INTRODUCTION
About 55,000 children per year are born in the United States at birth
weights <1,500 g (3 lb, 5 oz), defined as very low birth weight
(VLBW: Martin et al., 2002; Volpe, 2001). These children represent slightly
more than 1% of annual live births. Advances in obstetric and neonatal
care have yielded dramatic increases in the survival of these children
over the past few decades, especially among the least mature neonates
(Fanaroff et al., 2003; Hack et al., 1996a). The more extreme low
birthweight subset of the VLBW population is identified on the basis of
birth weight <1,000 (2 lb, 3 oz), <800 g (1 lb, 12 oz), or
<750 g (1 lb 10 oz). Shortened gestational periods provide an
alternative means for describing preterm infants. Gestational age,
however, is less reliably measured than birth weight. For this reason,
birth weight is often used as the major marker of prematurity (Cole et al., 2002).
Sequelae of VLBW include neurosensory deficits and other health
problems (cerebral palsy, poor vision, hearing loss, seizures,
pulmonary disease, growth impairments), cognitive weaknesses, poor
academic achievement, and attention deficits and other behavior
disorders (Taylor et al., 2000a). Outcomes
are highly variable, ranging from severe motor and mental handicaps to
more subtle developmental and cognitive disorders, such as learning
disabilities and attention deficits (Whitfield et
al., 1997). Some children with VLBW have no detectable
impairments (Taylor et al., 2000b).
This variability is explained in part by the extent of prematurity
and attendant medical complications. The children at greatest risk for
sequelae are those with extremely low birth weight and/or neonatal
complications. These complications include evidence for
intraventricular hemorrhage (IVH) or periventricular leukomalacia (PVL)
on neonatal cranial ultrasounds, birth hypoxia and chronic lung
disease, septicemia, raised bilirubin levels, hypothyroxemia, and
composite indices of neonatal medical complications (Brazy et al., 1993; Frisk &
Whyte, 1994; Hopkins-Golightly et al.,
2003; Horwood et al., 1998; Landry et al., 1993; Marlow et
al., 1989; McGrath & Sullivan,
2002; Perlman, 2001; Taylor et al., 1998). Children with VLBW who are
small for gestational age due to intrauterine growth failure are also
considered at higher risk (Hutton et al.,
1997; Pinto-Martin et al., 1999),
though it is unclear if low weight for gestational age confers
additional risk over and above the degree of low birth weight and
neonatal medical complications (Hack, 1997).
Factors that predict outcomes in term-born children of normal birth
weight, such as sociodemographic status and the family environment, are
also related to these outcomes in children born prematurely (Taylor et al., 2000a). Social–environmental
factors, moreover, may moderate the effects of VLBW (Taylor et al.,
1998, 2004).
Hypoxic ischemia, leading to germinal matrix IVH, associated
hemorrhagic infarction, and PVL, is a primary cause of brain damage in
preterm children (Perlman, 1998; Volpe, 1998, 2001) and thus the
most likely reason for the adverse developmental consequences of VLBW.
Infection is also considered a source of damage to the periventricular
region (Dammann & Leviton, 1999). More
severe grades of IVH and PVL have been documented in only a minority of
children with VLBW; and the likelihood of these insults increase with
the degree of prematurity (Inder & Volpe,
2000; Perlman, 1998). In view of
recent magnetic resonance imaging (MRI) findings of diffuse white
matter abnormality in 33 of 50 children with a mean birth weight of
1,086 g (Counsell et al., 2003), white matter
pathology is likely to be substantially more common than suggested by
cerebral ultrasound findings (Volpe,
2003).
Cerebral insults are most evident in sensorimotor,
parietal–occipital, and temporal regions adjacent to the
ventricles, the hippocampus and other subcortical structures, and
circuits connecting these structures to the frontal lobes (Frisk & Whyte, 1994; Goto et
al., 1994; Holling, 1999; Lou, 1996; Perlman, 2001;
Peterson et al., 2000). Evidence of damage
includes periventricular white matter abnormalities, reduction in white
and gray matter volumes, and ventricular enlargement. Cortical
development is also compromised (Inder et al.,
1999). The nature of arterial border zones in the premature
brain, together with the special vulnerability of white matter to
proinflammatory cytokines and free radical attack at this stage in
neural development, places periventricular and subcortical regions at
greatest risk for injury (Perlman, 2001;
Volpe, 2001). Hypoxic ischemia in term
infants is more likely to result in multifocal cortical necrosis,
especially to parasagittal areas and the depths of the sulci (Volpe, 2001).
If frontostriatal or temporal lobe/hippocampal pathology has
similar effects in children with VLBW as in persons with later
occurring lesions, the survivors of VLBW may be especially prone to
deficits in visual–motor skills, memory, and executive function,
at least relative to deficits in verbal–semantic skills (Aram & Eisele, 1992; Broadbent et al., 2002; D'Esposito & Postle, 2002; Knowlton, 2002; Salmon et al.,
2001). Demonstration of specific deficits in the former ability
domains in children with VLBW would help establish early
structure–function relations and would support developmental
congruence in these relations (Dennis et al., 1999).
Evidence for specificity in impairment would also indicate constraints on
neural plasticity (Taylor & Alden, 1997).
Major methodological difficulties faced in examining these
relationships in the VLBW population are at least twofold. First,
clinical management involves only gross assessment of neuropathology
during the neonatal period, including neonatal cerebral ultrasounds and
estimates of the likelihood of brain insult as indexed by other
neonatal complications or the extent of prematurity (i.e., birth
weight, gestational age, or birth weight relative to gestational age).
