Objectives. To examine the prevalence, stability, and predictors of clinically significant behavior problems in 869 preterm low birth weight (LBW) infants at 3, 5, and 8 years of age.
Methods. A prospective cohort study was conducted. Clinically significant behavior problems were assessed using dichotomized total problem Child Behavior Checklist scores in LBW children at ages 3, 5, and 8 years. Baseline sociodemographic and obstetric data were collected. Maternal General Health Questionnaire performed at 40 weeks' gestation was dichotomized at a score of 12 to give a measure of maternal psychological distress. Prevalence and stability of behavior problems at ages 3, 5, and 8 were determined and potential predictors of behavior problems at age 3, 5, and 8 were examined using multiple logistic regression.
Results. Prevalence of behavior problems remained at ∼20% at 3, 5, and 8 years. Stability of behavior problems between different ages was ∼50%. Significant behavior problems at ages 3, 5, and 8 were predicted by maternal psychological distress at 40 weeks (odds ratio [OR]: 1.59; 95% confidence interval [CI]: 1.21–2.09), maternal cigarette smoking during pregnancy (OR: 1.57; 95% CI: 1.20–2.04), Hispanic ethnicity (OR: 2.00; 95% CI: 1.24–3.24), and maternal age (OR: 0.97; 95% CI: 0.94–0.99).
Conclusions. This sample had double the prevalence of behavior problems expected in the general child population. These problems showed stability over time. Cigarette smoking in pregnancy, maternal psychological distress at 40 weeks' gestation, maternal age, and Hispanic ethnicity all were significant predictors of the development of behavior problems from ages 3 to 8. These findings have implications for health policies on smoking and postnatal depression.
Preterm low birth weight (LBW) infants are at an increased risk for the subsequent development of both cognitive and behavioral problems in childhood.1,2 These problems have implications for school performance3 and later life.4 Although the factors that affect cognitive development in this group have been examined extensively, less attention has been paid to the factors that affect behavior.
LBW infants have higher average scores on measures of problem behavior than control subjects1 and are more likely to have clinically significant behavior disorders.5 In particular, there is a well-documented association between prematurity and later development of attention-deficit/hyperactivity disorder.6 Although the prevalence of behavior disorders in LBW infants has been established in the short term, few studies have examined stability of behavior disorders in these children in the longer term. Two studies that have reported that 40% to 50% of LBW children with significant behavior problems continue to display them 3 years later.5,7
As with cognitive development in LBW infants, the development of child behavioral problems is known to be influenced by a number of environmental factors,8,9 but many of the existing follow-up studies of LBW children have reported on relatively few predictors of outcome other than gender, ethnicity, and socioeconomic status. This may be in part because they have included small populations from single sites. These studies generally have not identified factors during pregnancy or the neonatal period that might be modified to reduce the risk of behavior problems, but this is important if we want to improve the content of early intervention programs. Therefore, there is an urgent need for studies on preterm LBW children that 1) recruit from multiple sites and in larger numbers, 2) measure prevalence of child behavior problems in a clinically meaningful way across pre- and midschool periods, and 3) consider a wide range of potential predictors of outcome, particularly predictors that are potentially modifiable through intervention.
The present study had 3 objectives: 1) to establish the prevalence of clinically significant behavior problems in a sample of preterm LBW infants at 3, 5, and 8 years of age; 2) to determine the stability of clinically significant behavior problems between 3 and 8 years of age; and 3) to identify potentially modifiable factors during pregnancy or the neonatal period that might be altered to reduce the risk for behavior problems. To address these objectives, we relied on data collected for those who participated in the Infant Health and Development Program (IHDP).10–12
Study Sample and Design
The IHDP was a multisite, randomized, controlled trial of the effect of early childhood support and education from birth to age 3 years corrected for duration of gestation on the social, emotional, and cognitive outcomes in LBW premature infants. Subject selection and method of the IHDP is described fully elsewhere.13 Briefly, infants were eligible when they were born preterm (<37 weeks' gestation) and weighed <2500 g at birth. A total of 985 participating mother–child dyads were recruited at 8 sites in the United States between January 7, 1985, and October 9, 1985. One third of the group were randomly assigned to the intervention and compared with the remainder, who received routine follow-up only. The study design included stratification by birth weight: one third of the sample weighed between 2001 and 2500 g, and two thirds weighed 2000 g or less. The intervention group was provided with both home-visiting services and center-based child care from birth to age 3, corrected for gestational age.
