Background. Cognitive development in very low birth weight (VLBW, ≤1500 g) infants typically has been reported based on mean endpoints in cross-sectional studies. These overall group means mask individual patterns of cognitive development. Given the heterogeneity of VLBW infants, it is important to identify individual patterns of development and the factors associated with the different patterns.
Objective. We sought to determine individual patterns of cognitive development over the first 6 years in VLBW children and to examine the relative influence of selected biomedical and sociodemographic factors on these patterns.
Method. VLBW infants (N = 203) were followed from birth to six years. Cognitive scores were obtained at four yearly intervals, and biomedical and social data were obtained beginning with the neonatal period. Cluster analysis was used to identify individual patterns of cognitive development.
Results. Five developmental patterns were identified: average–stable (13% of the sample); average–declined to low average (24% of the sample); average–declined to below average (43% of the sample); very low–increased to low average (8% of the sample); and very low–stable (12% of the sample). The patterns could be differentiated by several biomedical factors, including birth weight, gestational age, neonatal health, and 1-year assessments of neurological status and head circumference, as well as by level of maternal education. In particular, abnormal neurological status at 1 year was associated with a pattern of very low stable scores, and a suspicious status was associated with a pattern of improving cognitive development. Maternal education was influential among children born at the upper end of the VLBW range, who had a more favorable set of biomedical factors.
Conclusions. Biomedical factors are of major importance for the cognitive development of VLBW infants, and their influence increases as birth weight declines. Differences in neurological integrity at 1 year were an important indicator of different patterns of cognitive development, especially for infants at the lower end of the VLBW range.
- very low birth weight
- cognitive developmental patterns
- cluster analysis
- biomedical risk
- sociodemographic risk
Low birth weight (LBW, ≤2500 g) preterm infants are widely reported to be at increased risk for developmental disabilities, with the risk increasing as birth weight decreases.1-6Almost universally, the increased morbidity reported for LBW children has been based on methods defining various outcomes for them as a group and examining the statistical effects of various factors on these group outcomes. However, LBW infants are also known to be a very heterogeneous group, both in biomedical and sociodemographic characteristics and in developmental outcome. The typical reports of lower cognitive status, based on mean endpoints, mask important differences in individual patterns of development. Yet the differentiation of these patterns is crucial, so that we can begin to identify factors that are associated with diverse developmental courses. It is particularly crucial to explore this problem in infants of very low birth weight (VLBW, ≤1500 g), because medical technical advances have dramatically increased their rates of survival, but they continue to be at very high risk for poor cognitive development.
Only one study7 has attempted to identify patterns of individual difference in cognitive development in LBW children, and these researchers examined patterns only to the age of 3 years. In that study, only 25% were VLBW; the remainder were relatively heavier infants (≥1500 g and ≤2500 g). They identified five patterns of cognitive development over the first 3 years of life: three patterns declining to average or below average; one high stable; and one very low stable. Some of these patterns had been hypothesized; for example, because LBW children are likely to show lower intelligence scores especially during the first few years of life, a below average stable group was expected. A pattern of increasing scores was also expected, in part because of “catch-up” and also because the evaluation instruments used at the endpoint of age 3 years rely less on the perceptual/motor skills known to be problematic for many LBW infants. Yet this pattern of upward trend was not found, even though a large portion of these children received intensive intervention. An upward trend might be more likely in children of lower birth weights and higher perinatal risk, who initially have lower scores but who may recover from their early disadvantage, given additional time.
The present study focused on VLBW children, using cluster analysis to identify individual patterns of cognitive development. The association of these patterns with several important biomedical and sociodemographic factors was then examined. Low birth weight and gestational age are well-recognized biomedical factors associated with increased risk for developmental delays, and they were associated with patterns of development in the study of LBW children.7These and other biomedical factors would be expected to be even more important for the lower birth weight, higher risk children in the present study. Therefore, birth weight, gestational age, and additional biomedical risk factors were examined. In particular, an index of perinatal medical risk was included, because this might be expected to influence both “catch-up” and extent of medical complications associated with very early birth. Also included were three variables known to be associated with cognitive outcome in VLBW children,8-10 but whose effects on patterns of individual differences have not been explored: the infant's birth weight in relation to its gestational age, that is, whether they were small for gestational age (SGA), and head circumference and neurological status at 1 year of age.
Sociodemographic factors have a well-known impact on cognitive outcome. This has been demonstrated both in children of normal birth weight11 and in children of LBW.5,10,12,13Although sociodemographic factors would certainly be expected to have an impact on patterns of cognitive development observed in VLBW children, their effects might vary with different combinations of biomedical factors.
