Printer Friendly

Early identification of developmentaly disabled and at-risk preschool children.

Early Identification of Developmentally Disabled and At-Risk Preschool Children

ABSTRACT: This study examined child-centered data (from birth to 7 years) and familial factors

as possible predictors of disabilities in adolescence. The sample was taken from original

participants in the National Collaborative Perinatal Project in Rhode Island who were also

judged as handicapped after school entry. Results of the current study indicated that parental

traits (i.e., maternal education) are more accurate predictors of adolescent status than the

child's own behavior from birth to 3 years, whereas child-centered skills assessed at 4 and 7 year

of age are better predictors than are familial factors. Overall, data suggest that early

identification models which focus upon developmental delay or adverse medical events from birth to

3 years of age are inadequate in fully identifying children eventually judged to be handicapped.

Screening initiatives must be developed that are multivariate (child and family focused) and

account for differential weights of risk factors over time. * Concern for young children and their families has reached a high level of prominence within the past 2 years. This amplified attention to the early childhood period originates from multiple sources. Research advances in the field of infant behavior have resulted in improved understanding of the learning capacities of newborns, the role of perinatal risk in compromising growth and development, and the enormous impact of life experiences on the psychological development of the infant. Until recently, many parents were led to believe that their infants could not taste fluids, were incapable of sensing odors, and experienced marginal feeling of pain. Studies have indicated that, in fact, neonates are excellent discriminators of taste, and they can detect their mother's own fragrance by 4 days of age (Lipsitt, 1986).

Central to the escalation of interest in young children has been a significant acknowledgement of infants and toddlers with special needs and their families. This attention has both empirical and legislative underpinnings. Many researchers have focused on the outcomes of early intervention programs for infants and toddlers with special needs (Casto & Mastropieri, 1986; Guralnick & Bennett, 1987; White, Mastropieri, & Casto, 1984). This concentrated effort has prompted increasingly sophisticated methodological and statistical analyses (Dunst & Synder, 1986; Dunst, Snyder, & Mankinen, 1987; Shonkoff & Hauser-Cram, 1987). As a result, research has moved from inquiring, "Is it effective?" to an examination of more subtle questions about which factors and their interactions are primary determinants of child and family well being (Coates & Lewis, 1984; Dunst, McWilliam, Trivette, & Galant, 1987).

So persuasive has this body of literature become that the reauthorization of the Education of the Handicapped Act in 1986 (Public Law 99-457, Part H.) included provisions for states to launch major program development initiatives, underwritten by the federal government, so that by 1991, a comprehensive, national early intervention system would result. This legislation granted states considerable latitude in conceptualizing and articulating a system, but did prescribe 14 essential components that must be represented in each statewide plan, including a definition of the population and valid procedures enabling such children and their families to be promptly and accurately identified. Consequently, as states begin to develop operational and functional responses to these stipulations, a significant and complex policy question remains unanswered: To whom should early intervention services be directed?

P.L. 99-457 requires participating states to develop policies ensuring the identification and service of children who experience significant developmental delay or who have established conditions (e.g., chromosomal, neurological, and metabolic disorders) resulting in such delays. In addition, states may elect to identify and serve children at risk of having substantial developmental delays if appropriate early intervention services are not provided. A brief review of investigations that have attempted to design screening models for both groups reveals several significant and relevant findings.

Attempts to identify children at risk for developmental disabilities were initially reported more than a century ago (Little, 1861). Since that time, numerous studies (Denhoff, Hainsworth, & Hainsworth, 1972; Drillien, 1964; Graham, Pennoyer, & Caldwell, 1957; Kawi & Pasamanick, 1958; Parmelee & Michaolis, 1971; Pasamanick & Knobloch, 1961; Schacter & Apgar, 1959; Sigman & Parmelee, 1979; Wiener, 1970) have attempted to isolate prenatal and perinatal traumata that adversely affect developmental pathways.

Studies that examined isolated factors (e.g., anoxia) concluded that models that attempt to predict school age functioning on the basis of single factors (Gottfried, 1973) were plagued by unacceptably high rates of error and misclassification. One of the most compelling findings emerged from the National Collaborative Perinatal Project (Broman, Bien, & Shaughnessy, 1985; Nichols & Chen, 1981), which reported that at 7-year follow-up, significant causal factors reside not so much in the child's medical history as in the ecological context within which a child is reared.