Baseline data were collected by interview at entry to the trial, maternal self-reported mental health was measured using the 12-item General Health Questionnaire (GHQ)14 at 40 weeks, gestation and parent-reported behavior problems in the child were assessed at age 3 using the Child Behavior Checklist for Ages 2 to 3 (CBCL/2–3)15 and at ages 5 and 8 using the CBCL/4–18.16
The CBCL/2–3 lists various types of behavior and emotional problems that occur in children aged 2 and 3 years. Each of 99 problem items is scored 0 = not true, 1 = somewhat or sometimes true, and 2 = very true or often true, based on the preceding 2 months. Scores on each item are summed to give a raw total problems score. Higher scores indicate more behavior problems. A suggested cutoff point dichotomizes “clinical” and “normal” groups and has been shown to discriminate efficiently between children who are referred to mental health services and nonreferred children.15
The CBCL/4–18 is a widely used, parent-completed checklist that identifies various types of behavior and emotional problems that occur in children aged 4 and above. Each of 118 problem items is scored 0 = not true, 1 = somewhat or sometimes true, and 2 = very true or often true, based on the preceding 6 months. Scores on each item are summed to give a raw total problems score. Higher scores indicate more behavior problems. Gender-specific cutoff points dichotomize clinical and normal groups and have been shown to discriminate efficiently between children who are referred to mental health services and nonreferred children.16 Fifty-four of the problem items are common to both the CBCL/2–3 and the CBCL/4–18.
The 12-item GHQ has been used to screen for short-term changes in mental health in a number of large-scale, community-based surveys as well as in clinical studies. The respondent assesses her present state compared with her usual state by answering 12 items on a 4-point scale: better than usual, same as usual, worse than usual, much worse than usual. We use a cutoff point of 12 to dichotomize a group with significant psychological distress from a normal group.14
Baseline measures of interest in predicting child behavior problems were identified. These included marital status and living arrangement, trimester of first prenatal care, alcohol consumption during pregnancy, cigarette consumption during pregnancy, gestational age at birth as defined using the Ballard method,17 and the child's birth order. In keeping with previous work13 on the IHDP, we also included the more established risk factors of child gender, race/ethnicity, birth weight, neonatal health,18 maternal age at birth, and level of education. Treatment group assignment was included to adjust for any potential treatment effect. The categorizations of these variables are summarized in Table 1.
All of the categorical variables were treated as nominal variables and dummy coded. The unadjusted prevalence of being in the clinical group at each age (3, 5, and 8) was determined. The prevalence stratified by birth weight and each of the categorical variables was then derived. Then the percentages that remained in the clinical and normal groups at ages 5 and 8 were computed. Next, using a logit model, the odds ratios (ORs) of being in the clinical group at ages 5 and 8, conditional on being in the clinical group at ages 3 and 5, were determined.
Finally, alternating multiple logistic regression19,20 was used to model the outcome (clinical vs normal group assignment on the CBCL/2–3 and the CBCL/4–18 from ages 3 through 8) as a function of the baseline measures and GHQ. This method is appropriate with repeated observations when the responses at each time point may be correlated. It gives more conservative estimates of confidence intervals (CIs) than ordinary logistic regression. The method uses the data on all 869 subjects (not just data on the 804 with complete data). The longitudinal element of time is incorporated into this model using age of testing (3, 5, and 8 years) as an additional covariate dummy coded to make age 3 the baseline. SAS 8.221 was used for the analysis. The full main effects logistic regression model included 2 sets of predictor variables. The first set was site of randomization, gender of child, race/ethnicity of child, maternal education, maternal age, birth weight, neonatal health index, and treatment group assignment. These variables were kept in a reduced model to be consistent with previous analyses in this sample. The second set of predictor variables was marital status, maternal psychological distress at 40 weeks' gestation, living arrangement, trimester of first prenatal care, alcohol consumption during pregnancy, cigarette consumption during pregnancy, gestational age at birth, birth order, and age at testing. If these variables did not contribute significantly to the model, then they could be dropped.