Thus, the purpose of the present study is (1) to determine patterns of individual differences in cognitive development over the first 6 years of life in VLBW children, and (2) to examine the relative impact of selected biomedical and sociodemographic factors on these patterns.
Subjects followed were 203 VLBW infants, 92 boys and 111 girls, born between 1975 and 1989 and recruited from the Neonatal Intensive Care Units of three hospitals in the Bronx that were part of the Department of Pediatrics of the Albert Einstein College of Medicine. Criteria for inclusion in the present study were a birth weight ≤1500 g and cognitive data at four points in time (1; 2; 3 or 4; and 5 or 6 years of age).
Table 1 presents background data on the study population. The range of birth weights of this group was broad, with 25% weighing ≤1000 g. One third of the infants were classified as SGA, defined as less than the third percentile.14 The degree of neonatal morbidity was also heterogeneous, as indicated by the Neonatal Health Index (NHI15). The NHI was calculated on the basis of length of stay in the neonatal intensive care unit, adjusted for birth weight and standardized to a mean of 100, with higher scores indicating better health. This index had the advantage that it could be calculated for every child in the sample, whereas data on specific health problems were missing for some children. (The NHI was significantly related to the presence and severity of various health problems. For example, infants with bronchopulmonary dysplasia had a significantly lower mean NHI than those without bronchopulmonary dysplasia [71 vs 89.7, t = −4.63, df= 173, P < .001]; NHI was significantly correlated with intraventricular hemorrhage, coded on a four-point scale,r = .48, P < .001). The children were primarily from minority ethnic and racial backgrounds; 41% of their mothers did not complete high school.
Cognitive performance was measured using the Mental Development Index (MDI), of the Bayley Scales of Infant Development, at 12 and 24 months of age, the Stanford-Binet Intelligence Scale at 4 years of age (the 3-year score was used if the 4-year score was missing, which occurred for 27% of the sample), and the Wechsler Intelligence Scale for Children—Revised at 6 years of age (the Wechsler Preschool and Primary Scale for Intelligence at 5 years was used if the score at age 6 was missing, which occurred for 36% of the sample). These are referred to as IQ 3/4 and IQ 5/6. Combining these ages ensured a cognitive index at each age point (ie, 1; 2; 3/4; and 5/6) for all children in the sample. Scores to age 4 were all corrected for gestational age at birth; scores at ages 5 and 6 were not corrected.
Neurological status was based on examination of the child at 1 year corrected age, using three classifications: (1) abnormal (16% of the sample), a severe abnormality, such as cerebral palsy, global hypotonicity, chronic seizures, hydrocephaly, blindness, or severe sensorineural hearing loss, for which a diagnosis could be established; (2) suspicious (46% of the sample), some atypical or questionable signs in tone, reflexes, gait, or movement, but for which there is no definitive diagnosis or syndrome; and (3) normal (38% of the sample), based on a completely normal neurologic examination.
Head circumference at 1 year corrected age was dichotomized into groups of ≤10th percentile (32% of the sample) and >10th percentile, based on separate growth charts for boys and girls.
Patterns of Cognitive Development
Using the cognitive test scores described above, a hierarchical agglomerative cluster analysis (Ward's method) was performed to identify relatively homogeneous subsets of children with similar developmental patterns.16,17 In the first step of this analysis, each individual is its own cluster, and each new cluster is formed by the merging of previous clusters; once a cluster is formed, it cannot be split. The cluster solution that best represents the data can be determined by examination of the coefficients in the agglomeration schedule, seeking a point where a sizable change occurs and selecting the number of clusters just before that point.18
The clusters were validated in two ways. First, the variables that wereinternal to the cluster analysis, that is, the cognitive test scores, were used in a discriminant analysis, to determine the percentage of children correctly classified, and in ANOVAs, to compare the mean cognitive scores across the clusters at each age. Second, variables external to the cluster analysis, that is, biomedical and sociodemographic characteristics of the children, were compared across the clusters, and discriminant analysis was performed to determine the relative contribution of the independent variables to cluster membership.
All analyses were performed using SPSS/PC+ programs.
Patterns of Cognitive Development
Based on the criteria described above, a five-cluster solution was selected as best representing the data. The five patterns of cognitive development are illustrated in the Figure, and the mean cognitive scores of the children in the clusters are presented in Table2. Discriminant analysis indicated that a total of 90% of the sample was correctly classified and that only one cluster was <90%; clusters D and E had the best fit (100% correctly classified).