What becomes apparent, therefore, is that a child's development cannot be predicted independent of caretaking experiences (Werner & Smith, 1982). Powerful cross-cultural evidence by Susser, Hauser, & Kelly (1985) also underscores the impact of social environment on mental performance in that epidemiologic surveys in Sweden show the prevalence of severe developmental disabilities to be approximately .3%, comparable to rates observed in the United States. Conversely, the prevalence of mild mental retardation in Sweden is about .4%, 10 times lower than U.S. rates. Clearly, domains other than biologic must be examined to fully account for such remarkable differences.

Sameroff, Seifer, Barocas, Zax, and Greenspan (1987) have offered additional insight into multiple risk models by examining the impact of 10 factors on verbal IQ scores derived at 4 years of age. Specific risk factors included such conditions as maternal anxiety and mental health, stressful life events, family social support, occupation and education levels, and mother/child interactive behaviors. Results indicated that as the number of risk factors increased, intellectual performance decreased, with the difference between the lowest and highest groups being approximately 30 IQ points. Of greatest interest is that no child-centered information was entered into the multiple risk analyses.

Consistent with these findings, the Collaborative Project data set was used to conduct a follow-up study of project participants into adolescence (Kochanek, Kabacoff, & Lipsitt, 1987) by examining the usefulness of information collected before 12 months of age in predicting the presence of a disability reported between 14 and 20 years of age. Major findings indicated that maternal characteristics (e.g., level of educational attainment) were more accurate predictors of adolescent status than child performance data gathered at infancy and 4, 8, and 12 months of age. Moreover, addition of child-centered behaviors into the regression equation did not increase accuracy in identifying students with disabilities beyond that from maternal education alone. This study concluded that models of screening and multi/transdisciplinary team diagnosis founded on child-centered data alone are of suspect validity.

Although this conclusion is clearly apparent for data gathered from birth to 12 months of age, studies have not assessed the differential contribution of child performance data, collected serially over time in conjunction with familial traits, in predicting adolescent outcome. A key, unanswered question is: Which child and parental attributes accurately predict adolescent status, and how do these factors change over time? Of interest to researchers, such questions are also of immense importance to states and interagency coordinating councils (P.L. 99-457, Part H.) in conceptualizing plans for "a comprehensive child find system."

The major purpose of this study, therefore, was to (a) examine whether child-centered data (from birth to age 7) and familial factors can predict the presence of disabilities in adolescence; and, if so, to (b) identify the most significant child or ecological factors collected at different time points to maximize the accuracy of such predictions. Within the present study, child-centered data are defined as measures, factors, and events that have been derived from or directly pertain to the child. Included are standardized test scores, appraisal of developmental competency via clinical judgment, and stresses which the child has endured (e.g., respiratory distress, infectious diseases) or are descriptive of the child's status (e.g., low birthweight).



A group of 268 handicapped adolescent students and a control group of 268 nonhandicapped students matched on sex, race, and age were studied. Sixty-five percent were male, 72% were white, and ages ranged from 14 to 20 years, M = 16.20, SD = 1.38, at the time of follow-up. Students in the handicapped group were participants of the National Collaborative Perinatal Project (NCPP) in Rhode Island and were identified as handicapped by school-based multidisciplinary diagnostic teams during the period 1978-1980. Forty-five students (17%) were judged to have behavior disorders, 47 (17%) were developmentally disabled, and 138 (51%) were learning disabled. The remaining handicapped students were distributed in low numbers across a range of categories including visual and auditory impairments, speech and language disorders, and other health impairments. Subjects in the control group were also participants in NCPP and were not judged to be handicapped. Details of subject selection and diagnosis are described in the following section.


Subjects were Rhode Island participants of the NCPP, a major multi-institutional, interdisciplinary collaborative project developed to study the relationship of prenatal and perinatal conditions and events on subsequent physical, intellectual, and educational capacity. Thirteen university and medical centers, including Brown University and affiliated hospitals, had participated in the national study which involved systematic collection of data through prospective observation and examination of some 60,000 pregnancies with obstetrical intake between January 2, 1958, and December 31, 1965. At the conclusion of the study, 55,908 births were recorded nationally, approximately 4,000 of them in Rhode Island. Data on family health and obstetric, pediatric, and psychologic status were collected at infancy, 4, 8, and 12 months, and 3, 4, and 7 years of age. The NCPP has been described in detail elsewhere (Niswander & Gordon, 1972).

For more than a decade, the Rhode Island Department of Education has maintained an automated handicapped student census system containing information on each student served in any given academic year. Students are included in this system if they have been judged by a multidisciplinary team to be handicapped as defined by P.L. 94-142. Representative data include diagnostic categorization and type and frequency of special education service provided per week.