The IHDP was approved by both local and national Human Subjects Committees. This analysis has been approved by the Harvard School of Public Health Human Subjects Committee.
Description of Sample
Of the 985 subjects who were randomly allocated to intervention or follow-up only, data were complete at 3 years of age on 869. These form the population for this study. Complete outcome data at all 3 time points were present on 804 (92.5%) of the 869 children.
Comparison of the 869 subjects with 116 excluded because of incomplete data showed no significant differences in gender (χ2 <0.01, P = .96, df = 1), race/ethnicity (χ2 = 0.44, P = .80, df = 2), site (χ2 = 7.37, P = .39, df = 7), treatment group assignment (χ2 < 0.01, P = .94, df = 1), marital status (χ2 = 1.78, P = .41, df = 2), maternal age (t test P = .46), birth weight (t test P = .49), or Neonatal Health Index (t test P = .09). However, there was a significant difference for maternal education (χ2 = 4.58, P = .03, df = 1): 49.1% of mothers with missing data did not complete high school compared with 38.8% of those who did.
The characteristics of the sample for the categorical variables are shown in Table 1. The mean birth weight for the sample was 1799.1 g (standard deviation: 453 g), and the mean maternal age at time of birth was 24.8 years (standard deviation: 6 years).
Table 2 shows estimates of point prevalence at each age. Consistent with the way the instrument is normed and scored, the results show no significant variation in point prevalence over time: 20.9% (95% CI: 18.2–23.6) at age 3, 20.0% (95% CI: 17.2–22.7) at age 5, and 19.9% (95% CI: 17.2–22.7) at age 8.
Table 3 shows prevalence at each age of testing stratified by the 2 birth weight strata and by the categorical variables. The prevalence did not seem to be influenced by birth weight at any age. At age 3, prevalence was significantly increased in children whose mother did not complete high school compared with those whose mother did and in children whose mother either did not attend at all for prenatal care or did not attend until the third trimester. These differences were no longer apparent by ages 5 and 8. Children of mothers with psychological distress at 40 weeks' gestation and children of lone mothers had a consistently higher prevalence between ages 3 and 8 of clinically significant behavior problems, but this difference was statistically significant only at age 5. Similarly, children of mothers who smoked during pregnancy showed a consistently increased prevalence of clinically significant behavior problems, but this difference was statistically significant only at age 8. Children of Hispanic mothers had a consistently higher prevalence rate than mothers with white or black ethnicity, and children of divorced single or widowed mothers also showed higher prevalence rates.
Figure 1 shows both numbers and proportions of the sample in the clinical and normal groups at ages 3, 5, and 8 years. A total of 311 (35.8%) of the 869 subjects were in the clinical group on at least 1 occasion. Of the 182 children in the clinical group at age 3, 50.0% remained in the clinical group at age 5, and 51.7% of the group who were in the clinical group at both ages 3 and 5 remained in the clinical group at age 8. However, these 47 subjects, who were in the clinical group at all ages (3, 5, and 8), made up only 5.4% of the entire sample of 869.
If a child was in the clinical group at age 3, then that child was 9.27 (95% CI: 6.29–13.67) times more likely to be in the clinical group at age 5 than a child who was in the normal group at age 3 and 4.17 (95% CI: 2.87–6.06) times more likely at age 8 to be in the clinical group. If a child was in the clinical group at age 5, then that child was 8.91 (95% CI: 6.00–13.24) times more likely to be in the clinical group at age 8 than a child who was in the normal group at age 5. Finally, if a child was in the clinical group at both age 3 and age 5, then that child was 6.48 (95% CI: 4.06–10.35) times more likely to be in the clinical group at age 8 than those in the clinical group at 1 time point and in the normal group at another or in the normal group at both time points.
