Predicting children with pervasive developmental disorders using the Wechsler Intelligence Scale for Children–Third Edition
*Tomonori Koyama, PhD, Department of Child and Adolescent Mental Health, National Institute of Mental Health, National Center of Neurology and Psychiatry, 4-1-1 Ogawa-Higashi, Kodaira, Tokyo 187-8553, Japan. Email: firstname.lastname@example.org
An original combination score (i.e. the sum of Vocabulary and Comprehension subtracted from the sum of Block Design and Digit Span) was created from the four Wechsler Intelligence Scale for Children–Third Edition (WISC-III) subtests identified by discriminant analysis on WISC-III data from 139/129 children with/without pervasive developmental disorders (PDD; mean, 8.3/8.1 years) and its utility examined for predicting PDD. Its best cut-off was 2/3, with sensitivity, specificity, positive and negative predictive values of 0.68, 0.61, 0.65 and 0.64, respectively. The score seems useful, so long as clinicians are aware of its limitations and use it only as a supplemental measure in PDD diagnosis.
RECENTLY, BAIRD ET AL. reported a prevalence of pervasive developmental disorders (PDD) of 1.16%,1 which is much higher than previously thought. Such a significant rise is primarily the result of broadened recognition and early detection of autistic conditions. But some PDD children are often misdiagnosed with non-PDD conditions, such as attention-deficit–hyperactivity disorder (ADHD).2
The Wechsler intelligence scales (Wechsler Intelligence Scale for Children [WISC], Wechsler Adult Intelligence Scale [WAIS]) are familiar cognitive tests, and because some of the characteristics have been identified in PDD children, such as a high score on Block Design or a low score on Comprehension,3–7 information from those subtests may be useful to differentiate PDD children from non-PDD ones. Mayes and Calhoun classified each child's WISC-III profile type by ranking the index and the subtest scores from lowest to highest, and found that low Coding with low Comprehension or low Freedom from Distractibility index with low Comprehension profile correctly identified 56% of children with high-functioning (IQ ≥ 80) autism and 76% without autism.8
Compared to reviewing all of the subtests and analyzing them as in the aforementioned study, it may be more useful in clinical settings to use a simple parameter for the selected subtests. In the present study we propose an original combination score from the Japanese version of the WISC-III9 and examine its utility for predicting PDD in children.
The present study was conducted as part of a comprehensive study approved by the ethics committee of the Tokyo University Graduate School of Medicine. Through consecutive referrals to three clinics specializing in developmental disorders in and near Tokyo, we selected all children who had completed the WISC-III. In each clinic, a clinical team consisting of experienced professionals led by the same child psychiatrist (H. K.) made DSM-IV10 diagnoses by consensus based on detailed clinical assessments of children and semi-structured comprehensive interview to parents for approximately 1.5 h.
The participants were 139 PDD children (autistic disorder, n = 26; Asperger's disorder, n = 24; PDD not otherwise specified, n = 89; mean age, 8.3 ± 2.1 years, range, 5–12 years; 111 boys; mean Full Scale IQ [FIQ], 93.2 ± 17.3, range, 49–134) and 129 non-PDD children (ADHD, n = 86; communication disorders, n = 15; borderline intellectual functioning/mental retardation, n = 13; adjustment disorders, n = 5; learning disorders, n = 2; selective mutism, n = 1; obsessive–compulsive disorder, n = 1; and without any DSM-IV axis I or II disorder, n = 6; mean age, 8.1 ± 1.9 years, range, 5–12 years; 105 boys; mean FIQ, 92.8 ± 19.0, range, 43–129). No significant differences in the mean age, sex ratio, or mean FIQ were observed between the two groups.
We identified WISC-III subtests useful to predict PDD using discriminant analysis and created a simple combination score from identified subtests, because such a score is clinically more useful than the discriminant function, which is not applicable to other databases. We compared the means between the two groups on t-test. The correlations between the combination score and children's age or FIQ were examined in each group.
We determined an appropriate score cut-off in order to differentiate PDD from non-PDD by examining sensitivity, specificity, positive and negative predictive values for each cut-off.
All tests were two-tailed and statistical significance was set at P < 0.05. All statistical analyses were performed using SPSS 15.0J for Windows (SPSS Japan, Tokyo, Japan).
Discriminant analysis correctly classified 95 of the 139 PDD children (sensitivity, 0.68) and 91 of the 129 non-PDD children (specificity, 0.71). The standardized canonical discriminant function coefficients were highest on Block Design (0.60) and Digit Span (0.46) and lowest on Vocabulary (−0.81) and Comprehension (−0.40). Based on this result, we simply created a combination score by summing Block Design and Digit Span and subtracting Vocabulary and Comprehension.
As expected, the mean combination score was significantly higher in PDD (mean, 5.5 ± 6.7) than non-PDD (mean, 0.6 ± 5.9), t = 6.39, d.f. = 266, P < 0.001. There was no significant correlation between the combination score and children's age or FIQ in each group.
Table 1 shows the cut-off and its related measures to differentiate PDD from non-PDD. We considered a combination score of ≤2/≥3 to be the best to predict non-PDD/PDD.
Table 1. Cut-off and related measures of combination score for predicting PDD
Our original combination score derived from Vocabulary, Comprehension, Digit Span and Block Design could differentiate PDD and non-PDD conditions moderately. These subtests have been known to show which autistic children performed well or badly,3–7 reflecting their unique cognitive characteristics, such as detail-focused visual processing or weakness to comprehend social contexts.
The sensitivity and specificity of approximately 0.65 are not satisfactory, possibly because the score is derived only from cognitive tests and does not consist of any items that assess autistic symptoms. Even discriminant analysis did not yield a satisfactory result. The present findings and those of Mayes and Calhoun8 indicate that clinicians should not rely heavily on the cognitive profile when diagnosing PDD. Nevertheless, because the four subtests are easy to administer and the calculating formula of the combination score is very simple, careful interpretation of the score could provide additional information on PDD to clinicians.
The utility of the present combination score may not be able to be generalized, although the sample size was very large, due to the fact that it was a specialized clinic-based sample and hence may contain a sampling bias. In addition, its utility in adolescents or adults who have a history of developmental disorders remains to be clarified.
In conclusion, the combination score calculated from the four WISC-III subtests seems useful, so long as clinicians are aware of its limitations and use it only as a supplemental measure in PDD diagnosis.
We thank Ms Yoko Hayashi, Ms Keiko Shimoyamada, Mr Hiromi Ishida, Mr Junichi Yukimoto, Ms Tomoko Nakano, and Ms Mika Tobari for their help with data collection.