Bilirubin and Brain Damage: What Do We Do Now?

(1)

1062

PEDIATRICS

Vol. 83 No. 6 June

1989

Bilirubin

and Brain

Damage:

What

Do We Do Now?

In many ways, the study reported by van de Bor

and her colleagues’ has a message akin to the classic

telegram: “Bad news. Start worrying. Details to

follow.” Perhaps we should start worrying about

low levels of bilirubin in very low birth weight

infants, but it is premature to do anything else,

other than wait for the critical details that might

dictate a change in current practice.

Compared with similar previous investigations,

this study has several strengths. First, the sample

of 831 surviving infants includes many more very

low birth weight infants than previous studies.25

Second, follow-up evaluations were performed on

100% of the surviving children-a truly remarkable

achievement. Third, with the exception of the

Na-tional Institute of Child Health and Human

Devel-opment (NICHD) randomized phototherapy trial,6

in which enrollment was completed in 1976, it is

the only evaluation of a large cohort of infants in

the modern era of neonatal intensive care. This is

important because such care includes vigorous

treatment of hyperbilirubinemia with phototherapy

as well as exchange transfusion. Finally, the

con-sideration of possible confounding variables

(in-cluding intracranial hemorrhage) is more complete

and more sophisticated than that found in previous

investigations.

There are some weaknesses, of course. These

infants were tested at a corrected age of 2 years and

many “minor” abnormalities found may be less

obvious or absent by 4 to 7 years of age.4 The

measurement of intracranial hemorrhage was crude

and bilirubin determinations were performed only

when there was clinical indication. Thus, there is

concern that the measured maximal bilirubin

con-centration was not, in fact, the maximal bilirubin

concentration in certain infants. If so, maximal

bilirubin concentration is less likely to be a

predic-tor of long-term outcome and, as the authors point

out, this would only underestimate the full effect of

bilirubin.

The Dutch investigators found a dose-response

relationship between peak total serum bilirubin

levels and the risk of neurologic handicap at 2 years:

the odds of handicap increased by about 30% for

each 50-mol/L (2.9-mg/dL) increase in maximal

bilirubin concentration.

Note that the odds of disease and the probability

(or risk) of disease are related but not the same. If

the probability of disease is P, the odds are P/(1

-P). When P is small, the probability and odds are

almost the same. For example, if P = .01, odds are

0.01/0.99 = 0.0101; if P = .1, odds are 0.1/0.9 =

0.11. However, as P increases, odds get higher

rap-idly. For example, if P = .25, odds =

0.33,

if P = .5,

odds = 1, and if P = .99, odds = 99! The logistic

model assumes that increases in the level of each

risk factor multiply the odds of disease by a fixed

amount. If bilirubin and the other risk factors are

related to the odds of handicap in that simple way,

then the odds ratio of 1.3 per 50 mol/L of bilirubin is the estimate that best fits the data. For every

50-mol/L increase in maximal bilirubin

concentra-tion, the predicted odds of handicap are multiplied

by 1.3. Thus, a baby with a maximal bilirubin

concentration of 250 mol/L would be predicted to

have odds of handicap (1.3) =

2.86

times higher

than a baby with a maximal bilirubin concentration

of 50 zmol/L. (This is equivalent to an odds ratio

of 1.57 per 5-mg/dL increase in maximal bilirubin

concentration.)

Given this observed association, we must ask: (1)

is the association real? (2) If so, is it causal?, and

(3) If

causal, what should we do about it?

Is the association real? Spurious associations

may be observed either because of random error

(chance) or systematic error (bias).7 The best way

to assess random error is to calculate a confidence

interval around the parameter used to measure the

strength of the association. In general, as sample

size increases, random error becomes smaller and

the confidence interval narrower. In this study,

however, the 95% confidence interval for the

esti-mated odds ratio of 1.3 ranged from 1.03 to 1.62.

Thus, the actual effect of an elevated bilirubin level

may be trivial (or even nonexistent), or it may be

considerably greater than observed. In spite of the

large sample size, the estimate of the odds ratio is

imprecise.