After fitting a full main effects model, age at testing, marital status, living situation, trimester of first prenatal care, birth order, gestational age, and alcohol during pregnancy were shown to have no significant effects; therefore, these variables were removed in a reduced model. The reduced model included the following variables: site of randomization, gender of child, race/ethnicity of child, maternal education, maternal age, birth weight, neonatal health index, treatment group assignment, maternal psychological distress, and cigarette consumption during pregnancy.
The effects of birth weight and neonatal health were not significant in the model. Sociodemographic variables, effects of treatment assignment, and significant predictors are presented as ORs. These results show that after adjustment, the average odds of being in the clinical as opposed to the normal group at ages 3, 5, and 8 were significantly and independently predicted by 4 baseline characteristics: maternal psychological distress at 40 weeks, maternal cigarette smoking during pregnancy, Hispanic ethnicity, and maternal age. Controlling for other variables, the odds of a child's being in the clinical as opposed to the normal group were 59% higher in children whose mother had psychological distress at 40 weeks, 57% higher in children whose mother smoked during pregnancy, and doubled in Hispanic children compared with children in the white or other group. The odds were reduced by ∼3% for each year of increase in maternal age. This means that the risk was nearly halved in a mother aged 33 at birth compared with a mother aged 16 at birth. The results of fitting the reduced model are shown in Table 4.
Our first objective was to establish the prevalence of clinically significant behavior problems. We found that the prevalence of behavior problems in this sample of preterm LBW children remained at ∼20% at 3, 5, and 8 years of age.
The prevalence of clinically significant behavior problems using the dichotomized total problems scores on the CBCL/2–3 and CBCL/4–18 was twice the 10% prevalence expected in normative population samples, suggesting that, on average, prematurity and LBW double the risk of behavior problems. This is supported by another study showing that LBW children were twice as likely as normal birth weight children to have clinically significant behavior problem scores.22 Additional support comes from a random effects meta analysis of 6 case-control studies.1 This study reported a relative risk of 2.64 for the development of attention-deficit/hyperactivity disorder in LBW children compared with normal birth weight control subjects. The higher prevalence of clinically significant behavior problems in children of smokers, Hispanic mothers, and those with psychological distress at 40 weeks' gestational age was reflected in the multivariate model.
Our second objective was to establish the stability of clinically significant behavior problems. We found that once these problems had developed, they showed moderate stability over time. This stability is greater between time periods 2 or 3 years apart than over the interval of 3 to 8 years. The stability in the short term is consistent with the literature.5
Our third objective was to identify which sociodemographic and obstetric characteristics of the mother and the child at the time of the birth were independent predictors of clinically significant behavior problems at ages 3, 5, and 8. Maternal smoking during pregnancy, maternal age at time of birth, and being Hispanic as opposed to white race/ethnicity were significant predictors of behavior problems at 3, 5, and 8. These results for significant predictors are consistent with the existing literature. Longer established predictors of child behavior problems such as gender, birth weight, gestational age, and neonatal health were not significant predictors of outcome when included in the multivariate model. These predictors might well have been significant in a different sample of predominantly very low or extremely low birth weight infants.
