus arteriosus. J PEDIATR 1979;95:865-6. 17. Jacob J, Gluck L, DiSessa T, Edwards D, Kulovich M, Kuvlinski J. The contribution of PDA in the neonate with severe: RDS. J PEDIATR 1980;96:79-87. 18. deSa D. Pulmonary fluid content in infants with respiratory distress. J Pathol 1969;97:469-78. 19. Bland R. Edema formation in the newborn lung. Clin Perinatol 1982;9:593-609. 20. Palta M, Gabbert D, Weinstein MR, Peters ME. Multivariate assessment of traditional risk factors for chronical lung disease in very low birth weight neonates. J PEDIATR 199 l;119:285-92. 21. Lorenz JM, Kleinman LI, Kotagal UR, Reller MD. Water balance in very low-birth-weight infants: relationship to water and sodium intake and effect on outcome. J PEDIATR 1982; 101:423-32.

The Journal o f Pediatrics June 1992

22. Frank L, Sosenko I. Prenatal development of lung antioxidant enzymes in four species. J PEDIATR 1987;110:106-10. 23. Berg TK, Pagtakhan RD, Reed MH, Langston C, Chernick V. Bronchopulmonary dysplasia and lung rupture in hyaline membrane disease: influence of continuous distending pressure. Pediatrics 1975;55:51-4. 24. Northway WH Jr, Petriceks R, Canty E, Bensch KG. Maturation as a factor in pulmonary oxygen toxicity: a preliminary report. J PEDIATR 1979;95:859-64. 25. Merritt T. Oxygen exposure in the newborn guinea pig lung lavage cell populations, chemotaxis, and elastase response: a possible relationship to neonatal bronchopulmonary dysplasia. Pediatr Res 1982;16:798-805. 26. Frank L, Sosenko I. Development of lung antioxidant enzyme system in late gestation: possible implication for the prematurely born infant. J PEDIATR 1987;110:9-14. 27. Rhodes P, Hall R, Leonidas J. Chronic pulmonary disease in infants with assisted ventilation. Pediatrics 1975;55:788-96. 28. Kleinbaum D, Kupper L, Morgenstern H. Epidemiologic research. London: Wadsworth, 1982. 29. Hodson W, Truog W, Mayoek D, Lyrene R, Woodrum DE. Bronchopulmonary dysplasia: the need for epidemiologic studies. J PEDIATR 1982; 101:848-51. 30. Stahlman M, Hedvall G, Lindstrom D, Snell J. Role of hyaline membrane disease in production of later childhood lung abnormalities. Pediatrics 1983;103:634-7. 31. Rhodes PG, Graves GR, Patel DM, Campbell SB, Blumenthal BI. Minimizing pneumothorax and bronchopulmonary dysplasia in ventilated infants with hyaline membrane disease. J PEDIATR 1983;103:634-7. 32. Sullivan J. Iron, plasma antioxidants and the "oxygen radical disease of prematurity." Am J Dis Child 1988;142:1321-44.