Three of the clusters had mean IQs within the average range at 1 year; two of these had parallel profiles and one diverged somewhat. Cluster A (13% of the sample), labeled as average–stable, had mean cognitive test scores that remained in the average range and above the standardization mean of 100 throughout the 6 years. Cluster B (24% of the sample), labeled as average–declined to low average, had scores that decreased between 12 and 24 months, and then remained in the low average range, with means around 90. Cluster C (43% of the sample), labeled as average–declined to below average, had a profile of decline similar to that of cluster B, but cluster C started almost 20 points below cluster B and then was ∼10 points below after that; thus, cluster C remained >1 SD below the normative mean after 12 months. The last two clusters initially had mean cognitive test scores indicating significant delay. Cluster D (8% of the sample) was labeled as very low–increased to low average, because they began with scores >2 SD below the normative mean at 1 year and then steadily increased, ending with a low average mean IQ of 88 at 5/6 years. Cluster E (12% of the sample), labeled as very low–stable, displayed mean cognitive test scores that remained >2 SD below the normative mean at every age.
Associations With Patterns of Cognitive Development
As indicated in Table 3, the clusters differed significantly on all the independent variables examined except SGA. Clusters A, B, and C had comparable mean birth weights (∼1200 g), significantly higher than clusters D and E. They also had significantly higher percentages of children who were neurologically normal than clusters D and E (partitioning df,19χ2 = 16.25, df = 1, P < .001). Cluster A, with the best pattern of cognitive development, had a significantly higher level of maternal education than all other clusters combined (χ2 = 11.76, df = 1,P < .001). Cluster B had the highest NHI, significantly higher than that of cluster C, and close to average for their birth weight; their neurological status was the most optimal, significantly better than that of either cluster A (χ2 = 5.12, df = 2, P < .05) or cluster C (χ2 = 8.79, df = 2, P < .025), but their level of maternal education was not as favorable as that of cluster A (χ2 = 5.22, df = 1,P < .05). In cluster C, the level of maternal education was even less favorable than that of cluster B (χ2 = 12.27, df = 1, P < .001), and the children also had a significantly lower NHI than did children in cluster B. Clusters D and E were similar in very low birth weight, early gestational age, and a high proportion of males, significantly higher than clusters A, B, and C combined (χ2 = 12.24, df = 1, P < .001). However, cluster D also differed from cluster E; the children in cluster D had a significantly better NHI, and fewer had an abnormal neurological status (χ2 = 9.92, df = 2,P < .025) or a subnormal head circumference (≤10th percentile; χ2 = 4.74, df = 1,P < .05). Cluster E had indicators of greater biomedical risk than every other cluster. The mean birth weight of the children in this cluster was <1000 g, the NHI was significantly below that of all other clusters, more than two thirds were classified as neurologically abnormal at 1 year, and a similarly large proportion had a subnormal head circumference at 1 year. (Information on specific neurological diagnoses is based on the 3-year neurological examination; these diagnoses were essentially unchanged for the 5/6-year examination. At 3 years, 20% of the cohort were neurologically abnormal; the diagnosis for 62% was spastic diplegia. Specific diagnoses for abnormal cluster members were the following: cluster A, 3 abnormal [3 spastic diplegia]; cluster B, 2 abnormal [1 hemiplegia and 1 hemiplegia + seizure disorder]; cluster C, 12 abnormal [6 spastic diplegia, 1 spastic diplegia + porencephalic cyst, 1 triplegia, 2 microcephaly, 1 quadriplegia, 1 severe sensorineural hearing loss]; cluster D, 4 abnormal [3 spastic diplegia, 1 triplegia]; cluster E, 18 abnormal [5 spastic diplegia, 3 triplegia, 3 quadriplegia, 1 spinal lesion, 2 microcephaly, 2 severe visual deficit, 2 multiple diagnoses including hypotonia, microcephalia, and seizure disorder; and microcephalia, severe visual deficit, seizure disorder, and spastic diplegia]).
Discriminators of Developmental Patterns
Independent variables used in the discriminant analysis were birth weight, gender, NHI, neurological status, and maternal education. Omitted from the analysis were (1) SGA, because it was found to be unrelated to cluster membership (see Table 3), (2) head circumference at 1 year, because of missing observations, which disproportionately affected some of the clusters, and (3) gestational age, which failed to reach the statistical requirement for entry into the analysis (because the correlation with birth weight was high). Two discriminant functions were obtained (Table 4) that differentiated among the five clusters. The first, labeled biomedical, was significant (Wilks' λ = .58, P < .0001, after the first function was derived), and the second, labeled sociodemographic, was marginal (Wilks' λ = .90, P < .09, after the second function was derived). The biomedical function accounted for 84% of the total between-group variability. Examination of the discriminant function coefficients of the independent variables on each function presented in Table 4 indicates that only neurological status contributed positively to the biomedical function (>.50) when other variables were held constant. The remaining variables made only minor contributions. Maternal education and the child's gender each contributed positively, and at a similar level, to the sociodemographic function, which accounted for 10% of the total between-group variability.
