Functional Assessment of a Multicenter Very Low-Birth-Weight Cohort at Age 5 Years

Background  Very low-birth-weight newborns (birth weight ≤1500 g) experience serious neonatal complications, but long-term outcomes are not completely known. Most studies reflect an era of neonatal care that was fundamentally different from the present.

Objectives  To compare the functional level of very low-birth-weight children before and after surfactant introduction and to relate functional level to clinical and socioeconomic factors.

Design  Inception cohort followed up from birth to an average age of 5 years.

Setting  Six regional neonatal intensive care units in a contiguous geographic area.

Participants  Four hundred twenty-five very low-birth-weight children, born between August 1, 1988, and June 30, 1991, of 438 located among 626 whose parents provided follow-up information before neonatal intensive care unit discharge.

Interventions  None.

Main Outcome Measures  Diagnosis of cerebral palsy and standardized scores for self-care, mobility, and social function from the Pediatric Evaluation of Disability Inventory.

Results  Cerebral palsy was present in 12.6% of the children, with no change after surfactant introduction. Intraventricular hemorrhage (odds ratio, 2.3 per grade; 95% confidence interval, 1.8-2.8) and bronchopulmonary dysplasia (odds ratio, 2.3; 95% confidence interval, 1.2-4.6) were independently predictive of cerebral palsy and of functional outcome. For self-care, mobility, and social function, 11.7%, 29.5%, and 10.7% of the children, respectively, scored at least 2 SDs below the normative means. Social function was 0.25 to 0.50 normative SDs lower after general surfactant availability than before general surfactant availability.

Conclusions  While there was no increase in major disability after surfactant introduction, there may have been a decrease in social function associated with the lower neonatal mortality. Most very low-birth-weight children functioned within the normal range in everyday tasks. Several predictors of outcome were identified.

VERY low-birth-weight (VLBW) newborns (birth weight ≤1500 g) often experience serious neonatal complications such as bronchopulmonary dysplasia (BPD) and intraventricular hemorrhage (IVH). However, the long-term consequences of VLBW and its complications are not completely understood. In 1995, a review by Hack et al¹(p191) stated "At present there is a lack of solid information about the long term follow-up of low birth weight children in the United States."

Most follow-up information from the United States pertains to VLBW children born when neonatal care was fundamentally different from the present. Studies^2-5 have demonstrated a change in survival of VLBW newborns after the availability of surfactant and with increasing use of antenatal and postnatal corticosteroid therapy. Studies to address the childhood consequences of the greater survival have either been small,^6,7 limited to clinical trial populations,^8,9 or of short duration.^6,8,10,11 Two studies^12,13 with at least 200 VLBW children born after 1989 found developmental delays up to age 3 years. However, to our knowledge, no recent study has assessed functional level close to the time of kindergarten enrollment.

One study¹² of 122 children with BPD and 84 control subjects found that BPD, neonatal neurologic status, and socioeconomic level were related to Bayley Motor and Mental scale scores. Another regional study¹³ of 367 VLBW children at a median postterm age of 15 months found delays in development compared with higher-birth-weight children but considered a limited set of neonatal risk factors within the VLBW group. Thus, to our knowledge, there has not yet been a comprehensive assessment of how recent treatment trends, neonatal complications, and socioeconomic level relate to childhood outcome in an unselected, regional cohort. Interest is turning to how the children function in everyday tasks and social interactions,^14-16 while most previous studies addressed the incidence and prevalence of developmental disorders and delays.

By following up our cohort born between August 1, 1988, and June 30, 1991, to an average age of 5 years, we assess the long-term functional consequences of neonatal complications and of increased survival. Our cohort is representative of general neonatal intensive care unit (NICU) care in a geographic region with 6 perinatal centers.