Systematic error (bias) is a more vexing cause of

spurious associations, because it persists in spite of

increases in the sample size. Bias is much more

likely to cause a spurious association when it is

differential (when the systematic error occurs

dif-ferentially according to whether the risk factor is

present). In the study by van de Bor et at, we must

be sure that the (somewhat subjective)

measure-ment of neurodevelopmental outcome was not

in-fluenced by past bilirubin levels. The best way to

prevent this would have been to ensure that those

responsible for evaluating neurodevelopmental

out-come were unaware of (“blinded” to) the degree of

previous bilirubin exposure.

Is the association causal? Given that the

associ-ation does not seem to be (entirely) due to chance

or bias, the next question is whether it represents

at Viet Nam:AAP Sponsored on September 7, 2020

www.aappublications.org/news

(2)

COMMENTARIES 1063

a cause-effect relationship.7 The two rival

expla-nations are effect-cause and effect-effect. The

for-mer is primarily a consideration in cross-sectional

studies and is ruled out in the present study by the

temporal relationship between the variables. (It is

not possible for neurologic handicap to cause

ele-vation of bilirubin levels obtained 2 years

previ-ously.) The latter explanation, also called

“con-founding,” is much more difficult to rule out. The

question is whether bilirubin and neurologic

hand-icap are associated only because both are related to

some third factor, such as intracranial hemorrhage.

We know that severe intracranial hemorrhage is

associated with neurologic handicap; if intracranial

hemorrhage also increases maximal bilirubin

con-centration, then maximal bilirubin concentration

would be associated with handicap simply because

babies with higher maximal bilirubin concentration

would be more likely to have had an intracranial

hemorrhage.

How do we go about determining whether

con-founding is a likely basis for an association? To be

a confounder, a variable must be associated with

both the risk factor (maximal bilirubin

concentra-tion) and the outcome (neurologic handicap). Thus,

a good approach to evaluating the likelihood of

confounding is to consider which determinants of

handicap might also be associated with maximal

bilirubin concentration. Van de Bor et al considered the major known predictors of neurodevelopmental

outcome: gestational age, birth weight, seizures,

intracranial hemorrhage, respiratory distress

syn-drome, ventriculomegaly, and bronchopulmonary

dysplasia. Gestational age, birth weight, and

respi-ratory distress syndrome turned out not to be

as-sociated with handicap in the present study. Of the

remainder, based on our understanding of bilirubin

metabolism, only intracranial hemorrhage might

have been expected to be associated with maximal

bilirubin concentration, and, when examined, that

association has been either weak8 or absent.9 Thus,

even before the multivariate analysis, the suspicion

that confounding by any of the variables measured

could cause the observed association is fairly low.

Confounding by unknown or unmeasured variables

can never be ruled out in an observational study,

but such unknown variables would have to be

strongly (and independently) related to both

out-come and maximal bilirubin concentration,1#{176} and

this does not seem likely.

The logistic regression analysis performed by van

de Bor et al further strengthens the case for

caus-ality. As mentioned before, logistic regression uses

a model in which the different risk factors are

assumed independently to increase the odds of the

disease. This allows the independent effect of the

risk factor (maximal bilirubin concentration) to be

estimated with potential confounding factors (like

intracranial hemorrhage) held constant

statisti-cally. An important limitation of this approach is

that, to control for confounding statistically, the

measurement of the potential confounder must be

valid. In the van de Bor et a! study, the

measure-ment of intracranial hemorrhage was somewhat

crude: intracranial hemorrhage was treated as

pre-sent or absent, and ultrasound evaluations were not

performed in all infants. The investigators assumed

that significant intracranial hemorrhage would

have been apparent and would have led to an

ultra-sound evaluation, but such hemorrhages may not

be clinically obvious.’1 Nevertheless, the fact that

the association is of similar magnitude with and

without the potential confounders included in the

logistic model (our own analysis of table 1 gives an

odds ratio of 1.28 per 5O-mol/L maximal bilirubin

concentration)’2 suggests that significant

con-founding is unlikely.

In addition to seeking evidence to suggest that

an association is not causal, as discussed before, it

is important to consider positive evidence for a

causal relationship. Four important characteristics

to consider are consistency of the association, its

strength, the presence of a dose-response relation-ship, and its biologic plausibility. We will discuss, rather briefly, how these issues relate to the effects

of bilirubin on developmental outcome. (They have

been thoroughly reviewed elsewhere.13”4)

Except for babies with erythroblastosis and

ker-nicterus, the association between serum bilirubin

and neurologic handicap has not been consistent.