By merging NCPP data files for Rhode Island with Department of Education census files, 268 students were identified who were original NCPP participants and also later judged to be handicapped by school district. The frequency of special education service provided for this group depended on diagnostic categorization. The median number of hours (hr) per week of specialized services was 22.6 (maximum = 30 hr) for the behavior disorder group, 25.0 for those judged as developmentally disabled, and 4.2 for the learning disabled. Though the reliability and validity of diagnostic categorization by school districts could not be ascertained for these handicapped students, again, all children received comprehensive, multidisciplinary team evaluations and were determined to be in need of special education services. Given the variability in frequency of such services, it would appear that students judged as handicapped in this sample included a broad range of youngsters with mild through severe impairments.

Following identification of these handicapped students, a comparison group of 268 students was selected from the original NCPP files, controlling for age, sex, and race. NCPP source data (from birth to 7 years) were reviewed manually, and any evidence for or suspicion of a handicapping condition resulted in exclusion from the comparison group.


NCPP files provided extensive physical, social, and psychological information on subjects from the prenatal period through 7 years of age. These data were employed as predictors of adolescent criterion group membership (handicapped or nonhandicapped). Given the large number of potential variables to be considered, a three-stage method of analysis was employed.

In the first stage, variables that demonstrated a significant (p [is less than] .01) univariate relationship with the handicap variable were retained for further analysis. Each predictor variable was evaluated for a significant univariate relationship with the predicted binary variable (handicapped vs. nonhandicapped) using a chi-square test if the predictor variable was categorical, a Kruskal-Wallis test if the predictor variable was ordinal, and a t-test if the predictor variable was continuous.

In the second stage, the variables so identified were entered into stepwise logistic regressions employing handicapping condition as the outcome variable. Logistic regression was used to form a weighted combination of predictor variables for predicting the binary outcome variable (Fienberg, 1980). In cases where data do not meet the requirements for discriminant function analysis (i.e., multivariate normality and homogeneity of covariance matrixes), logistic regression is a preferred procedure (Press & Wilson, 1978). Because predictor variables in this study were nonnormal, the logistic regression model was preferred.

Separate logistic regressions were developed for the environmental variables and for each time period (infancy, 4, 8, 12, 36, 48, and 84 months). In each stepwise analysis, a variable was included if it contributed significantly to the discrimination of the handicapped and nonhandicapped groups, at the .05 level, beyond the discrimination yielded by other variables present in the equation.

Finally, a stepwise logistic regression was performed with all variables retained in the previous stage. Environmental variables assessed at birth were entered first, followed by variables for 4, 8, 12, 36, 48, and 84 months. Again, group membership was the dependent variable, and a .05 significance level was employed for variable entry. It is important to again note that all data analyses were completed for handicapped versus nonhandicapped group membership. Analyses by diagnostic categorization are not reported because of inadequate sample size for the multivariate procedures used; in addition, verification of valid assignment to diagnostic categories by school districts could not be determined by the investigators.


In the first stage, potential predictor variables were screened for a univariate relationship with a handicapped versus nonhandicapped outcome. This reduced the potential number of factors to 7 environmental variables, 8 variables at age 4 months, 11 variables at age 8 months, 6 variables at age 12 months, 10 variables at 3 years, 11 variables at 4 years, and 13 variables at 7 years. A listing of these variables appears in Figure 1. Though paternal occupation demonstrated a significant, univariate relationship with the handicap variable (p [is less than] .01), a large incidence of missing data (i.e., 47%) for this factor required that it be deleted from further analyses.

No obstetrical or neonatal variables were entered into the regression analyses. That is, no prenatal or perinatal factors yielded statistically significant findings between handicapped versus nonhandicapped groups when examined with univariate analyses. These specific factors are presented in Figure 2.

In the second stage, separate stepwise logistic regressions were performed for each period. Among the environmental variables, maternal education was retained. No other environmental variable contributed significantly at the .05 level. At 4 months, response to mirror and placing were identified. At 8 months, the Bayley Motor score, physical development, and mental development were retained. For data collected at 12 months, neurological abnormalities, nonneurological abnormalities, and head circumference were identified as significant predictors. Finally, speech production and verbal expression emerged as significant, non-redundant factors at age 3 years, while IQ was retained at both 4 and 7 years.

In the third stage, all significant variables obtained from the second stage were entered into a stepwise logistic regression with order of variable entry determined by time of measurement as noted above. The analysis identified maternal education, placing at 4 months, physical development at 8 months, and neurological and nonneurological abnormalities at 12 months as nonredundant predictors of an adolescent handicapped diagnosis. At 3 years of age, speech intelligibility proved to be a nonredundant predictor, whereas a child's IQ score was significant both at 4 and 7 years of age.