From August 1, 1988, through June 30, 1991, 1024 VLBW neonates were admitted before 24 hours of age to 6 regional NICUs covering a contiguous geographic area of Wisconsin and Iowa. The births represent 3 periods with different surfactant availability: 1, sporadic randomized clinical trials (August 1, 1988, to July 31, 1989); 2, investigational new drug (August 1, 1989, to July 31, 1990); and 3, general availability (August 1, 1990, to June 30, 1991). Antenatal corticosteroid use also increased across the 3 periods in our region.² State death certificate and follow-up clinic record searches were performed to identify children who had died after hospital discharge. Of the 803 children who survived to age 5 years, the parents of 626 agreed to potential follow-up and provided recontact information before NICU discharge. Those agreeing to recontact made up the target group for follow-up and were representative of the group as a whole on a wide range of characteristics, including birth weight, Apgar scores, and fraction of inspired oxygen at 24 hours (Table 1). A group of 438 was located close to age 5 years, with 13 refusing participation. The 425 children evaluated were similar to the target group for clinical and socioeconomic characteristics at baseline (Table 1). The follow-up sample had fewer African Americans and other races, but the number of nonwhites was too low in all 3 samples to draw conclusions regarding race or to influence overall results.

Designated neonatal nurses recorded the neonate's race, sex, and medical record information on a standardized form. Respiratory records were photocopied, and all radiographs from 25 to 35 days of life were obtained. Clinical predictors in 3 areas are considered herein: (1) early predictors (characteristics and treatments during the first 3 days of life), (2) later neonatal predictors (treatments and diagnoses up to discharge from the NICU), and (3) nonspecific clinical variables (hospital of birth and days of NICU stay). Early predictors are multiple birth, birth weight, gestational age, Apgar scores, antenatal corticosteroid therapy, surfactant therapy, and supplemental oxygen and ventilation at 24 hours of age. Respiratory disease in the first 3 days of life was scored 0 to 100 by increasing severity.¹⁷ Later neonatal predictors are grade of IVH, presence of necrotizing enterocolitis, retinopathy of prematurity (ROP), and BPD. Several criteria for BPD or neonatal persistent lung disease were considered.¹⁸ Radiographs were interpreted for "radiographic abnormality" indicative of BPD according to previously established procedures.¹⁹ Use of oxygen at 36 weeks' postconceptional age was recorded. Finally, due to the increasing use of postnatal corticosteroid therapy, postnatal corticosteroid therapy, mechanical ventilation, or both on day 30 were considered indicators of "severe BPD" at 1 month. Pairwise agreement between the 3 criteria ranged from 63% to 80%.

Data collectors were pediatric nurses trained at workshops where they reviewed written instructions on how to administer the assessment and listened to actual interviews. Extensive discussion focused on the range of responses, and data collectors conducted practice interviews. Telephone interviews were conducted as close as possible to age 5 years (75.8% of the children were aged 5 years, 11.5% were younger, and 12.7% were older). Physician diagnosis of cerebral palsy (CP) was ascertained at the parent interview and confirmed by clinic record abstracting. Blindness and use of corrective lenses were similarly ascertained. Functional assessment used the Pediatric Evaluation of Disability Inventory (PEDI),²⁰ which includes self-care, mobility, and social function domains (Table 2). This instrument is age standardized from 6 months to 7½ years to normative scores with mean (±SD) 50 ± 10 for each domain. The parents of 4 children refused the PEDI portion of the interview.

Socioeconomic information included maternal educational level (coded as 1 [less than high school] through 6 [postgraduate education completed]), caretakers' occupations, and family composition (eg, number of parents in the household). Socioeconomic levels of occupations were determined by the system of Stevens and Cho.²¹ This system assigns numerical socioeconomic levels to occupations from 14 to 90 (highest). The resulting mean ± SD for the 1980 labor force is 36.4 ± 18.9. In our study, the highest of the child's caretakers' occupational socioeconomic levels was used. ZIP codes at birth and follow-up were linked to 1990 census data to obtain neighborhood median household and per capita incomes. All socioeconomic information was combined into 2 different socioeconomic scores by principal components analysis.²²

The difference between the age of a child at the time of the interview and the average interview age within the study year is greater when the parents were more difficult to schedule (due to initial nonreturn of consent forms, not being reachable by telephone, or being initial no-shows at scheduled interview times). The percentile of this difference among all interviews in a year provided a parent "reachability" index.

A total of 42 randomly selected study parents were interviewed with a second interviewer listening and scoring. Duplicate scores correlated at 0.98, 0.97, and 0.92 for self-care, mobility, and social function, respectively. To monitor the consistency of interviews across calendar time and child age, parents of 164 healthy full-term children volunteered to be interviewed once each, spaced across the period. The only potential inconsistency detected was a yearly decrease in the social function domain of 1.0 point (P = .09). Analyses of the social domain were, therefore, performed with and without correction for this potential drift, and differences in results were noted where relevant.