MRI scans are rarely carried out for clinical management of preterm
infants, and cranial ultrasounds are insensitive to more subtle white
matter abnormalities (Inder et al., 2003;
Maalouf et al., 1999). Second, the neuropathology
associated with VLBW varies widely across individuals, with detectable brain
damage being absent, relatively localized, or diffuse. This variation,
together with the global effects on cognition that would be expected in
cases of diffuse insult, may obscure specific neuropsychological consequences
of localized periventricular lesions.
Nonetheless, deficits in the above-mentioned skills among cohorts of
survivors are generally out of proportion to global impairments or to
deficits in other cognitive domains. Deficits in motor and
visual-perceptual skills, memory, and executive function are common
sequelae of VLBW. These sequeale are present even in children without
neurosensory disorders or mental retardation, or when controlling for
IQ (Goyen et al., 1998; Hack et al., 1992; Holsti et
al., 2002; Jacobson et al., 2001;
Luoma et al., 1998; Roth
et al., 1993; Selzer et al., 1992;
Taylor et al., 2000c; Waber & McCormick, 1995; Whitfield et al., 1997). Numerous findings suggest
more marked effects of VLBW on perceptual or performance skills than on
language functions (Fletcher et al., 1997;
Frisk & Whyte, 1994; Gabrielson et al., 2002; Luoma
et al., 1998; Taylor et al., 2000b).
Deficits on delayed memory tasks have been reported in preterm infants
and preschoolers with VLBW (Espy et al.,
2002; Rose, 1983; Ross et al., 1992, 1996). Specific
impairments in working memory and spatial planning, relative to IQ or
receptive vocabulary scores, have also been documented in young
school-age children with <1,000 g birth weight (Frisk & Whyte, 1994; Harvey
et al., 1999). Frisk and Whyte (1994)
found that these deficits were related to presence of IVH or PVL as
identified on neonatal ultrasound. Deficits in older school-age
children with VLBW have been found on tasks assessing visual working
memory and speed of visual processing, visual recognition memory, and
spatial planning (Luciana et al., 1999; Rickards et al., 2001; Rose
& Feldman, 1996).
Results from our own longitudinal study of children with VLBW offer
further support for specific cognitive deficits (Taylor et al., 2004). Even when controlling for IQ,
the children in this study with <750 g birth weight scored more
poorly than term controls on tests of memory, visual–spatial
skill, and executive function. Follow-up of the children between mean
ages 7–14 years revealed increasing deficits over time in the
<750 g group relative to controls on some tests of
visual–motor skill and executive function, suggesting that these
impairments may become even more pronounced with advancing age.
Cognitive outcomes in a lower-risk 750–1,499 g group fell
consistently between those of the <750 g group and term control
children, though few deficits and no specific impairments could be
detected in the heavier VLBW group.
The present study reports on the cognitive outcomes in our sample at
a recently completed follow-up conducted at mean age 16 years. The
rationale for undertaking this assessment was to assess executive
function and memory more comprehensively than we had previously, to
determine if the specific cognitive impairments we had observed in
children with VLBW at younger ages would continue to be evident in
later adolescence and early adulthood, and to examine neonatal risks
for long-term neuropsychological deficits. Studies of outcomes in
adults born preterm or with VLBW suggest persistent effects of
prematurity on cognition, educational attainment, and behavior (Hack et al., 2002; Pharoah et
al., 2003; Tideman, 2000). However,
cognitive assessments at later ages have been largely limited to IQ
testing. We do not know if individuals with VLBW continue to have
specific weaknesses, or if these weaknesses resolve or lead to less
differentiated global cognitive deficiency. Comprehensive testing also
permitted a more detailed evaluation of the nature and specificity of
cognitive impairments than was possible in our earlier follow-ups or in
other studies.
Based on school-age outcomes of VLBW and on the widespread
neuropathology likely present in some survivors, our first hypothesis
was that children with VLBW would have poorer cognitive outcomes in
general than term-born controls. In view of evidence for a gradient
effect, with outcomes varying in relation to the degree of low birth
weight and/or prematurity (Breslau et al.,
1996; Horwood et al., 1998; Klebanov et al., 1994), we anticipated that
cognitive sequelae of VLBW would be more evident in the <750 g group
than in the 750–1,499 g group.
Given the previous support for specific neuropsychological
consequences of VLBW, our second hypothesis was that cognitive
impairments in our VLBW sample would be more evident on measures of
visual–motor skills, memory, and executive function than on
language testing. This hypothesis also examined the possibility that
risks for early hypoxic ischemia and other insults to periventricular
structures would result in long-term deficits similar to those found in
later-onset lesions to this brain region. Two sets of secondary
analyses were conducted to further explore the specificity of cognitive
deficits. In one set of analyses, we repeated our primary group
comparisons after excluding children with neurosensory disorders or low
IQ; and in the second set, we repeated group comparisons for the total
sample, controlling for verbal-semantic ability as measured by a
vocabulary measure. The purpose of the first set of analyses was to
determine if sequelae of VLBW could be detected in children without
severe disabilities, and the aim of the second set was to identify
areas of relative cognitive impairment.
Our final hypothesis was that cognitive outcomes for children in the
two VLBW groups would be worse for individuals at higher risk for brain
insult. Lacking more precise measures of neonatal brain insult, these
risks were defined in terms of ultrasound findings of neonatal
periventricular pathology, lower birth weight, lower weight for
gestational age as an index of intrauterine growth retardation, and
severity of chronic lung disease as measured by the duration of
neonatal oxygen requirement. These factors increase the likelihood of
hypoxic ischemia and associated early brain insults in the
periventricular region and have been associated with poorer cognitive
outcomes in children with VLBW (Farel et al.,
1998; Hutton et al., 1997; Luciana et al., 1999; Perlman,
2001; Pinto-Martin et al., 1999; Rose & Feldman, 1996; Volpe,
2001).