The predictive value of maternal age independent of socioeconomic status is in keeping with previous reports.23 The finding that Hispanic children were at double the risk of clinically significant behavior problems is consistent with previous work24 showing higher risk in Puerto Rican children for behavioral and developmental disorders, but very little research has been done in this area for preterm LBW children.25
We identified 2 modifiable factors: maternal psychological distress and maternal cigarette smoking during pregnancy. The marked effects of maternal psychological distress are in keeping with the literature suggesting that maternal distress modifies the relationship between mother and infant at a number of levels and in a reciprocal manner to increase the risk of behavior problems and other psychiatric disorders.26 The persistence of this effect, arising from maternal psychological distress at 40 weeks' gestation, has not to our knowledge been demonstrated before in preterm LBW children. It may reflect long-term impairment of the mother–child relationship arising from maternal psychological distress at 40 weeks' gestation. However, it seems more likely that the maternal psychological distress at 40 weeks' gestation represents a snapshot of recurrent or chronic maternal psychological distress or even unmeasured or residual confounders reflecting underlying maternal vulnerability. The association of smoking during pregnancy and increased risk of behavior problems also, to our knowledge, has not been shown before in a preterm LBW cohort. Smoking and postnatal depression are frequently found together and may represent part of a pattern of risk factors found in association with social deprivation, which in turn is also associated with increased risk of child behavior problems. However, this study controlled for maternal education as a proxy for social deprivation, and the smoking effect was independent of the effect of psychological distress. Therefore, the association of child behavior problems with smoking may represent a real teratogenic effect as has been shown in other recent work27 on term infants. If so, then this finding adds to the increasing body of literature showing an association between maternal smoking during pregnancy and later behavioral problems in children after controlling for birth weight.28
There are a number of limitations in this study. First, the cohort was born 17 years ago, when the mortality of very low birth weight infants was considerably higher than it is now.29 A second limitation is that cigarette smoking during pregnancy was assessed by maternal recall with no independent validation performed (eg, measuring cotinine levels). However, this would not have been possible given that recruitment was not until after the birth. Mothers may not have reported smoking if they did, but it seems unlikely that they would report smoking if they did not smoke. Failure to detect maternal smoking would lead to a misclassification error, which might have diluted the relationship of smoking with behavior, but we still obtained a relationship. A third limitation is that outcome data were available only on 869 of the 985 involved in the IHDP. Those mothers for whom data were not available were less likely to have completed high school and therefore may have been at increased risk for child behavioral problems. This means that the prevalence rates found in the study may be underestimates. Another limitation of maternal report on child behavior problems is that depressed mothers may be more likely to rate their children as more disturbed. This may have biased the analysis.
The study has a number of implications for clinical practice. First, it suggests that greater attention be paid to maternal smoking during pregnancy. Smoking cessation programs are worthwhile in any event: a recent systematic review30 concluded that smoking cessation programs in pregnancy reduce smoking, LBW, and preterm birth. Our results suggest that additional benefits may accrue. Also, assessment of maternal mood with widely used screening questionnaires such as the Edinburgh Postnatal Depression Scale31 should be performed while the infant is in the neonatal intensive care unit. Appropriate intervention could reduce maternal morbidity as well as reduce the risk of later child behavior problems. We suggest interpreting the increased risk for behavior problems in Hispanic children compared with the white (and black) population with caution. The numbers involved were small, and Hispanic mothers may have had a lower threshold for reporting behavior problems than the other ethnic groups in this study.
In summary, in a sample of 869 preterm LBW children who were recruited at 8 US sites and followed up to age 8, we found a prevalence of clinically significant behavior problems approximately double what might be expected in the general population. We found that these problems showed moderate stability once they developed. Finally, we have identified 2 potentially modifiable factors for which interventions might reduce the risk of behavior problems: cigarette smoking in pregnancy and maternal psychological distress at 40 weeks' gestation.
Recent follow-up studies of LBW children in adolescence and young adulthood have found catch-up growth, a reduction in acute health problems,32 and a lower rate of drug and alcohol misuse4 compared with control subjects. These findings are not fully understood and suggest the need for additional work. Data collection on phase 4 of the IHDP is currently in progress (the participating children are now 18 years of age) and hopefully will shed some light on these interesting findings.
This study was partially supported by funds from the Robert Wood Johnson Foundation (I.D. 039543) and the National Institutes of Health (MH01880). Dr Gray is supported by The Commonwealth Fund, a New York City–based private independent foundation. The views presented here are those of the authors and not necessarily those of The Commonwealth Fund and its director, officers, or staff.