All analyses were performed using SAS statistical software.²² Descriptive statistics were obtained on the 3 functional domains and compared with normative distribution parameters by t tests for means and χ² tests for variances.²³ Potential predictors were examined by computing subgroup differences and partial Pearson correlations, adjusted for interviewer indicator variables.²⁴

Predictors for CP were investigated by logistic regression analysis. Among children without CP, ordinary linear regression was used for each of the 3 functional domains. Predictors were sequentially added and retained when significant (P<.05 for at least 1 domain) in the following order: (1) the 9 interviewers (indicator variables); (2) the 2 socioeconomic scores and sex; (3) early predictors; (4) later neonatal predictors; (5) period of birth (indicator variables); (6) all interaction and quadratic terms of variables in item 2 to item 4 (P<.01 required) and interactions with item 5; (7) parent reachability; and (8) nonspecific clinical variables, where days spent in the NICU was log transformed to reduce the influence of a few long stays. Model fit was examined graphically and by influence statistics.

To examine participation bias, results were extrapolated to the whole target group by models including the parent reachability index as main and interaction effects. These analyses showed that functional levels may be slightly lower among nonresponders than among responders but that regression coefficients were unlikely to be affected by participation bias.

Table 1 presents descriptive statistics on all key variables for the children who were followed up.

Table 3 shows descriptive statistics for the PEDI self-care, mobility, and social domains. The means were significantly lower than the normative 50 (P<.001). The SDs of 14, 17, and 12 were significantly (all P<.001) higher than the normative 10.

Cerebral palsy was present in 12.6% of the children, with no differences between the 3 birth periods. Table 3 shows descriptive statistics for the 3 PEDI domains by CP status. All means in the table were below the norm (P<.001). Four children with CP (7%) were blind in both eyes, and 1 (2%) was blind in 1 eye. Blind children with CP had mean scores of 15, 10, and 22 in the 3 domains, respectively.

Gestational age (odds ratio, 0.82 per week; 95% confidence interval, 0.73-0.93) and the baseline respiratory severity score (odds ratio, 1.2 per 10-point increase; 95% confidence interval, 1.0-1.3) were independently predictive of CP. However, their significance vanished after IVH grade (odds ratio, 2.3 per grade; 95% confidence interval, 1.8-2.8) and severe BPD or radiographic evidence of BPD (odds ratio, 2.3; 95% confidence interval, 1.2-4.6 for both) were taken into account. No other variables were significantly predictive of CP.

Table 4 shows how 2 socioeconomic scores were constructed. The first score (overall socioeconomic level) is essentially an average of all socioeconomic information available on the child. The second score (family status) is essentially the difference between the family's individual socioeconomic level and that of the neighborhood.

The overall socioeconomic level score correlated at 0.12 (P = .02) with self-care and at 0.24 (P<.001) with social function. The family status score correlated with the self-care, mobility, and social function domains at 0.14 (P = .01), 0.15 (P = .004), and 0.11 (P = .04), respectively.

Among early clinical predictors, birth weight correlated the strongest with the functional domains at 0.16 (P = .003), 0.24 (P<.001), and 0.16 (P = .003) (Figure 1). Multiple births had mobility 4.1 points higher (P = .01) than singleton births. After surfactant became available, children receiving it had mobility 4.8 points lower (P = .01) and social function 2.6 points lower (P = .06) than those who did not.

Intraventricular hemorrhage grade correlated with all 3 domains at −0.16 (P = .002), −0.20 (P<.001), and −0.18 (P<.001) (Figure 2). Presence of severe BPD was associated with scores 5.7 (P<.001), 6.6 (P<.001), and 3.7 (P = .007) points lower in the 3 domains. Presence of radiographic abnormalities at 25 to 35 days of life and use of oxygen at 36 weeks' postconceptional age had significant, but weaker, associations with functional domains. Children with ROP had mobility 7.3 (P<.001) points lower than those without. One child without CP was blind in both eyes, and another was blind in 1 eye. Children without (7.7%) and those with (20.4%) ROP wore corrective lenses. There was an indication (P = .12) that the lower mobility in those with ROP was most pronounced among those wearing corrective lenses.