METHODS
Sample Recruitment and Long-Term Follow-Up
The sample for this study included children and caregivers who were
originally recruited for a longitudinal study of early school-age
outcomes of extremely low birth weight (Hack et al.,
1994) and who remained in the study at the long-term follow-up.
The initial sample comprised the 65 children with <750 g birth
weight, 65 children with 750–1,499 g birth weight, and 61
term-born controls. Children in the <750 g group were 93% of the
survivors born at neonatal intensive care units in Region V of Ohio
from July, 1982 through December, 1986. Children in the 750–1,499
g group were matched to those in the <750 g group based on hospital
of birth, age, sex, and race. Term-born children were matched to those
in the <750 g group based on school attended, age, sex, and race.
Because of difficulties in recruitment, we were unable to find matches
for all children in the <750 g group. However, the three groups were
similar in background characteristics at the initial assessment and
across subsequent follow-up during the middle school-age years (Hack et al., 1994; Taylor et al., 2000b, 2004).
In recruiting children for the most recent, or long-term, follow-up
at mean age 16 years, we attempted to contact all families previously
enrolled (Taylor et al., 2004). Of the 201
previous participants, 147 (73%) took part in this follow-up, including
48 in the <750 g group, 47 in the 750–1,499 g group, and 52
term-born controls. The long-term follow-up was conducted at a mean age
of 16.8 years (SD = 1.2, range = 14.6–21.2). The mean
interval between this assessment and initial recruitment was 10.0 years
(SD = 0.6, range = 8.7–11.9). Mean grade level for the
129 participants still in high school was 10.5 (SD = 0.9,
range = 9–12). The groups were similar in these characteristics,
with no significant differences.
Reasons for drop-out included lack of ability to be tested
(n = 3), moves out of the area (n = 11), inability to
locate families (n = 27), and disinterest or lack of
follow-through by families (n = 13). To assess attrition bias,
we compared participants who dropped out with those remaining in
follow-up in background characteristics. Results failed to reveal
differences in the distributions of children by group, sex, or race.
However, children who dropped out had lower socioeconomic status (SES)
as measured by the Hollingshead Four Factor Index (Hollingshead, 1975) and lower initial cognitive
ability as measured by a short-form of the Kaufman Assessment Battery
for Children (Kaufman & Applegate, 1988)
than did children who completed the study. Comparison of participants
with VLBW who dropped out with those remaining in follow-up failed to
reveal differences in neonatal risk factors.
Table 1 summarizes sample characteristics
at the long-term follow-up for the three birthweight groups.
Differences in birth weight, gestational age, and other neonatal
factors reflected recruitment criteria. The groups were similar in sex
and race distribution, as well as in SES and in age at follow-up. Eight
participants had neurosensory impairments, including 5 in the <750 g
group (1 with cerebral palsy, 2 with hearing impairment, and 2 with
visual impairment), and 3 in the 750–1,499 g group (1 with
cerebral palsy, 1 with hearing impairment, and 1 with both cerebral
palsy and hearing impairment). None of the controls had these
disabilities.
Sample characteristics at long-term follow-up
Procedures and Measures
The participants were assessed in a single half-day session. Although
only results of cognitive testing are considered in this study,
follow-up also included tests of academic achievement, parent
interviews to obtain information about the family environment and
participants' adaptive behavior skills, and parent and teacher
ratings of behavior and school performance (Taylor
et al., 2004). Test order was counterbalanced across
participants and breaks were provided to reduce fatigue. Institutional
review board approval and informed parental consent and child assent
were obtained prior to participation.
Table 2 lists the tests comprising the
neuropsychological battery by skill domain, together with brief
descriptions of the procedures and dependent variables. Global
cognitive function was assessed using a short form of the Wechsler
Intelligence Scale for Children–Third Edition (WISC–III;
Wechsler, 1991) or of the Wechsler Adult
Intelligence Scale–Third Edition (WAIS–III; Wechsler, 1997). Prorated IQ was based on the
Vocabulary and Block Design subtests (Sattler,
1992). The 53 participants aged 17 years and older who received
the WAIS–III subtests were similarly distributed across the three
groups. Tests of language, in addition to the Wechsler Vocabulary
subtest, included measures of naming, word knowledge, pragmatic
language, and fluency in generating letter–word associations.
Visual–motor skills were surveyed using tests of fine motor
dexterity, design copying, and visual matching/search, and spatial
judgment. The California Verbal Learning Test, Second Edition
(CVLT–II; Delis et al., 2000) provided
a comprehensive assessment of verbal learning and memory. Nonverbal
memory was measured using the spatial memory tasks from the Cambridge
Neuropsychological Test Automated Battery (CANTAB; CeNes
Cognition, 1996). To evaluate executive function, we administered
other CANTAB tasks and the Contingency Naming Test (Anderson et al., 2000).
Neuropsychological test battery administered at long-term
follow-up
The CANTAB was selected because of its sensitivity to frontostriatal
and medial temporal lesions in adults and to the effects of preterm
birth in children (Luciana, 2003; Luciana et al., 1999; Luciana
& Nelson, 1998). CANTAB tasks are appropriate for persons of
diverse cognitive abilities and assess several dimensions of executive
function, including self-guided searching, set switching, spatial
planning, and spatial working memory. The CANTAB was administered on a
portable PC Pentium 233 MMX 64/4300 12.1-inch TFT color LCD
touch-screen using Windows 98 software.