Correct classification rates, sensitivity, and specificity for each function are described in Table 1. Sensitivity is defined as the percentage of handicapped children correctly identified as handicapped. Specificity is defined as the percentage of nonhandicapped children correctly identified as nonhandicapped. The results of logistic regressions include the predicted probability of group membership (handicapped vs. nonhandicapped) for each subject. By default, if a subject's predicted probability for group membership was greater than .50, the subject was classified into that group. The sensitivity and specificity values reported were based on this cutoff procedure. The setting of the cutpoint will affect sensitivity and specificity rates for a particular logistic model. The cutpoint procedure used was chosen to maximize the overall percent correctly classified for all subjects combined (i.e., the overall hit rate). Sample sizes did not permit the development of separate classification models for each subtype of the handicapped group. Sample size varies across age in Table 1 because all data were not available for all subjects over the 7-year time span.


All prenatal and perinatal data (i.e., pregnancy history, maternal medical and reproductive history, labor and delivery information, birth presentation and trauma) failed to yield significant results with univariate analyses. Therefore, despite the substantial differences observed between handicapped and nonhandicapped groups in adolescence, these groups, as a whole, were indistinguishable from each other during infancy.

When each variable cluster is viewed separately, the classification rates indeed reveal important differences. More specifically, whereas parental traits (i.e., maternal education) are more accurate predictors of handicapped adolescent status than the child's own behavior from birth to 3 years of age, child-centered skills assessed at 4 and 7 years of age prove to be better indicators of handicapping conditions than maternal level of educational attainment. Moreover, even when environmental traits are combined with child performance data gathered at 4 and 7 years of age, the accuracy of group classification is not appreciably influenced. Overall, therefore, while results underscore the initial role of ecological determinants of adolescent status, findings also suggest that the effects of these determinants are not constant over time; in fact, individual factors (e.g., maternal education) contribute differently to group classification depending on the age at which child level of functioning is examined.

In addition, specificity (i.e, accurately predicting the nonhandicapped) is consistently higher than sensitivity across time periods. While this is attributable to the algorithm which was used to assign students to either outcome group, these criteria, or "break points," can be adjusted and would therefore increase sensitivity. Accurately identifying a greater number of handicapped infants and toddlers can be improved, but an associated consequence would be decreased specificity, and thus, a higher number of false positives (children inaccurately identified as handicapped). Of significance here is that as screening models are conceptualized, it is critical to examine sensitivity/specificity relationships, for important cost and mislabeling issues are embedded in such decisions.

With respect to stepwise logistic regression findings, data appear to suggest that from birth to 12 months of age, the ability to accurately identify handicapped children can be facilitated by analyzing and adjusting factors within the environmental cluster. Conversely, at 4 and 7 years of age, measures relating to the child's developmental competence are of greatest predictive value, and thus warrant the closest attention in crafting screening initiatives. Overall, findings in this study indicate that early identification models that focus solely on developmental delay or adverse medical events from birth to 3 years of age are inadequate in fully identifying children eventually judged to be handicapped. While such models will identify youngsters with established conditions, they will ultimately identify only a small segment of the total handicapped population. If states elect to serve at-risk children, screening models must be multivariate (i.e., child and family focused) and account for differential weights of these factors over time.

Several important policy implications are inherent in these findings. First, all 50 states and U.S. territories have elected to pursue funds (P.L. 99-457, Part H.) to develop comprehensive early identification and intervention systems for developmentally delayed and "at risk" children from birth to 3 and the families of these children. Furthermore, 38% (N = 19) of the "Lead Agency" designations reside in State Departments of Education. Whereas such agencies are certainly most accustomed to developing and overseeing services for handicapped children under P.L. 94-142, examination of their respective regulatory policies also indicates screening and program eligibility criteria that are dominated by child-centered characteristics (Smith & Schakel, 1986). Data reported here suggest that as these agencies develop policies for special education services for children from birth to 5 years of age, decision-making models must reflect the complex and dynamic relationship that exists among child and family-centered factors. Our findings indicate that variables that are salient in screening and assessment processes from birth to 12 months are different from those at 4 and 7 years of age. Moreover, not only are changes evident over time in the factors themselves, but the relative weight, or discriminative ability, of specific factors (e.g., maternal education) also changes.