The logarithm of the number of days in the NICU was correlated at −0.22, −0.23, and −0.21 with the 3 functional domains (all P<.001).

Table 5 shows effect sizes in 2 multiple regression models for variables that remain independently predictive for at least 1 functional domain. Model 2 differs from model 1 by including parent reachability and nonspecific medical factors.

In model 1, early predictors (except multiple birth) and other BPD criteria were not significant in the model when IVH grade and severe BPD at 30 days (as indicated by ventilation, postnatal corticosteroids, or both) were taken into account. Excluding ROP from the models would have made the effects of IVH grade and severe BPD larger. The higher mobility among multiple births was not explained by higher gestational age among the multiple births. The effects of sociodemographic and medical factors were completely independent.

The higher self-care among girls was most pronounced at the highest socioeconomic levels. For the self-care and social domains, the socioeconomic effect was weaker among multiple births (P<.01 for interactions).

In model 2, for the social domain, IVH grade and severe BPD lost statistical significance as length of hospitalization was entered. Lower parent reachability was independently associated with lower mobility (P = .04) and social function (P<.01). After taking into account all other variables, differences between hospitals of birth were independently predictive only for mobility (P = .001).

Social function was lowest in the last period (Table 5). This difference is explained neither by nonspecific medical factors nor by early predictors. Adjustment for a possible shift in interview technique decreased the difference to −3.6 (P = .02) and −2.6 (P = .08) in the 2 models. The difference between periods was almost twice as large among children exposed to antenatal corticosteroids (P = .04 for the interaction). It was also greater among those with higher family status (P = .03 for the interaction). Although there was no overall difference between periods in mobility, there was a difference between periods 3 and 1 of −11.5 (P<.001) among those with severe BPD (P<.001 for the interaction).

Table 6 shows sequential increments in the percentage of variability explained with the addition of variable groups to the model. While clinical neonatal variables are predictive for all 3 domains, they are most important for mobility. Socioeconomic variables are most important for social function. These results pertain to children without CP. Including children with CP in the correlation analysis dramatically increases the incremental percentage variance explained by clinical complications to 17%, 19%, and 13%, for the 3 domains.

To our knowledge, this study of NICU survivors who weighed 1500 g or less at birth in a contiguous midwestern geographic area is unique in the United States in that a large, recently born, multicenter VLBW cohort was followed up to an average age of 5 years. In addition, our study is among the few that evaluated outcomes, and the only one that evaluated functional status of VLBW children across the period when surfactant therapy was introduced.

The incidence of CP in our cohort was 12.6%, while the review by Hack et al¹ would lead us to expect a 16% to 18% incidence of CP for our birth weight distribution.⁷ Of all VLBW children, 29.5% fell more than 2 SDs below the norm for mobility function. This includes almost all of the children with CP. However, the proportion with low mobility function is high even among children without CP. For the self-care and social domains, 11.7% and 10.7% fell more than 2 SDs below the normative mean among all VLBW children, respectively, while 2.5% would be expected from the normative distribution. Although the relation between functional measures and developmental tests is unknown, the percentages classified as below normal by the PEDI resemble those of studies using other tests. For example, 13% of VLBW children at age 3 years fell at least 2 SDs below the norm for Bayley Mental scale scores,¹² and a similar percentage abnormal was found at age 4 years in McCarthy scales among 55 VLBW children born between 1986 and 1991.⁷

One of our main goals was to detect any changes in function following the introduction of surfactant therapy. We found a decline in social function following the general availability of surfactant. Interestingly, we found that the lower social function in the postsurfactant era was most pronounced in subgroups of children whose survival may have been due to the new therapies, such as those exposed to antenatal corticosteroids. Previous analyses² indicated that antenatal corticosteroid therapy was the major contributor to the improvement in survival. It has further been reported that the combination of antenatal corticosteroid therapy with surfactant therapy is especially effective.^25-27 In the last period, we found surfactant therapy most intensely applied.²

A comprehensive review¹⁶ did not report worse developmental outcome following the greater survival ensuing from surfactant therapy or its availability. Clinical trials^8,28 of surfactant treatment under select, closely controlled conditions have shown encouraging results up to age 24 months. A comparison of 114 children who weighed less than 750 g at birth in the presurfactant and postsurfactant eras at age 20 months showed no notable differences in Bayley Scales scores or in the prevalence of neurologic deficits.¹⁰ However, among VLBW neonates ventilated on day 21 of life, the same research group found a nonsignificant decrease of about 0.20 SD in Bayley Scales scores between an early (1986-1988) and late (1988-1990) period.¹¹ Our significant difference in social function between 1988 and 1989 and 1990 and 1991 is of similar magnitude after adjustment for all known and suspected influences on the score.