Some investigators have found an association in

low birth weight infants,2’3”5”6 but others have not,4

particularly when bilirubin levels did not exceed 20

mg/dL (342 mol/L).4”2#{176} In two studies, signifi-cantly elevated bilirubin levels appeared to improve

developmental outcome in very low birth weight2

and term infants.2’

The major neurologic deficit was cerebral

palsy-not associated classically with bilirubin

encepha-lopathy in the full-term infant. Nevertheless, this

and other nonspecific neurodevelopmental

abnor-malities have been the predominant findings in

previous studies identifying an association between

bilirubin levels and handicap in preterm and low

birth weight infants.2’3

The strength of an observed association is also

important; stronger relationships are more likely to

be causal. But estimates of the strength of the

association between bilirubin levels and brain

dam-age have varied in direction as well as magnitude,

and the estimate in the present study was

insuffi-ciently precise to provide evidence of a strong

re-lationship.

Van de Bor and co-workers do provide evidence,

(3)

1064

PEDIATRICS

Vol. 83 No. 6 June

1989

however, of a dose-response relationship, and the

association is biologically plausible. Bilirubin is

neurotoxic in a variety of animal models,2225 and

recent studies have demonstrated transient effects

of bilirubin on brainstem-evoked potentials,2m

in-fant behavior, and the characteristics ofthe baby’s

cry.3#{176}

The evidence presented by van de Bor et al is not

overwhelming, but it does suggest a causal

relation-ship between maximal bilirubin concentration and

neurodevelopmental outcome, although the

mag-nitude of the effect may be small.

What do we do now? In brief, nothing much.

First, as noted before, although the association

between maximal bilirubin concentration and

neu-rologic handicap was statistically significant, the

95% confidence interval was wide, the association

(in other studies) has not been consistent, and panic

about bilirubin in the 100- to 2O0-mol/L (6 to 12

mg/dL) range is certainly premature.

Second, even if the observed effect size were true,

we lack a crucial piece of information: would an

intervention that reduced maximal bilirubin

con-centration also reduce the incidence of neurologic

handicap? Randomized clinical trials are necessary

before we can make therapeutic recommendations,

and one such trial has been completed. We have

preliminary data from the NICHD collaborative

phototherapy trial in which approximately 500 of

the infants 2,000 g were followed up for 6 years.6

Phototherapy reduced peak bilirubin levels

signifi-cantly (by about 70 mol/L [4 mg/dL]), but this

did not reduce the incidence of motor deficits or

cerebral palsy, and there was no effect on IQ scores.6

The implications of the study by van de Bor et

al for larger or more mature babies are worth

men-tioning. Even if the relative risk of neurologic

hand-icap for term babies with hyperbilirubinemia were

the same as that found by van de Bor et al for

preterm babies, the absolute risk would be much

smaller. Thus, instead of changing the odds of

handicap from 10% to 13%, a 50-imol/L change in

maximal bilirubin concentration might alter the

odds from 1% to 1.3%. If this were the case, and if

treatment could decrease the maximal bilirubin

concentration by 50 imol/L, about 33 premature

infants would need to be treated to prevent one

case of neurologic handicap. To achieve similar

benefit in term infants, we would have to treat 333

babies. Clearly, the benefits of interventions to

reduce bilirubin levels are much less likely to exceed the risks in larger or more mature infants.

What do we do now? The first step is to confirm

these results with additional observational studies.

We could begin by further analyzing the NICHD

trial results and by assembling data from infants in

intensive care nursery follow-up programs in the

United States. Many neonatal groups have been

systematically following up their surviving infants

with birth weights of <1,500 g. Maximal bilirubin

values are available, as are data concerning

intra-cranial hemorrhage and developmental outcome. If

the data of van de Bor et al are correct, then there

appears to be justification for another randomized

trial of the effect on neurologic outcome of

decreas-ing serum bilirubin concentrations (perhaps by

ad-ministering tin protoporphyrin in the delivery

room) in a group of premature low birth weight

infants.

Meanwhile, start worrying. Details to follow.

REFERENCES

THOMAS

B.