Related work from this study has been presented in abstract form at the Pediatric Academic Societies' Annual Meeting; May 3–6, 2003; Seattle, WA (Pediatr Res. 2003;53:351A, 460A).
We thank Dr S. Buka for comments on an earlier draft and the 2 anonymous reviewers for helpful comments.
- ↵Buka SL, Lipsett LP, Tsuang MT. Emotional and behavioral development of low birthweight infants. In: Friedman SL, Sigman MD, eds. The Psychological Development of Low Birthweight Children. Norwood, NJ: Ablex Publishing; 1992:187–214
- ↵Miceli PJ, Goeke-Morey MC, Whitman TL, Kolberg KS, Miller-Loncar C, White RD. Brief report: birth status, medical complications, and social environment: individual differences in development of preterm, very low birth weight infants. J Pediatr Psychol.2000;25 :353– 358
- ↵McCarton C. Behavioral outcomes in low birth weight infants. Pediatrics.1998;102(suppl E) :1293– 1297
- ↵Goldberg D, Williams P. A User's Guide to the General Health Questionnaire. Windsor, United Kingdom: NFER-Nelson; 1988
- ↵Achenbach T. Manual for the Child Behavior Checklist/2-3 and 1992 Profile. Burlington, VT: University of Vermont Department of Psychiatry; 1992
- ↵Achenbach T. Manual for the Child Behaviour Checklist/4-18 and 1991 Profile. Burlington, VT: University of Vermont, Department of Psychiatry; 1991
- ↵Ballard JL, Novak KK, Driver M. A simplified score for assessment of fetal maturation of newly born infants. J Pediatr.1979;95(suppl) :769– 774
- ↵Scott DT, Bauer CR, Kraemer HC, Tyson J. The Neonatal Health Index. In: Gross RT, Spiker D, Haynes CW, eds. Helping Low Birth Weight Premature Babies. Palo Alto, CA: Stanford University Press; 1997:341–357
- ↵Carey V, Zeger SL, Diggle PJ. Modeling multivariate binary data with alternating logistic regression. Biometrika.1991;80 :517– 526
- ↵Lipsitz SR, Laird NM, Harrington DP. Generalized estimating equations for correlated binary data: using the odds ratio as a measure of association. Biometrika.1991;78 :153– 160
- ↵SAS Institute. SAS/STAT User's guide, Version 8.0. Cary, NC: SAS Institute Inc; 2000
- ↵McCormick MC, Workman-Daniels K, Brooks-Gunn J. The behavioral and emotional well-being of school-age children with different birth weights. Pediatrics.1996;97 :18– 25
- ↵Campbell S. Behavior Problems in Preschool Children: Clinical and Developmental Issues. 2nd ed. New York, NY: The Guilford Press; 2002
- ↵Law KL, Stroud LR, LaGasse LL, Niaura R, Liu J, Lester BM. Smoking during pregnancy and newborn neurobehavior. Pediatrics.2003;111(suppl) :1318– 1323
- ↵Lemons JA et al. Very low birth weight outcomes of the National Institute of Child health and human development neonatal research network, January 1995 through December 1996. NICHD Neonatal Research Network. Pediatrics.2001;107(1) . Available at: www.pediatrics.org/cgi/content/full/107/1/e1
- ↵Lumley J, Oliver S, Waters E. Interventions for promoting smoking cessation during pregnancy. Cochrane Database Syst Rev.2000;(2):CD001055
- ↵Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression. Development of the 10-item Edinburgh Postnatal Depression Scale. Br J Psychiatry.1987;150 :782– 786
- ↵Saigal S, Stoskopf BL, Streiner DL, Burrows E. Physical growth and current health status of infants who were of extremely low birth weight and controls at adolescence. Pediatrics.2001;108 :407– 415
- Copyright © 2004 by the American Academy of Pediatrics