While we recognized that several of the measures could be grouped
under multiple domains, classification of tests into domains was based
on the primary demands of each test. For example, although some demands
on memory and executive function are imposed by the language tests,
they draw more heavily on verbal knowledge than do other tests in the
battery. Classification of recall of the Rey-Osterrieth Complex Figure
(ROCF; Bernstein & Waber, 1996) as a
visual–motor rather than memory test was justified by evidence
for shared loadings of ROCF copy and recall on a
perceptual–planning factor (Taylor et al.,
2002). Memory for spatial sequences (CANTAB Spatial Span) could
be considered a measure of attention span or spatial working memory,
and thus of executive function. Classification of this task as a memory
measure was justified by its demands on nonverbal short-term memory
(Luciana et al., 2001). Our decision to
analyze individual test scores, rather than a smaller set of ability
composites, reflected the lack of an empirical basis for combining
scores and an interest in examining test sensitivity to the sequelae of
VLBW.
Age-standardized scores were used for tests with norms that applied
to all participants. In most instances, however, tests norms were not
available for older participants or were of questionable applicability;
hence raw scores were used for these measures. To reduce redundancy of
measurement, most tests were represented by single scores. The only
multiple-measure tests were the ROCF and California Verbal Learning
Test, Second Edition (CVLT–II; Delis et al.,
2000). The copy and recall administrations of the ROCF were
examined separately due to the differing demands of these two
procedures. Because structural and incidental scores on the ROCF were
at ceiling levels for most participants, only the organization scores
were analyzed. The multiple response measures obtained from the
CVLT–II were reduced to three measures summarizing initial
learning (Trials 1–5 Total score), recall (average score on the
four recall trials), and recognition (Recognition Discriminability
score).
Neonatal risks for hypoxic ischemia and other periventricular brain
insults were assessed in terms of (1) the presence/absence of IVH,
PVL, or dilated ventricles on neonatal cranial ultrasounds conducted
during the neonatal hospitalization (based on the most severe
abnormality observed), (2) birth weight in kilograms, (3) birth weight
for gestational age as a measure of intrauterine growth failure, and
(4) duration of oxygen requirement in weeks as a measure of chronic
lung disease. Birth weight for gestational age was defined as the
difference between actual weight and expected weight for gestational
age in standard deviation units, using standards provided by Usher and
McLean (1969). Birth weight was of interest
due to the increased risk of brain insult with lower weight. The
rationale for examining weight for gestational age is that this factor
may also be associated with intrauterine hypoxic ischemia (Hutton et al., 1997). The length of oxygen
requirement was used as a marker of chronic lung disease (also referred
to as bronchopulmonary dysplasia, or BPD), a complication of VLBW that
increases risks for chronic recurrent postnatal hypoxemia and other
postnatal pulmonary complications (Eichenwald et
al., 1997; Perlman, 2001). Birth
weight was correlated with weight for gestational age (r =
.53, p < .01) and length of oxygen requirement (r
= −.44, p < .01), but weight for gestational age was
not related to length of oxygen requirement (r = −.06).
Because the duration of oxygen requirement was highly associated with
the duration of ventilatory support and days of hospitalization
(rs of .85 and .89, respectively, p < .01), the
latter variables were not considered.
Data Analysis
Group differences in the neuropsychological outcomes listed in Table 2 were examined using analysis of covariance
(ANCOVA), with SES, race, and sex as covariates. Preliminary analysis
indicated that one or more of these factors was associated with most of
the outcomes, but failed to reveal significant interactions of these
factors with group. These factors were thus included as covariates in
all analyses (a table summarizing the associations of these factors
with the outcomes is available from the first author). Age and
age2 were entered as additional covariates to
adjust for the linear and quadratic effects of age on the raw scores.
Differences between each VLBW group (<750 g, 750–1,499 g) and
the term group were analyzed as preplanned contrasts. Given the
substantial number of dependent measures, only effects with p
< .01 were considered significant in these and all other analyses.
Transformations were made to some scores to normalize distributions,
but analyses were conducted on untransformed scores initially to aid in
interpretability.
To gauge the extent and potential clinical significance of group
differences, we also examined rates of deficient test scores. Expected
scores were computed by regressing scores from the term-born
participants on SES, sex, race, and (for raw score measures) age and
age2. Deficits were then defined as scores that
were more than 1 standard error of prediction below the expected scores
(or above them for error measures). Comparisons of the rates, or odds,
of these deficits in each VLBW group relative to term group were
conducted via logistic regression.
To determine if long-term cognitive consequences of VLBW were present
in children without severe disabilities, the above analyses were
repeated after excluding the 26 participants (16 from the <750 g
group, 7 from the 750–1,499 g group, and 3 term controls) with
either a neurosensory disorder or prorated IQ < 70. We also repeated
the above analyses on the total sample, but with the Wechsler
Vocabulary scaled score included as an additional covariate. The latter
analyses were justified by previous findings suggesting relative
sparing of verbal–semantic abilities in children with VLBW (Frisk & Whyte, 1994). The performance component
of IQ, Block Design, was highly correlated with many of the tests of
executive function (e.g., its partial correlation with CANTAB Working
Memory was .68, p < .01). Furthermore, Block Design could
itself be construed as a measure of visual–motor skills or
executive function. For these reasons, areas of relative deficit were
identified by controlling for Vocabulary rather than IQ.
Linear regressions were conducted on scores of participants in the
two VLBW groups to identify the neonatal risk factors that predicted
variations in long-term cognitive outcomes. The covariates in these
analyses were the same ones included in analysis of group differences.