The group means of the five clusters on the two discriminant functions are presented in Table 5. Cluster A (average–stable) was average on the biomedical function and highest on the sociodemographic function. Cluster B (average–declined to low average) was highest on the biomedical function and average on the sociodemographic function. Cluster C (average–declined to below average) was average on the biomedical function and moderately low on the sociodemographic function. Cluster D (very low–increased to low average) was low on both functions. Cluster E (very low–stable) was lowest on both, and especially low on the biomedical function.
This study examined the patterns of cognitive development in a sample of VLBW children, many of whom had poorly educated mothers, placing them at risk for developmental delays on both biomedical and sociodemographic grounds. Five developmental patterns were identified by cluster analysis. The clusters were subjected to validation analyses and found to be sufficiently different from one another on the criterion variables to indicate good validity. Their developmental patterns indicated that throughout the first 6 years of life, 37% (clusters A and B) of the children performed consistently in the average range, 12% did consistently poorly (E), and 8% (D) showed a steady improvement. The largest group (43%) (C) declined from average to below average.
With one exception, all the independent variables examined were significantly associated with cluster membership. The exception was SGA. The absence of a relationship between SGA and outcome is probably attributable to the relatively limited birth weight range of the children studied (≤1500 g), because a report of the LBW (≤2500 g) cohort from which the present sample was drawn has shown SGA to be related to lower cognitive scores.8
Four of the five clusters identified in the present study had counterparts in the Liaw and Brooks-Gunn7 study of larger LBW children. Two clusters, C (average–declined to below average) and E (very low–stable), were quite similar to clusters in their study, both in profile (ie, the shape of the curve) and absolute level (ie, the magnitude of the mean test scores). Two other clusters, A (average–stable) and B (average–declined to low average) were similar in profile to two of their clusters but had lower absolute levels of cognitive performance.
The cluster that had no counterpart in the Liaw and Brooks-Gunn study was Cluster D (very low–increased to low average), which showed a pattern of scores that increased with age. It may be, as was suggested earlier, that this increasing pattern is associated with the greater vulnerability of the present sample and that “catch-up” occurs as the early medical complications resolve.
In both studies, the children were comparable in socioenvironmental background, with similar proportions of mothers who did not complete high school, and, with one exception, there were similar distributions of maternal education in clusters with similar profiles. The exception was the very low stable cluster, in which 21% of the mothers in the Liaw and Brooks-Gunn study did not finish high school, compared with 63% in the present study. Liaw and Brooks-Gunn suggested that biomedical problems may have overwhelmed the social factors in their very low stable cluster. However, other researchers point to evidence that the combined effects of severe neonatal insult and high social risk can be devastating.5,20,21 This appears to have occurred in the present study, in which children with the poorest set of biomedical factors and the lowest level of maternal education had a pattern of persistently very low scores (cluster E).
In general, given the overall comparability in level of maternal education, it appears that the major differences in patterns of cognitive development between the two studies—the lower absolute levels of cognitive performance in clusters A and B and the emergence of a “catch-up” cluster in the present study—are most likely attributable to the greater biomedical insults experienced by VLBW infants than by LBW infants.
Within the present study itself, there were two interesting contrasts. First, two clusters had similarly poor mean cognitive scores at 1 year, but one (cluster E) remained very low throughout, whereas the other (cluster D) improved and eventually attained a mean in the low average range by 5/6 years of age. Comparison of these two clusters reveals that although birth weights and gestational ages were similar, the group that improved had better neonatal health, more normal 12-month head circumference, and better neurological status. Particularly striking is the finding that two thirds of those whose scores remained low (cluster E) had clear neurological abnormality at 12 months; in contrast, two thirds of those whose cognitive scores improved (cluster D) were diagnosed as neurologically suspicious, but only 19% had clear neurological abnormality. The group that improved across age presents a picture of resilient children whose early depressed development was primarily associated with extreme LBW and general immaturity of the nervous system but without clear neurological abnormality. By 5/6 years of age, their mean IQ was on a par with that of children born at higher birth weight, greater gestational age, and a far better set of biomedical factors (cluster B). Moreover, 73% of those who had been neurologically suspicious at 1 year were now considered neurologically normal. This is consistent with reports of other researchers22-24 that mild hypotonia (classified here as neurologically suspicious) has been found to diminish between infancy and 2 to 3 years of age in preterm children.