As departments of education and other lead agencies attempt to conceptualize screening criteria for children from birth to age 5 and their families, traditional perspectives such as categorical versus noncategorical models (Smith & Schakel, 1986) and a dominant focus on child ability testing (Lambert, 1988) may not be adequate to develop reliable, valid, and cost-effective early detection systems. The merits of noncategorical policies are indisputable for infants and toddlers; but approaches that emphasize the "use of a well-defined and measurable level of developmental delay" (Smith & Schakel), particularly for children less than 1 year of age, are highly likely to underidentify those children subsequently judged as handicapped.

Fortunately, an ever-increasing body of literature offers guidance and direction to those formulating policy in this area (Meisels & Provence, 1989). Although the utility of several influential macroscopic factors (e.g., maternal age and education) continues to be reported (Larson, Collet, & Hanley, 1987), a range of microscopic variables, such as maternal/child interaction and family needs/resources/support systems (Dunst, McWilliam, Trivette, & Galant, 1987; Mitchell, Magyary, Barnard, Sumner, & Booth, 1985) have also been identified. These latter variables are potentially useful for decision-making models and, more important, in the preparation of meaningful individualized family service plans.

These findings are indeed very convincing regarding the ecological determinants of adverse outcomes in children. It is crucial to note, however, that while no one child factor (e.g., anoxia, prematurity, neurological status) can accurately predict outcome, it is also true that isolated environmental factors alone cannot account for significant variance and, accordingly, should not be used exclusively in the development of alternative screening and assessment models. Caution must be exercised in interpreting and applying results from this study in that Collaborative Project data were collected 25-30 years ago, and may not accurately reflect relationships between predictor and criterion variables at present. Significant advances in neonatology and pediatric medicine that have occurred in the last decade may alter relationships reported here. In addition, dramatic increases in the number of infants victimized by teratogenic exposure (Jones & Lopez, 1988; Zuckerman et al., 1989) are not reflected in this study. While the developmental jeopardy resulting from toxic exposure is not fully understood (Select Committee on Children, 1989), this hazard must be reflected in child identification models.

We must note, also, that an important consequence of a child- and family-focused screening system is that a population may be identified for whom exclusively child-directed services may not be fully responsive to identified needs. Consequently, states must carefully examine screening and service relationships to ensure that the array of available services is congruent with child and family needs. The Sameroff et al. (1987) study is illuminating in this regard. In examining significant factors (e.g., life stress, mental health, education, maternal anxiety, and maternal/child interaction) that are predictive of child competence, the need becomes apparent for innovative service configurations.

P.L. 99-457 creates the opportunity for states to identify and serve high-risk children and their families. Electing to serve such populations represents a major conceptual shift from rehabilitative treatment to prevention in the field of special education. Recent survey data (Gallagher, Harbin, Thomas, Clifford, & Wenger, 1988) reveal that 18 states have decided to do so. While there is considerable variability across these states regarding their vision of risk (Graham & Scott, 1988), the specific factors selected and the manner in which they are combined will markedly affect not only the number of children seved but also the services required. A high-risk condition exists when a child has a greater-than-average chance of developing a disability. Risk is not a condition but rather a circumstance to indicate an elevated probability that a disorder will occur. Identifying, defining, and assessing these risk factors over time is an enormous challenge which will require the collaborative efforts of specialists in education, medicine, psychology, and epidemiology will be necessary for the task.

It is apparent that the forthcoming developmental period of statewide early intervention planning is replete with considerable challenges and complexities. We must pay thoughtful attention to these recommendations if we are to begin to comply with the lofty expectations of this federal statute on behalf of infants and toddlers with special needs and their families. [Figures 1 to 2 Omitted] [Tabular Data 1 Omitted]

THOMAS T. KOCHANEK is Professor, Department of Special Education, Rhode Island College, Providence. ROBERT I. KABACOFF is Associate Professor, School of Psychology, Nova University, Fort Lauderdale, Florida. LEWIS P. LIPSITT is Director, Child Study Center, Brown University, Providence, Rhode Island.
COPYRIGHT 1990 Council for Exceptional Children
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1990 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Kochanek, Thomas T.; Kabacoff, Robert I.; Lipsitt, Lewis P.
Publication:Exceptional Children
Date:Apr 1, 1990
Previous Article:Are regular education classes equipped to accomodate students with learning disabilities?
Next Article:Effects of self-evaluation on the independent work skills of preschool children with disabilities.

Related Articles
Groundbreaking Celebration for Bais Sonya Gutte Campus: A Gift of the Winnick Family Foundation; Collaboration Between Chabad of California and...
Research into practice.
New IDEA heralds some changes for instruction of students with disabilities.
Learning disabilities and young children: identification and intervention: a report from the National Joint Committee on Learning Disabilities *...

Terms of use | Copyright © 2016 Farlex, Inc. | Feedback | For webmasters