Analyses considered each neonatal risk factor separately. In
preliminary analyses, absolute birth weight predicted many outcomes
independent of group (<750 g vs. 750–1,499 g). This
finding, along with the lack of evidence for differential effects of
birth weight in the two groups, justified inclusion of birth weight as
a continuous predictor of outcome in analysis of data from the
aggregate VLBW sample. Following the primary analyses, risk factors
that were separately related to outcomes were entered into regressions
as sets to identify risks that predicted unique variance in cognitive
performance.
RESULTS
Group Differences in Neuropsychological Performance
(Total Sample)
Table 3 presents group means and standard
deviations for each neuropsychological measure, together with results
from the ANCOVAs and effect sizes for the group differences. The
<750 g group scored more poorly at the long-term follow-up than the
term controls on all measures except for the Boston Naming Test (Spreen & Strauss, 1998), CVLT–II
composite delayed recall, and CANTAB
Intradimensional–Extradimensional (IED) Shift. As evident from
the effect sizes, the largest group differences were on tests of
visual–motor skills, spatial memory, and executive function.
Transformations that were made to correct for non-normal distributions
in three scores (Verbal Cancellation: Mesulam,
1985; CANTAB Pattern Recognition; and CANTAB IED Shift) did not
alter this pattern of findings. Analyses failed to reveal significant
differences between the 750–1,499 g group and term controls,
though nonsignficant trends (p < .1) in favor of the
control group were found for Purdue Pegboard (Gardner, 1979) and Contingency Naming.
Results of group comparisons of neuropsychological outcomes
Secondary analyses provided additional support for impairments in
executive function. ANCOVA conducted on other CVLT–II measures
revealed that the <750 g group had significantly more recall
intrusions, lower Recall Discriminability, and lower Semantic
Clustering than the term group. On CANTAB Stockings of Cambridge, the
<750 g group had significantly shorter “initial planning
time” (suggesting poorer planning or greater impulsivity) and
greater “subsequent planning time” (suggesting slower
subsequent processing) than the term group. On CANTAB Spatial Working
Memory, the <750 g group also had a significantly higher Strategy
score, indicating a less organized search process.
According to results from logistic regression analysis, the group
differences were clinically meaningful. As shown in Table 4, deficits on some measures were 3 or more
times more common in the <750 g group than in the term group.
Consistent with results from ANCOVA, differences in rates of impairment
were found for all tests of visual–motor skills (including Block
Design), as well as for measures of spatial memory and executive
function, but not for the language measures. Relative to controls, the
750–1,499 g group also had higher rates of deficits on CANTAB
Pattern Recognition Memory and Spatial Span. Figure
1 graphs the distribution of Block Design scaled scores to
illustrate the effects of VLBW on score distributions.
Group differences in rates of neuropsychological deficits
Distribution of scaled scores on the Wechsler Block Design subtest
for the three birthweight groups.
Evidence for Specificity of Deficits
Findings from analyses that excluded children with neurosensory
disorders or IQ <70 provided further evidence for specific cognitive
sequelae of <750 g birth weight. Even when excluding children with
severe disabilities, the <750 g group scores significantly more
poorly than term controls on measures of IQ (Wechsler Block Design and
prorated IQ), visual–motor skills [Purdue Pegboard,
Developmental Test of Visual Motor Integration (VMI, Beery, 1989), ROCF copy and recall, Judgment of
Line Orientation (Spreen & Strauss,
1998)], memory (CANTAB Pattern Recognition Memory and
Spatial Span), and executive function (CANTAB Stockings of Cambridge,
Spatial Working Memory, and Rapid Visual Processing, and Contingency
Naming). In contrast, the initial differences in Wechsler Vocabulary
and the other language measures, as well as in verbal learning and
memory, were no longer significant. Analysis that included the total
sample but controlled for Vocabulary revealed a nearly identical
pattern of findings, the only exceptions being that the group
differences in IQ and CANTAB Rapid Visual Processing were no longer
significant. Analysis of transformed scores for the three nonnormal
measures yielded results similar to analysis of the untransformed
scores.
Rates of deficits also continued to be higher in the <750 g group
than in the term group after excluding children with neurosensory
disorder or IQ <70. These differences were significant for Block
Design, Purdue Pegboard, the VMI, ROCF copy, CANTAB Pattern Recognition
Memory, Spatial Span, Stockings of Cambridge, and Spatial Working
Memory, and Contingency Naming. Figure 2
illustrates group differences in rates of deficits for this subset of
the sample.
Illustration of group differences in rates of deficit scores for the
subset of the sample without neurosensory disorder and with Wechsler
prorated IQ ≥ 70. CANTAB = Cambridge Neuropsychological Test
Automated Battery. Deficits were defined as scores > 1 standard
error of prediction below expectations based on the term group's
performance, and considering the effects of socioeconomic status, race,
sex, and age. For all three measures, rates (or odds) were
significantly higher for the <750 g group relative to term group
(for Purdue Pegboard, odds ratio, OR = 7.77, 95% confidence interval,
CI = 2.44–24.68, p < .005; for CANTAB Pattern
Recognition Memory, OR = 9.20, CI = 2.34–36.20, p <
.005; and for CANTAB Stockings of Cambridge, OR = 5.17, CI =
1.79–14.92, p < .005). The findings demonstrate
clinically significant group differences in visual–motor skills,
visual memory, and spatial planning.
Predictors of Outcomes Within the VLBW Groups
Abnormality on cranial ultrasound failed to predict any of the
cognitive measures. More severe indications of abnormality, including
Grade III–IV IVH, PVL, and ventricular dilatation, also failed to
predict any of the outcomes. Results of analyses of the other neonatal
risk factors are summarized in Table 5.