We could not test another important hypothesis that a change in the pattern of admissions to tertiary NICUs may have played a role in generating different outcomes in different periods, as implied in another study.²⁹ The previous comparison of our study cohort with regional statistics implied that NICU admissions to tertiary centers may have decreased in the last period as secondary units acquired ventilation facilities.²

Our large sample size allowed comprehensive evaluation of the risk factors for lower functional outcome. We found that IVH grade and severe BPD were independently associated with CP. This result was also found in a large clinical trial population.³⁰ Among children without CP, several moderately predictive sociodemographic and medical conditions combined to explain 15% to 20% of the variability in functional level. This percentage is comparable with the 22% to 23% found for Bayley Mental and Motor scale outcomes, using a similar but not identical set of predictors.¹² However, the latter percentages may have been increased by the sampling technique that produced a greater spread in socioeconomic levels. Similar to Singer et al,¹² we also found that prematurity acts through its neonatal complications of IVH and BPD. There was an indication of variation in outcome between hospitals, which was most pronounced for mobility. Our hospitals differed on a wide range of therapies and characteristics of neonates admitted, making it almost impossible to identify specific causes of the variation.

Our composite socioeconomic scores were designed to capture most of the variation across the population in individual socioeconomic variables, and together with sex explain 2.7% to 7.5% of the variation in functional domains. The scores we constructed were stronger predictors than mother's educational level alone. Correlations between mother's educational level and McCarthy scale scores reported by another study⁷ were stronger than our relations. McCarthy scales, being intelligence related, may be more sensitive to this influence. Our group was too racially homogeneous to reliably examine race as a predictor.

The relation of parent reachability to the social domain may also reflect the child's social environment or parental role modeling. Parents who function at a high level socially may be more prone to agree to be interviewed in a timely manner. An alternative explanation of the influence of this variable may be that parents of medically compromised children are less ready to be interviewed. However, as parent reachability had much weaker relations with the other 2 functional domains, we suspect that this latter effect is relatively unimportant. While parent reachability did not confound the relations between social function and other factors, a direct relation between parent and child sociability may be implied.

In summary, our study has shown that, as in the past, most surviving VLBW children function at average levels in everyday tasks, while a minority of VLBW children is severely impaired. We found that functional level among children without CP is independently linked to neonatal complications and to the socioeconomic environment of the child. The percentage with severe impairment remained constant after the introduction of surfactant. However, there appears to have been a decrease in some functional variables. The decrease was not explained by known medical conditions or treatments but may have been a consequence of greater survival.

Editor's Note: I hope this study is only the first installment. It raises as many questions as it answers—like many good studies.—Catherine D. DeAngelis, MD

Accepted for publication June 9, 1999.

This study was supported by grant R01 HL38149 from the National Heart, Lung, and Blood Institute (University of Wisconsin Medical School, Madison) and grant M01 RR03186 from the National Center for Research Resources (University of Wisconsin Medical School), Bethesda, Md.

We thank Carol Razafindrakoto, MS, for linkage to socioeconomic data from the 1990 census and other sources.

Reprints: Mari Palta, PhD, Department of Preventive Medicine, University of Wisconsin, 504 N Walnut St, Madison, WI 53705 (e-mail: palta@cornfield.epi.wisc.edu).

Madison, Wis: Mari Palta, PhD; Mona Sadek-Badawi, MB,BCh; Aggie Albanese; Christina Iyama, MD; Nan Peterson, RN, MS; Marie Weinstein, MD; Laura Ziebarth, RN. Neenah, Wis: Paul Myers, MD; Cindy Wierichs, RN; Pamela Verhagen, RN. Iowa City, Iowa: Gail McGuinness, MD; Irma Kromer. Marshfield, Wis: James Opitz, MD; Joan Filbin, RNC. Green Bay, Wis: David Samuels, MD; Sue Vollmer, RNC, NNP; Lana Reinke, RN; Ruth Rodda, RN.