This study investigated the reliability and stability of an autism diagnosis in children under 3 years of age who received independent diagnostic evaluations from two clinicians during two consecutive yearly evaluations. Strong evidence for the reliability and stability of the diagnosis was obtained. Diagnostic agreement between clinicians was higher for the broader discrimination of autism spectrum vs. no autism spectrum than for the more specific discrimination of autism vs. PDD-NOS. The diagnosis of autism at age 2 was more stable than the diagnosis of PDD-NOS at the same age. Social deficits and delays in spoken language were the most prominent DSM-IV characteristics evidenced by very young children with autism.

One of the oldest and best known developmental screening tests was recently restandardized and revised as the Denver-II. Because it was published without evidence of its accuracy, the present study was undertaken with 104 children between 3 and 72 months of age attending one of five day-care centers. To determine the presence of developmental problems, children were administered individual measures of intelligence, speech-language, achievement, and adaptive behavior. A second psychological examiner, blind to the outcome of the diagnostic battery, administered the Denver-II. Developmental problems including language impairments, learning disabilities, mild mental retardation, and/or functional developmental delay were found in 17% of the children. The Denver-II identified correctly 83% and thus had high rates of sensitivity. However, more than half the children with normal development also received abnormal, questionable, or untestable Denver-II scores. Thus the test had limited specificity (43%) and a high overreferral rate. The alternative scoring method, categorizing questionable/untestable scores as normal, caused sensitivity to drop to 56% although specificity rose to 80%. Since neither scoring method produced acceptable levels of accuracy, an effort was made to locate the sources of accuracy and inaccuracy within the test. Only items in the language domain were modestly helpful in discriminating children with and without difficulties. The findings suggest that the authors of the Denver-II need to engage in further development of the instrument including revising scoring criteria and item placement in relation to children's ages. In the interim, test users should employ screening tests which are more accurate such as the Minnesota Inventories or the Battelle Developmental Inventory Screening Test.

To the Editor.— We commend Frankenburg et al for their efforts to revise the Denver Developmental Screening Test.1 The authors have wisely incorporated suggestions from many researchers who conducted validity studies on older versions of the measure.2-5 As a consequence, the Denver II appears to be substantially improved, particularly in its measurement of language. This is quite important because the more prevalent developmental disabilities share language deficits as a common characteristic. Yet despite its manny improvements, the Dennver II is clearly unnfinished adn unready for marketing.