Consistent with the group comparisons, lower birth weight and lower
weight for gestational age were related to many of the outcomes, but
neither factor was related to the language measures. A longer period of
oxygen requirement also predicted poorer outcomes on many of the tests,
although the pattern of associations differed somewhat for this risk
factor. Only oxygen requirement predicted Wechsler Vocabulary and the
Synonyms subtest of the Comprehensive Assessment of Speech and Language
(CASL; Carrow-Woodfolk, 1999), whereas only birth
weight and weight for gestational age predicted ROCF recall and CANTAB
Spatial Span. In analysis of the transformed scores, all three risk factors
listed in Table 5 predicted CANTAB IED Shift, but
results were otherwise similar to those from analysis of the
untransformed measures.
Results from regression analyses relating neuropsychological outcomes
at long-term follow-up to neonatal risk factors
To examine associations of the neonatal factors with outcomes that
were independent of verbal–semantic skill, the above analyses
were repeated with Vocabulary entered into the regressions as an
additional covariate. In these analyses, lower birth weight predicted
poorer performance on Block Design, all tests in the visual–motor
skills domain, CANTAB Spatial Span, Spatial Working Memory, and Rapid
Visual Processing, and Contingency Naming. A longer period of oxygen
requirement predicted lower scores on Purdue Pegboard, Verbal
Cancellation, CANTAB Spatial Working Memory, and Contingency Naming.
In analyses that included all three of the predictors listed in Table 5, oxygen requirement accounted for unique
variance in Wechsler Vocabulary (B =−0.09, SE = 0.03,
t = −2.80, p < .01), IQ (B = −0.53,
SE = 0.19, t = −2.75, p < .01),
Verbal Cancellation (B = −0.66, SE = 0.19, t =
−3.58, p < .005), Purdue Pegboard (B = −0.20,
SE = 0.05, t = −3.67, p < .005),
and CANTAB Spatial Working Memory (B = 0.74, SE = 0.24,
t = 3.08, p < .005). Birth weight accounted for
unique variance in ROCF copy (B = 4.88, SE = 1.65, t
= 2.95, p < .005), CANTAB Rapid Visual Processing (B =
6.70, SE = 2.08, t = 3.22, p < .005), and
Contingency Naming (B = 0.32, SE = 0.11, t = 2.86,
p < .01). Weight for gestational age did not account for
unique variance in any of the outcomes.
Secondary analyses examined the effects of several additional
neonatal risks, including neonatal jaundice as defined by indirect
serum bilirubin >10 mg/dL, neonatal apnea, necrotizing
enterocolitis, septicemia, and 5-min Apgar scores <6. Regression
analyses failed to reveal any associations of these factors with the
cognitive outcomes. Thus, of the factors available for study, lower
birth weight, lower weight for gestational age, and a longer period of
oxygen requirement were the major conveyors of risk for cognitive
sequelae.
DISCUSSION
Support for Hypotheses
The findings of this study support all three of our hypotheses.
First, neuropsychological impairments documented in previous follow-up
studies of middle school-age children with VLBW (Botting et al., 1998; Rickards
et al., 2001; Saigal et al., 2000;
Taylor et al., 2000b, 2004) persisted into later adolescence and early
adulthood. Consistent with past investigations, cognitive deficits were
wide-ranging but most pronounced in children with the lowest birth
weights (Breslau et al., 1996; Horwood et al., 1998; Klebanov
et al., 1994; Korkman et al.,
1996).
Despite limited evidence for gradient effects, the findings are also
in accord with research indicating that risks for cognitive sequelae
extend across the full range of <1,500 g birth weight (Breslau et al., 1996). Both VLBW groups had higher
rates of some cognitive deficits than the term group, and outcomes for
the 750–1,499 g group fell consistently between those for the 750
g and term groups (see Tables 3 and 4). Absolute birth weight, moreover, predicted
outcomes for the total VLBW sample, with no evidence that this factor
had differential effects in the two VLBW groups. Additionally, children
with VLBW with longer periods of oxygen requirement had poorer outcomes
on several measures even when controlling for birth weight and weight
for gestational age. Recent increases in the survival of neonates born
at the limits of viability justified recruitment of children with
<750 g birth weight (Hack et al., 1996b),
but this birth weight cannot be viewed as a threshold above which
outcomes are uniformly positive.
Second, several findings indicated specific effects on cognition.
Although the <750 g group had lower IQs than term controls, group
differences were more pronounced on measures of visual–motor
skills, memory, and executive function than on language tests; and no
group differences were found on the Boston Naming Test. Differences
between the <750 g group and term controls on the former measures
could not be attributed to the greater prevalence of severe
disabilities in the <750 g group, and these differences remained
when controlling for verbal–semantic skills. Similarly, in
regression analysis of outcomes for the aggregate VLBW sample, birth
weight and weight for gestational age predicted visual-motor skills,
memory, and executive function, but not language outcomes; and the
former associations held up even when controlling for Vocabulary.
Luciana et al. (1999) documented deficits
in executive function and spatial memory in a preterm sample using the
CANTAB tasks. However, their sample was assessed between 7 and 9 years
of age, controls were not matched to the preterm group in background
characteristics, and other cognitive domains were not assessed. The
present results extend their findings by demonstrating that deficits on
CANTAB tasks persist to later ages and are accompanied by an uneven
cognitive profile, with relative deficiencies in visual–motor
skills as well as in those abilities measured by the CANTAB.