NEWMAN, MD, MPH Department of Pediatrics

University of California San Francisco

M. JEFFREY MAISELS, MB, BCH Department of Pediatrics

William Beaumont Hospital

Royal Oak, MI

1. van de Bor M, van Zeben to Van der Aa TM, Verloove-Vanhorick SP et a!: Hyperbilirubinemia in very preterm infants and neurodevelopmental outcome at 2 years of age:

Results of a national collaborative survey. Pediatrics

1989;83:915-920

2. Scheidt PC, Mellit ED, Hardy JB: Toxicity to biirubin in neonates: Infant development during first year in relation to maximum neonatal serum bilirubin concentration. J Pe-diatr 1977;91:292-297

3. Naeye RL: Amniotic fluid infections, neonatal hyperbilirubi-nemia, and psychomotor impairment. Pediatrics 1978;62:

497-503

4. Rubin RA, Bellow B, Fisch RO: Neonatal serum bilirubin levels related to cognitive development at ages four through seven years. J Pediatr 1979;4:601-604

5. Johnson L, Boggs TR: Bilirubin-dependent brain damage: Incidence and indications for treatment, in Odell GB, Schaf-fer R, Siinopoulous AB (eds): Phototherapy in the Newborn: An Overview. Washington, DC, National Academy of

Sci-ences,1974, pp 122-149

6. Scheidt PC, Bryla DA, Nelson HB: NICHD phototherapy clinical trial: Six year follow-up results, abstracted. Pediatr Res 1988;23:455A

7. Newman TB, Browner WS, Hulley SB: Enhancing causal inference in observational studies, in Huiley SB, Cummings SR (eda): Designing Clinical Research. Baltimore, Williams &Wilkins, 1988, pp 98-109

8. Epstein MF, Leviton A, Kuban KCK, et al: Bilirubin, intra-ventricular hemorrhage, and phenobarbital in very low birth weight babies. Pediatrics 1988;82:350-354

9. Amato M, FauchereJC, von Muralt G: Relationship between periintraventricular hemorrhage and neonatal hyperbiliru-binemia in very low birth weight infants. Am J Perinatol

l987;4:275-278

10. Winkelstein W Jr, Shiffitoe EJ, Brand R, et a!: Further comments on cancer of the uterine cervix, smoking, and herpes virus infection. Am J Epidemiol 1984;119:1-7 11. Lazzara A, Ahmann P, Dykes F, et al: Clinical predictability

of intraventricular hemorrhage in preterm infants. Pediat-rics 1980;65:30-34

12. McGee DL: A program for logistic regression on the IBM

(4)

PEDIATRICS Vol. 83 No. 6 June 1989 1065

PC. Am J Epidemiol 1986;124:702-705

13. Maisels MJ: Neonatal jaundice, in Avery GB (ed): Neona-tology, Pathophysiology and Management of the Newborn, ed

3.Philadelphia, JB Lippincott, 1987, pp 583-585 14. Maisels MJ: Clinical studies of the sequelae of

hyperbiliru-binemia, in Levine RL, Maisels MJ (eds): Hyperbilirubine-mia in the Newborn, Report of the 85th Ross Conference on Pediatric Research. Columbus, OH, ROSS Laboratories, 1983,

pp 26-38

15. Crichton JU, Dunn HG, McBurney AK, et al: Long term

effects of neonatal jaundice on brain function in children of low birth weight. Pediatrics 1972;48:656-670

16. Koch CA, Jones DV, Dine MS. et a!: Hyperbilirubinemia in preterm infants: A follow-up study. J Pediatr 1959;55:23-29

17. Hugh-Jones K, Slack J, Simpson K, et al: Clinical course of hyperbilirubinemia in premature infants. N Engi J Med 1960;263:1223-1229

18. Odell GB, Storey GNB, Rosenberg LA: Studies in kernic-terus: III. The saturation of serum proteins with bilirubin during neonatal life and its relationship to brain damage at

5years.J Pediatr 1970;76:12-21

19. Shiller JG, Silverman WA: Uncomplicated hyperbilirubi-nemia of prematurity. Am J Dis Child 1961;101:587-592 20. Wishingrad L, Cornblath M, Takakuwa P, et al: Studies of

nonhemolytic hyperbilirubinemia in premature infante: Pro-spective randomized selection for exchange transfusion with observations on the levels of serum bilirubin with or without exchange transfusion and neurologic evaluations one year after birth. Pediatrics 1965;36:162

