We identified all infants born at 20 through 25 weeks of gestation (as determined by the date of the mother's last menstrual period and by early ultrasonography) in all 276 maternity units in the United Kingdom and Ireland from March through December 1995. Of the 4004 infants identified, 1185 had signs of life at birth, and 843 of these were admitted to neonatal intensive care units (the remainder died in the delivery room) (Table 1). For these 843 infants, gestation was recalculated by using the date of the last menstrual period, by review of ultrasound studies done before 20 weeks, or (in the case of 29 infants) on the basis of clinical examination by a pediatrician. Three of these 29 infants were considered by the pediatrician to have been born at 26 weeks or more of gestation and were therefore excluded. If the estimates based on the last menstrual period and on ultrasonography were both available and were discrepant by at least 14 days, the ultrasound estimate was used; this resulted in the exclusion of an additional 29 infants. Thus, there remained 811 infants admitted to the neonatal intensive care unit whose gestational age was not more than 25 weeks and 6 days. Of these, 497 died in the hospital and 314 were discharged home. Six infants died after discharge, and six left the United Kingdom or Ireland.
Between January 1998 and January 1999, the parents of the 302 survivors living in the United Kingdom or Ireland were offered the opportunity to have their children undergo a standardized follow-up assessment. The parents of 16 children declined the invitation, and the parents of 3 others failed to bring their children for their prearranged evaluation; thus 283 of the survivors (92 percent) participated in this study. Of the 25 surviving children who did not take part, information was available on 17 from assessments performed by their local pediatricians at about 30 months of age corrected for gestation, and on a further 6 from assessments performed locally at about 12 months of corrected age. The data obtained were sufficient to allow us to determine whether the child had a disability. Of these 23 children, 9 were classified as severely disabled, 4 as having other disabilities, and 10 as having no disability. Information on outcome was therefore unavailable for only 2 of 308 surviving children.
The study was approved by each local research ethics committee before collection of the original data. All families for whom outcome is reported gave written informed consent for the examination.
Each child was examined in a hospital outpatient clinic or health center near the family home (235 children) or at home (48 children). The assessment consisted of a detailed medical history obtained from a semistructured interview with the family; a clinical examination, including neurologic assessment10; a classification of the degree and type of disability11,12; a functional classification of hearing and visual ability12; and a growth assessment, including measurement of maximal occipitofrontal head circumference with a LASSO-O tape (Child Growth Foundation, London). Development was assessed with use of the Bayley Scales of Infant Development (second edition),13 which yielded scores for mental and psychomotor development with a standardization mean of 100 and standard deviation of 15 points. Scores over 100 indicated advanced development for age. Development was considered severely impaired if the scores were less than 55, moderately impaired if they were 55 to 69, and mildly impaired if they were 70 to 84. There are no standardization data for children in the United Kingdom, but in a study of 850 three-year-olds in the Avon region of the United Kingdom, most of whom were born at term, the mean mental development score was 99.3±11.3 (SD) and the mean psychomotor development score was 96.0±14.7 (unpublished data). We used the U.S. standardization norms,13 with a mean of 100 and a standard deviation of 15. Scores were corrected for gestational age because, even at this age, uncorrected results would alter performance by 12.5 to 15.0 percent. If the child was unable to complete the Bayley assessment (because of poor vision, for example), the pediatrician estimated the child's developmental level as severely or moderately impaired (equivalent to Bayley scores under 55 or 55 to 69, respectively) or as not impaired.
Ten experienced developmental pediatricians were recruited to perform the assessments. All attended a course in which they were trained in all aspects of the assessments. A psychologist provided instruction in the use of the Bayley scales. After the training course, all the pediatricians submitted video recordings of pilot Bayley assessments, which were scored independently. All the pediatricians reached a prespecified level of agreement of more than 80 percent with the psychologist for individual items on the Bayley scales before study testing commenced. Further tapes were submitted at random during the study, and scoring was rechecked. Overall agreement was reached on 91 percent of the items making up the Mental Development Index and 88 percent of the items of the Psychomotor Development Index. One pediatrician left during the study and was replaced by another, who achieved similar levels of agreement for the Bayley assessment. The developmental pediatricians were unaware of the neonatal courses of the children they evaluated.
Classification of Disability
Disability was classified according to a previously published scheme.11 A "severe disability" was defined as one that was likely to put the child in need of physical assistance to perform daily activities. Disabilities that did not fall into this category were referred to simply as "other disabilities." The criteria for severe and other disabilities in each neurologic domain are shown in Table 2. Cerebral palsy was classified retrospectively according to the descriptions of function for each limb12 in children with abnormal results on neurologic examination.10 The categories were diplegia, in which upper-limb function was better than lower-limb function, despite asymmetries; hemiplegia, in which lower-limb function was better than upper-limb function, usually with asymmetry; quadriplegia, in which all four limbs appeared equally affected; and other, nonspastic types of cerebral palsy, such as hypotonia and dyskinesia. This classification was independent of the degree of disability.
Data collected on standardized forms and Bayley records were sent by mail to the study center, where they were encoded for computer analysis with SPSS for Windows software (release 8.0.0, SPSS, Chicago). To check for accuracy, the data were double-entered and the files were compared. The data for each infant were examined and outliers were verified before the data were combined with the main study data set for analysis. Categorical outcomes were compared with use of chi-square tests for trends, as appropriate, or Fisher's exact test. Continuous outcomes were compared with use of independent Student's t-tests. All statistical tests were two-sided.