Third, cognitive outcomes in the VLBW sample were worse for
individuals at higher risk for brain insult. Consistent with previous
literature (Brazy et al., 1993; Horwood et al., 1998; Hutton et
al., 1997; Landry et al., 1993; Marlow et al., 1989; Perlman,
2001; Pinto-Martin et al., 1999; Taylor et al., 1998), lower birth weight, lower
weight for gestational age, and longer periods of oxygen requirement
predicted poorer cognitive performance. The failure of periventricular
abnormalities on neonatal ultrasounds to predict cognitive outcomes was
unexpected in light of previous findings from our follow-up study and
others (Frisk & Whyte, 1994; Jacobson et al., 2001; Leonard
et al., 1990; McGrath & Sullivan,
2002; Roth et al., 1993; Selzer et al., 1992; Taylor et
al., 1998). The relatively small sample size, reduced from
earlier follow-ups by sample attrition, may have restricted our power
to detect these associations. Another possibility is that longer-term
outcomes are influenced primarily by insults to which ultrasounds are
insensitive, such as the diffuse white matter abnormalities (Counsell et al., 2003; Volpe,
2003).
Birth weight, weight for gestational age, and length of oxygen
requirement were associated with many of the same outcomes. However,
the patterns of associations were somewhat different for these risk
factors. Longer periods of oxygen requirement explained variability in
language tests, while the weight indices did not. As further evidence
for differential effects of the risk factors, the duration of oxygen
requirement and weight indices accounted for unique variance in several
scores. These findings are consistent with results from other studies
of the effects of chronic lung disease on development (Farel et al., 1998; Hughes et
al., 1999) and suggest that this neonatal complication of VLBW
may predict more severe or generalized cognitive impairment. The
differential pattern of risk–outcome associations also raises the
possibility that the chronic lung disease has effects on development
that cannot be explained by prematurity alone, perhaps due to brain
insults secondary to recurrent hypoxemia and other postnatal
complications of this condition (Durand et al.,
1992; Garg et al., 1988; Perlman, 2001). The failure of weight for
gestational age to account for unique variance in outcomes suggests
that growth retardation may be of limited prognostic value independent
of other neonatal complications (Hack, 1997).
Alternatively, a larger sample may be needed to isolate the effects of
this risk factor. The restriction in range of low weight for
gestational age among children with VLBW, reflecting the improbability
of survival at the lower extremes, sets natural constraints on the
study of this predictor.
Study Implications
Given the vulnerability of periventricular white matter and
subcortical regions to hypoxic-ischemia and other complications of
prematurity (Dammann & Leviton, 1999;
Volpe, 2003), the pattern of impairment
observed in this study most likely reflects early and partially
localized brain insult. Later-occurring damage to structures in this
region, including frontostriatal circuits and parietal–occipital
and temporal cortex, affects cognition in similar ways (Broadbent et al., 2002; Knowlton, 2002; Salmon et al.,
2001). The concordance of the deficits in this VLBW sample with
expectations based on adult structure–function relations suggests
that these brain systems mediate behavior in a relatively fixed manner
from an early age. The early establishment of brain functions has been
documented in experimental studies of the effects of medial temporal
lobe lesions on memory in primates (Bachevalier
& Malkova, 2000), as well in human studies demonstrating
early specialization for some spatial and language skills in the
cerebral cortex and for motor control in the cerebellum (Bates & Roe, 2001; Dennis et al.,
1999; Neville & Bavelier, 2001; Stiles, 2000). In common with these and other
investigations (Taylor & Alden, 1997;
Taylor et al., 2000d), the present findings
imply constraints on early neural plasticity.
Evidence for wide-ranging cognitive impairments in this VLBW sample,
including lowered IQ in many of the participants, may be explained by
diffuse neonatal neuropathology or by the effects of injury to
subcortical white matter and the subependimal germinal matrix on
cortical development (Inder et al., 1999;
Inder & Volpe, 2000; Perlman, 2001). Another possibility is that
deficits in executive function and memory, even if due primarily to
periventricular damage, could have generalized effects on cognitive
development, including knowledge acquisition or growth in fluid
intelligence (Bennetto et al., 2001; Schatz et al., 2000). We emphasize, however, that
findings from this study and our previous follow-up of the VLBW sample
(Taylor et al., 2004) fail to support a
generalized slowing of cognitive development with increasing age. Early
damage to frontostriatal circuits appear to have ongoing effects on
some abilities without severely disrupting all processes of skill
acquisition (Denckla & Reiss, 1997; Pennington, 1994).
A further study implication is that the “subcortical”
profile of deficits observed in the present VLBW sample may be unique
to preterm children. Hypoxic ischemia in full-term children is more
likely to be associated with multiple areas of cortical–neuronal
loss (Volpe, 2001). The latter insults would
be expected to yield less uniform deficits in visual–perceptual
abilities and executive function, greater impairment in
verbal–semantic skills, and perhaps more frequent global
cognitive dysfunction. Differential outcomes of hypoxic ischemia in
preterm versus full-term children may nevertheless prove
difficult to tease out. Differences of this sort would be obscured by
the effects of early periventricular damage on cortical development,
the sensitivity of the hippocampus to hypoxic ischemia occurring at any
point in development, and the role of hippocampus in memory and spatial
processing (Luciana et al., 1999).
The results have several additional implications for clinical
practice. To begin with, many of the group differences had large effect
sizes, with substantial group differences in rates of deficient test
performance. Relative to term controls, one or both of the VLBW groups
had higher rates of abnormal scores on tests of visual–motor
skills, visual memory, and executive function. Higher rates of deficits
were found in the <750 g group even when excluding participants with
neurosensory disorders or low IQ, suggesting that the tests were
sensitive to more subtle manifestations of sequelae. Group differences
in rates of deficits for the total sample (see Table
4) took environmental risks such as low SES into account.