21. Holmes GE, Miller JB, Smith EE: Neonatal bilirubinemia in production of long term neurological deficits. Am J Dis Child 1968;116:37-43

22. Chen H-C, Lien IN, Lu T-C: Kernicterus in newborn rab-bits. Am J Pathol 1965;46:331

23. Jirka JH, Duckrow B, Kendig JW, et al: Effect of bilirubin on brain stem auditory evoked potentials in the asphyxiated rat. Pediatr Res 1985;19:556-560

24. Lucey JF, Hibbard E, Behrman RE, et a!: Kernicterus in asphyxiated newborn rhesus monkeys. Exp Neurol

1964;9:43-58

25. Wennberg RP, Hance AJ: Experimental bilirubin encepha-lopathy: Importance of total bilirubin, protein binding, and blood-brain barrier. Pediatr Res 1986;20:789-792

26. Perlman M, Fainmesser P, Sohmer H, et al: Auditory nerve-brainstem evoked responses in hyperbilirubinemic neonates. Pediatrics 1983;72:658-664

27. Nakamura H, Takada S, Shimabuku R, et a!: Auditory nerve and brainstem responses in newborn infants with hyperbil-irubinemia. Pediatrics 1985;75:703-708

28. Wennberg RP, Ahifors CE, Bickers R, et al: Abnormal auditory brain stem response in a newborn infant with hyperbilirubinemia: Improvement with exchange transfu-sion. J Pediatr 1982;100:624-626

29. Escher-Graub DC, Fricker HS: Jaundice and behavioral organization in the full-term neonate. Helu Paediatr Acta 1986;41:425-435

30. Golub HL, Corwin MJ: Infant cry: A clue to diagnosis. Pediatrics 1982;69:197-201

It’s Time

for Pediatricians

to

‘Rally

‘Round

the Pool Fence’

Pediatrics has distinguished itself as one of the

medical disciplines most concerned with preventive

medicine and strategies to prevent accidents and

illness. Immunizations and accident prevention

ed-ucation are a routine part of pediatric practice and

well-child care.

Poison prevention is one of the significant

ac-complishments of the 20th century in which

pedia-tricians played a major role and can feel justly

proud. Pediatricians played a decisive role in the

establishment of the Poison Control Center

net-work and in the development of child-proof safety

caps that ultimately led to the Poison Prevention

Packaging Act of 1970.’ The results have been

gratifying, with the death rate from poisonings in

children declining from 2.2 per 100,000 children in

1960 to 0.5 per 100,000 children in 1980.2 An

esti-mated #{189}to 1 million accidental ingestions have

been prevented by child-resistant closures.3

Poison-ings have been reduced from the second leading

cause of accidental death in children to fourth place

(Table). Deaths from unintentional ingestion of

potentially poisonous substances among children

less than 5 years of age decreased from a peak of

456 in 1959 to 57 in 1981.

Pediatricians and legislators now have the

op-portunity to make another significant impact on

accident prevention. The second leading cause of

accidental death in children and young adults has

become drowning, and swimming pools are a major

risk site. Various studies have shown that 40% to

80% of drowning and near-drowning accidents

oc-cur in swimming pools.’#{176} Using the new

epidemi-ologic designation, drowning is the second leading

cause of years of potential life lost from premature

mortality due to unintentional injury” (Table). Not

only is drowning a major cause of pediatric

mortal-ity, but near-drowning accidents are an important

cause of anoxic encephalopathy with permanent

neurologic dysfunction and morbidity. It is

esti-mated that one third of all survivors are moderately to severely neurologically damaged.’2

TABLE. Leading Causes of Prematu Unintentional Injury

re Mortality Due to

Accident Potential

Life Lost (yr)

Motor vehicle

Traffic accidents 1.4 x 106

Nontraffic accidents 3.4 x i0

Drownings 2 x 10

Fire and flames 1.4 x 10

Poisonings 1.3 x iO

Falls 8 x i0

Firearms 5.5 x i0

Choking on food or object 4.5 x i0

Transport

Water 3.5 x i0

Air 3.4 x iO

(5)

1989;83;1062

Pediatrics

THOMAS B. NEWMAN and M. JEFFREY MAISELS