The second interesting contrast involves two clusters that show the inverse picture, namely, both had relatively good mean cognitive scores at 1 year, but one maintained its good standing (cluster A), whereas the other declined (cluster B). The group that maintained its advantage had a relatively good biomedical history and the highest level of maternal education of all the clusters. The group that declined was the least compromised in terms of biomedical factors; however, its relatively benign medical course was apparently not enough to compensate for less favorable sociodemographic factors, namely, that three times as many mothers had not graduated from high school.
Overall, there was a distressingly large number of children who declined in cognitive performance both in the present study (67%) and in the Liaw and Brooks-Gunn7 study of larger LBW children (81%). The degree to which the decline might be spurious, attributable to an inflation of 12-month Bayley scores, as suggested by Liaw and Brooks-Gunn, or attributable to the lower social status of both samples, is not clear. Some researchers have noted that the negative impact of low social status on cognitive development is generally not evident for the first 18 to 24 months,12,25 and that may explain the precipitous drop seen at 2 years.
Several significant implications can be drawn from this investigation. First, a neurological assessment at 1 year may be an important marker of subsequent patterns of cognitive development, especially among children at the low end of the VLBW range, a group for whom survival rates have increased dramatically in recent years. Cluster E, which was overrepresented by children of abnormal neurological status, was characterized by a persistence of very poor test scores. Cluster D, which was overrepresented by children with a suspicious neurological status, manifested an improving pattern of cognitive development; by early school age, most of the children's suspicious signs had resolved, and the group displayed cognitive scores in the low average range. Without an investigation of patterns of cognitive development, the latter group would not have been clearly identified.
Second, a unique pattern (cluster D) and lower absolute levels of cognitive development differentiated the present study from a study of larger LBW children.7 Given the comparability of maternal education, the results point to the increased importance of biomedical factors in the cognitive performance of VLBW children, and demonstrate further that the influence of biomedical factors increases as birth weight declines. However, their influence varies, even when birth weights are similar.
Finally, in general, socioenvironmental conditions were found to be less important than biomedical factors. However, among the larger VLBW infants, favorable socioenvironmental conditions, as indicated by maternal education, were influential, with a higher level contributing to the maintenance of optimal cognitive development (cluster A) and a lower level contributing to a decline (cluster B). Again, this differentiation would not have been apparent without an examination of patterns of development.
By determining the various patterns of cognitive development present among VLBW children and examining their correlates, the present study helps to identify the characteristics of the children and their families that are associated with different developmental trajectories. Most important, it points up the individual differences in patterns of cognitive development, with some children remaining relatively stable over time and others demonstrating increases or declines. Such differences are masked in cross-sectional studies. The findings make clear that for VLBW children, the role of the pediatrician includes continued developmental surveillance and educational advocacy.
This work was supported by a grant from the National Institute of Child Health and Human Development to the Rose F. Kennedy Center for Research in Mental Retardation and Human Development (HD01799).
- LBW =
- low birth weight •
- VLBW =
- very low birth weight •
- SGA =
- small for gestational age •
- NHI =
- neonatal health index
- Kopp CB,
- Kaler SR
- Hoy EA,
- Bill JM,
- Sykes DH
- Hunt JV,
- Bruce AB,
- Cooper BAB,
- Tooley WH
- Eilers BL,
- Desai NS,
- Wilson MA,
- Cunningham MD
- McCarton CM,
- Wallace IF,
- Divon M,
- Vaughan HG
- Vohr BR,
- Garcia-Coll CT
- Hack M,
- Breslau N
- ↵McCall RB, Applebaum, MI, Hogarty PS. Developmental changes in mental performance. Monogr Soc Res Child Dev. 1973;32 (Serial No. 150)
- Escalona SK
- Scott DT,
- Bauer CR,
- Kraemer HC,
- Tyson J
- ↵Aldenderfer MS, Blashfield RK. Cluster Analysis. Beverly Hills, CA: Sage; 1984
- ↵Green JA. Analyzing individual differences in development: correlations and cluster analysis. In: Colombo J, Fagen J, eds. Individual Differences in Infancy: Reliability, Stability, Prediction. Hillsdale, NJ: Erlbaum; 1990:77- 109
- ↵Maxwell AE. Analysing Qualitative Data. London, England: Methuen; 1967:52–62
- Cohen S,
- Parmelee A,
- Sigman M,
- Beckwith L
- Ross G,
- Lipper EG,
- Auld PAM
- ↵McCall RB. The development of intellectual functioning in infancy and the prediction of later IQ. In: Osofsky JD, ed. Handbook of Infant Development. New York, NY: Wiley; 1979:707–741
- Copyright © 1997 American Academy of Pediatrics