Because practitioners typically do not adjust for the effects of
background factors other than age in identifying deficient test
performance, the relative risks of impaired test scores in children
with VLBW seen for clinical assessments may be even higher than those
suggested by our data.
VLBW also has more pronounced effects on some skills than on others.
Language abilities appear to be relatively spared, though such sparing
is likely to vary with the age at assessment and test demands (Frisk & Whyte, 1994; Luoma
et al., 1998). In contrast, VLBW has pronounced effects on
spatial recognition memory, spatial working memory, spatial planning,
and attention shifting. Examination of performance parameters from the
CVLT–II and CANTAB tasks provided additional insight into these
cognitive deficiencies. Children with more extreme low birth weight
were prone to excessive intrusions and poor use of semantic clustering
in verbal recall. Moreover, they displayed impulsive responding, slowed
response times, and poorly organized approaches to self-directed,
multi-step tasks. This pattern of cognitive strengths and weaknesses
has been noted in other studies of children with VLBW and is consistent
with descriptions of children with nonverbal learning disabilities
(Fletcher et al., 1997; Frisk & Whyte, 1994; Grunau
et al., 2002; Rourke, 1989; Woods et al., 2000).
Contrary to expectations based on group differences in verbal
learning at an earlier follow-up (Taylor et al.,
2000c), the CVLT–II had limited sensitivity to the
long-term effects of VLBW. This unexpected finding may reflect either
the participants' familiarity with the procedure as a result of
previous administrations of a similar task, or increases with age in
the use of compensatory strategies, such as verbal rehearsal.
Associations of birth weight and other neonatal risks with performance
on the Judgment of Line Orientation Test are also noteworthy in view of
questions regarding the effects of VLBW on “motor-free”
measures of spatial ability (Foreman et al., 1997;
Goyen et al., 1998; Jacobson
et al., 2001). The present results suggest that children with VLBW
are vulnerable to weaknesses in this area, though impaired executive
function also may have contributed to deficient performance.
A final clinical implication is that survivors of VLBW who exhibit
cognitive deficits are likely to have unique treatment needs, whether
in primary or secondary school or in higher education programs. Their
relatively good language skills may camouflage more significant
weaknesses in learning, including difficulties in following complex
instructions, poor organization, and slowness in acquiring information
and in mastering new concepts. Methods that may enhance learning
include direct teaching of organizational strategies, extra time and
direction in preparing for and taking tests, use of verbal mediation,
and repetition or alternative instructional approaches (Luciana et al., 1999; Thompson,
1999; Taylor et al., 2000c). Because
cognitive weaknesses may be associated with attention or other behavior
problems, behavior management may also be required (Botting et al., 1997; Woods et
al., 2000).
Limitations
The study's most serious limitation was the lack of direct
measures of neuropathology or brain oxygenation. VLBW survivors with
more extreme low birth weight and longer periods of neonatal oxygen
requirement were at higher risk for brain insults. However, we had no
means to investigate the nature or extent of brain insult. Other
limitations relate to sampling issues. Attrition occurred over the
course of follow-up, with disproportional drop-out in lower SES
families and in participants with lower general cognitive ability.
Although the drop-out rate was low relative to the length of the
follow-up period (10 years since initial recruitment) and SES was taken
into account in the analysis, caution is advised in generalizing
findings to the larger VLBW population. Despite our relatively large
group of survivors with <750 g birth weight, another concern is that
the size of the lower-risk 750–1,499 g group may have been
insufficient to detect some of the more subtle cognitive consequences
of VLBW in these children. Generalizations based on the present results
are further limited by changes in neonatal intensive care practices
since this VLBW sample was born. Because the increased survival of
children with extremely low birth weight has been associated with
higher morbidity rates (Hack et al., 1996b;
Fanaroff et al., 2003), outcomes may be
somewhat different for later-born cohorts. Studies of more recently
born cohorts of children with VLBW, however, suggest outcomes similar
to those observed in this and other samples of earlier cohorts
(Anderson et al., 2003, in
press).
Conclusions and Future Directions
Despite these limitations, this study is one of the few to
investigate the later outcomes of more extreme low birth weight. A
benefit of our recruitment procedure, which focused on children with
<750 g birth weight, is that it provided for more variability in
risks and outcomes than is typical in samples of children with VLBW.
Other unique aspects of the study were the comprehensiveness of the
test battery and assessment of neonatal risk factors as predictors of
these outcomes. The findings indicate that neuropsychological deficits
persist as survivors of VLBW enter later adolescence and early
adulthood. These impairments are consistent with insult to the
periventricular region, and neonatal risks are valid predictors of
long-term outcome.
Further research using more direct measures of residual
neuropathology, as well as more thorough assessment of environmental
influences, is needed to understand the biological basis of variations
in outcomes (Taylor et al., 2002). Rigorous
measures of brain status will also be useful in investigating
functional sparing, or the extent to which children can develop
normally despite early brain insults. If sparing can be demonstrated,
the factors associated with it may provide guidance as to how to
promote better outcomes. We are currently examining brain volumes in
our cohort using MRI morphometry. These findings will shed light on the
effects of early periventricular insults on neural development and
increase the precision with which we are able to map
brain–behavior relations. To determine the broader functional
implications of VLBW, we have also collected data on academic
achievement and behavior. Future research will be required to better
define the nature of neuropsychological strengths and weaknesses in
this population and to explore implications for adult educational and
vocational attainments and quality of life (Hack,
1999).
